Merge pull request 'fix(ai-avatar): 修复三个bug——封面重影/标题字号缩放/对口型音频截断' (#1876) from fix/ai-avatar-three-bugs-0913 into develop

1. 删除封面展示区重复叠加的标题div（成片帧已含标题，不再CSS叠字）
2. 标题字号按视频分辨率等比缩放，预览容器动态测量宽度计算scale
3. enable_video_loop默认改为true，防止TTS音频长于出镜视频时被截断

.gitea/workflows/ci-pipeline.yml

@@ -20,6 +20,7 @@ on:
         default: "手动触发 - CI漏触发补跑"
 permissions:
   contents: read
+  pull-requests: read
 concurrency:
   group: ci-pipeline-${{ gitea.ref }}
   cancel-in-progress: true
@@ -88,9 +89,22 @@ jobs:
         GITHUB_TOKEN: ${{ github.token }}
       run: |
         set -eu
+        # 优先用 git diff 判断 PR 改动范围（比 API 稳定）
         PR_NUMBER=$(echo "$GITHUB_REF" | sed 's|refs/pull/||; s|/.*||')
-        API_URL="${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/pulls/${PR_NUMBER}/files?limit=300"
-        FILES=$(curl -s -H "Authorization: token ${GITHUB_TOKEN}" "$API_URL" | python3 -c "import sys,json; [print(f['filename']) for f in json.load(sys.stdin)]")
+        if command -v git >/dev/null 2>&1 && [ -d .git ]; then
+          FILES=$(git diff --name-only origin/develop...HEAD 2>/dev/null || true)
+        fi
+        if [ -z "${FILES:-}" ]; then
+          # fallback 到 API
+          API_URL="${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/pulls/${PR_NUMBER}/files?limit=300"
+          FILES=$(curl -sf -H "Authorization: token ${GITHUB_TOKEN}" "$API_URL" | python3 -c "import sys,json; [print(f['filename']) for f in json.load(sys.stdin)]" 2>/dev/null || true)
+        fi
+        if [ -z "${FILES:-}" ]; then
+          echo "⚠️ 无法获取变更文件列表，保守运行完整 CI"
+          echo "skip_backend=false" >> $GITHUB_OUTPUT
+          echo "skip_frontend=false" >> $GITHUB_OUTPUT
+          exit 0
+        fi
         FRONTEND_COUNT=$(echo "$FILES" | grep -c '^apps/web/' || true)
         BACKEND_COUNT=$(echo "$FILES" | grep -cv '^apps/web/' || true)
         TOTAL=$(echo "$FILES" | grep -cv '^$' || true)
@@ -182,7 +196,7 @@ jobs:
     - name: Run style checks
       shell: bash
       run: bash scripts/ci/validate_style.sh
-    - name: Auto-fix formatting (black + isort)
+    - name: Auto-fix formatting (black + isort + ruff)
       if: failure()
       shell: sh
       env:
@@ -813,9 +827,6 @@ jobs:
         CACHE_REF="${REGISTRY}/${{ matrix.cache_name }}:develop"
 
         EXTRA_BUILD_ARGS="APP_VERSION=\"${GITHUB_SHA}\""
-        if [ "${{ matrix.service }}" = "web" ]; then
-          EXTRA_BUILD_ARGS="$EXTRA_BUILD_ARGS NGINX_CONF=infra/docker/nginx-staging.conf"
-        fi
 
         # Worker 与 API/Web 统一走持久 builder（ci-builder-persist），共享宿主机层缓存
         NO_CACHE_FLAG=""
@@ -1012,9 +1023,6 @@ jobs:
         CACHE_REF="${REGISTRY}/${{ matrix.cache_name }}:${GITHUB_REF_NAME}"
 
         EXTRA_BUILD_ARGS="APP_VERSION=\"${GITHUB_SHA}\""
-        if [ "${{ matrix.service }}" = "web" ]; then
-          EXTRA_BUILD_ARGS="$EXTRA_BUILD_ARGS NGINX_CONF=infra/docker/nginx-staging.conf"
-        fi
 
         NO_CACHE_FLAG=""
         for i in 1 2 3; do
@@ -1553,9 +1561,6 @@ jobs:
         CACHE_REF="${REGISTRY}/${{ matrix.cache_name }}:main"
 
         EXTRA_BUILD_ARGS="APP_VERSION=\"${TAG_NAME}\""
-        if [ "${{ matrix.service }}" = "web" ]; then
-          EXTRA_BUILD_ARGS="$EXTRA_BUILD_ARGS NGINX_CONF=infra/docker/nginx-production.conf"
-        fi
 
         # Docker build 带重试：失败自动重试2次，第2次重试加--no-cache
         NO_CACHE_FLAG=""

.gitea/workflows/debug-web-v2.yml

@@ -0,0 +1,60 @@
+name: "Debug: Web container v2 (mount conflict)"
+on:
+  push:
+    branches: [debug/web-crash-v2]
+  workflow_dispatch:
+jobs:
+  web-diag:
+    runs-on: runtime-builder
+    timeout-minutes: 10
+    steps:
+      - name: Setup SSH and diagnose
+        shell: bash
+        env:
+          STAGING_SSH_KEY: ${{ secrets.PREVIEW_SSH_KEY }}
+        run: |
+          set -x
+          which ssh || (apt-get update -qq && apt-get install -y -qq openssh-client)
+          mkdir -p ~/.ssh && chmod 700 ~/.ssh
+          printf "%s" "$STAGING_SSH_KEY" > ~/.ssh/id_rsa
+          chmod 600 ~/.ssh/id_rsa
+          H=47.98.113.167; P=22222
+          ssh-keyscan -p $P -H $H >> ~/.ssh/known_hosts 2>/dev/null
+          ssh -p $P -i ~/.ssh/id_rsa -o StrictHostKeyChecking=no root@$H 'bash -s' <<'REMOTE'
+          set -x
+          echo "=== Current staging containers ==="
+          docker ps -a --filter name=xiaoxia-*-staging --format "table {{.Names}}\t{{.Status}}\t{{.Image}}"
+          echo ""
+          echo "=== Web container logs (current/current-rolledback) ==="
+          docker logs xiaoxia-web-staging 2>&1 | tail -40
+          echo ""
+          echo "=== Web inspect: env & mounts ==="
+          docker inspect xiaoxia-web-staging --format 'Entrypoint: {{.Config.Entrypoint}} Cmd: {{.Config.Cmd}}'
+          docker inspect xiaoxia-web-staging --format '{{range .Config.Env}}{{.}}{{"\n"}}{{end}}' | grep -E "APP_ENV|VERSION"
+          echo "Mounts:"
+          docker inspect xiaoxia-web-staging --format '{{range .Mounts}}{{.Type}} {{.Source}} -> {{.Destination}} (rw={{.RW}}){{"\n"}}{{end}}'
+          echo ""
+          echo "=== Reproduce: rm on read-only bind mount ==="
+          docker run --rm --name nginx-ro-test \
+            -v /var/lib/xiaoxia-saas-staging/nginx-staging.conf:/etc/nginx/conf.d/default.conf:ro \
+            git.xiaoxiajianji.com/xiaoxia/xiaoxia-saas/xiaoxia-saas-web:387514c \
+            sh -c '
+              set -x
+              echo "Before:"
+              ls -la /etc/nginx/conf.d/
+              echo "Try rm (as entrypoint does):"
+              rm -f /etc/nginx/conf.d/default.conf
+              echo "rm exitcode=$?"
+              echo "After rm:"
+              ls -la /etc/nginx/conf.d/
+              echo "Test ln:"
+              ln -s /etc/nginx/nginx-staging.conf /etc/nginx/conf.d/default.conf
+              echo "ln exitcode=$?"
+              ls -la /etc/nginx/conf.d/
+              echo "nginx -t:"
+              nginx -t 2>&1
+            ' 2>&1
+          echo ""
+          echo "=== Also test with NEW fixed image (9c0d4b1 if present) ==="
+          docker images | grep xiaoxia-saas-web | head -5
+REMOTE

alembic/versions/073_add_lipsync_tts_fields.py

@@ -0,0 +1,45 @@
+"""lipsync_jobs 增加 TTS 直生字段（voice_id/script_text/speed/emotion）
+
+Revision ID: 073_add_lipsync_tts_fields
+Revises: 072_add_ai_avatar_render
+Create Date: 2026-09-09
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "073_add_lipsync_tts_fields"
+down_revision = "072_add_ai_avatar_render"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # 对口型支持「传音色 + 文案直接生成」：后端内部先 TTS 合成音频再提交对口型
+    op.add_column(
+        "lipsync_jobs",
+        sa.Column("voice_id", sa.String(200), nullable=False, server_default=""),
+    )
+    op.add_column(
+        "lipsync_jobs",
+        sa.Column("script_text", sa.Text(), nullable=False, server_default=""),
+    )
+    op.add_column(
+        "lipsync_jobs",
+        sa.Column("speed", sa.Float(), nullable=False, server_default=sa.text("1.0")),
+    )
+    op.add_column(
+        "lipsync_jobs",
+        sa.Column("emotion", sa.String(20), nullable=False, server_default=""),
+    )
+    # audio_url 改为可空：直生模式下音频由后端 TTS 合成后回填
+    op.alter_column("lipsync_jobs", "audio_url", existing_type=sa.Text(), nullable=True)
+
+
+def downgrade() -> None:
+    op.alter_column("lipsync_jobs", "audio_url", existing_type=sa.Text(), nullable=False)
+    op.drop_column("lipsync_jobs", "emotion")
+    op.drop_column("lipsync_jobs", "speed")
+    op.drop_column("lipsync_jobs", "script_text")
+    op.drop_column("lipsync_jobs", "voice_id")

alembic/versions/074_ai_avatar_render_script_id_optional.py

@@ -0,0 +1,36 @@
+"""ai_avatar_render_jobs.script_id 放宽为可空串（手动文案直生场景不关联文案库）
+
+Revision ID: 074_render_script_id_optional
+Revises: 073_add_lipsync_tts_fields
+Create Date: 2026-09-09
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "074_render_script_id_optional"
+down_revision = "073_add_lipsync_tts_fields"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # 列保持 NOT NULL（空串占位），仅应用层允许不传；这里显式补 server_default 防止历史约束歧义
+    with op.batch_alter_table("ai_avatar_render_jobs") as batch:
+        batch.alter_column(
+            "script_id",
+            existing_type=sa.String(length=36),
+            nullable=False,
+            server_default="",
+        )
+
+
+def downgrade() -> None:
+    with op.batch_alter_table("ai_avatar_render_jobs") as batch:
+        batch.alter_column(
+            "script_id",
+            existing_type=sa.String(length=36),
+            nullable=False,
+            server_default=None,
+        )

alembic/versions/075_add_sentence_timings.py

@@ -0,0 +1,27 @@
+"""add sentence_timings to lipsync_jobs
+
+Revision ID: 075_add_sentence_timings
+Revises: 074_ai_avatar_render_script_id_optional
+Create Date: 2026-09-12
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "075_add_sentence_timings"
+down_revision = "074_render_script_id_optional"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    with op.batch_alter_table("lipsync_jobs") as batch:
+        batch.add_column(
+            sa.Column("sentence_timings", sa.JSON(), nullable=True),
+        )
+
+
+def downgrade() -> None:
+    with op.batch_alter_table("lipsync_jobs") as batch:
+        batch.drop_column("sentence_timings")

apps/api/app/api/routes/ai_avatar_render.py

@@ -11,13 +11,16 @@
 from __future__ import annotations
 
 import logging
+from datetime import datetime, timezone
 
 from app.auth import AuthenticatedUser, get_current_user
 from app.dependencies import get_db_session
 from app.schemas.ai_avatar_render import (
     AiAvatarRenderJobResponse,
     CreateAiAvatarRenderRequest,
+    SmartCoverResponse,
 )
+from app.services.ai_avatar_cover_service import generate_smart_cover
 from app.services.ai_avatar_render_service import (
     AiAvatarRenderError,
     AiAvatarRenderService,
@@ -49,7 +52,7 @@ def create_render_job(
     """
     try:
         job = svc.create_render_job(
-            user_id=current_user.id,
+            user_id=current_user.user.id,
             lipsync_job_id=body.lipsync_job_id,
             script_id=body.script_id,
             b_roll_segments=[s.model_dump() for s in body.b_roll_segments],
@@ -74,10 +77,16 @@ def create_render_job(
         from app.tasks.ai_avatar_render import execute_ai_avatar_render
 
         execute_ai_avatar_render.delay(job.id)
-    except Exception:
-        logger.warning("Celery 任务提交失败，渲染任务已创建但未触发执行: %s", job.id)
+    except Exception as exc:
+        logger.exception("Celery 任务投递失败（创建）: job_id=%s err=%s", job.id, exc)
+        job.status = "failed"
+        job.error_message = f"任务提交失败：{exc}"
+        job.updated_at = datetime.now(timezone.utc)
+        svc.db.commit()
+        svc.db.refresh(job)
+        return AiAvatarRenderJobResponse.model_validate(job)
 
-    return job
+    return AiAvatarRenderJobResponse.model_validate(job)
 
 
 # ── GET /jobs — 任务列表 ─────────────────────────────────────────────────
@@ -94,7 +103,7 @@ def list_render_jobs(
 ):
     """获取 AI 数字人渲染任务列表."""
     items, total = svc.list_render_jobs(
-        user_id=current_user.id,
+        user_id=current_user.user.id,
         project_id=project_id,
         status=status,
         offset=offset,
@@ -118,7 +127,7 @@ def get_render_job(
     svc: AiAvatarRenderService = Depends(_get_service),
 ):
     """获取渲染任务详情."""
-    job = svc.get_render_job(job_id, current_user.id)
+    job = svc.get_render_job(job_id, current_user.user.id)
     if job is None:
         raise HTTPException(status_code=404, detail="渲染任务不存在")
     return job
@@ -134,7 +143,7 @@ def cancel_render_job(
     svc: AiAvatarRenderService = Depends(_get_service),
 ):
     """取消渲染任务（仅 pending 状态可取消）."""
-    job = svc.cancel_render_job(job_id, current_user.id)
+    job = svc.cancel_render_job(job_id, current_user.user.id)
     if job is None:
         raise HTTPException(status_code=404, detail="渲染任务不存在")
     if job.status != "cancelled":
@@ -155,7 +164,7 @@ def retry_render_job(
     svc: AiAvatarRenderService = Depends(_get_service),
 ):
     """重试失败的渲染任务."""
-    job = svc.retry_render_job(job_id, current_user.id)
+    job = svc.retry_render_job(job_id, current_user.user.id)
     if job is None:
         raise HTTPException(status_code=404, detail="渲染任务不存在")
     if job.status != "pending":
@@ -169,7 +178,82 @@ def retry_render_job(
         from app.tasks.ai_avatar_render import execute_ai_avatar_render
 
         execute_ai_avatar_render.delay(job.id)
-    except Exception:
-        logger.warning("Celery 任务提交失败，重试任务已重置但未触发执行: %s", job.id)
+    except Exception as exc:
+        logger.exception("Celery 任务投递失败（重试）: job_id=%s err=%s", job.id, exc)
+        job.status = "failed"
+        job.error_message = f"任务提交失败：{exc}"
+        job.updated_at = datetime.now(timezone.utc)
+        svc.db.commit()
+        svc.db.refresh(job)
+        return AiAvatarRenderJobResponse.model_validate(job)
 
-    return job
+    return AiAvatarRenderJobResponse.model_validate(job)
+
+
+# ── POST /{job_id}/smart-cover — 从最终成片智能抽封面（步骤②）────────
+
+
+@router.post("/{job_id}/smart-cover", response_model=SmartCoverResponse)
+def generate_render_smart_cover(
+    job_id: str,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    db: Session = Depends(get_db_session),
+):
+    """从最终渲染成片智能抽帧生成封面（MediaKit 抽帧 + 评分选最佳帧 + 转存 OSS）.
+
+    - 必须等渲染任务 completed 后才可调用（否则返回 400）
+    - 生成成功后自动更新 render_job 的 cover_config 与 output_cover_url
+    """
+    from app.services.ai_avatar_render_service import AiAvatarRenderService
+
+    svc = AiAvatarRenderService(db)
+    job = svc.get_render_job(job_id, current_user.user.id)
+    if job is None:
+        raise HTTPException(status_code=404, detail="渲染任务不存在")
+    if job.status != "completed":
+        raise HTTPException(status_code=400, detail="请先完成视频生成")
+    video_url = (job.output_video_url or "").strip()
+    if not video_url:
+        raise HTTPException(status_code=400, detail="渲染成片视频 URL 为空")
+
+    try:
+        # 从最终成片抽帧，帧本身已含标题/B-roll，直接转存 OSS
+        cover_url = generate_smart_cover(video_url, job_id=job_id, max_frames=5)
+    except Exception as exc:
+        logger.error(
+            "渲染成片智能封面生成异常: user=%s render_id=%s video_url=%s err=%s",
+            current_user.user.id,
+            job_id,
+            video_url[:80],
+            exc,
+            exc_info=True,
+        )
+        cover_url = ""
+
+    if not cover_url:
+        return SmartCoverResponse(
+            cover_url="",
+            status="fallback_failed",
+            message="智能抽帧失败（MediaKit 不可用或抽帧异常），请稍后重试",
+        )
+
+    # 更新 render_job 的封面字段（异步写入 DB；失败不影响返回）
+    try:
+        job.cover_config = {
+            **(job.cover_config if isinstance(job.cover_config, dict) else {}),
+            "mode": "auto_frame",
+            "url": cover_url,
+        }
+        job.output_cover_url = cover_url
+        job.updated_at = datetime.now(timezone.utc)
+        db.commit()
+    except Exception as exc:
+        logger.warning("更新 render_job 封面字段失败（不影响返回）: job_id=%s err=%s", job_id, exc)
+
+    logger.info(
+        "渲染成片智能封面生成成功: user=%s render_id=%s cover_url=%s",
+        current_user.user.id,
+        job_id,
+        cover_url[:120],
+    )
+    return SmartCoverResponse(cover_url=cover_url, status="completed")

apps/api/app/api/routes/chunked_upload.py

@@ -177,8 +177,8 @@ def _cleanup_expired_uploads() -> int:
                 meta_file.unlink()
                 cleaned += 1
                 logger.info(f"Cleaned up expired upload: {upload_id}")
-        except Exception as e:
-            logger.warning(f"Failed to cleanup upload metadata {meta_file}: {e}")
+        except Exception:
+            logger.exception("Failed to cleanup upload metadata: %s", meta_file)
 
     return cleaned
 

apps/api/app/api/routes/generation_variant_plans.py

@@ -108,7 +108,7 @@ def create_variant_plans(
             if _latest:
                 source_plan_id = _latest.id
         except Exception:
-            logger.warning("[variant-plans] 源 plan 解析失败", exc_info=True)
+            logger.exception("[variant-plans] 源 plan 解析失败")
 
     if not source_plan_id:
         raise HTTPException(
@@ -122,7 +122,7 @@ def create_variant_plans(
 
         voice_durations = _query_voice_durations(db, voices)
     except Exception:
-        logger.warning("[variant-plans] 配音时长查询失败（按占位段长选片）", exc_info=True)
+        logger.exception("[variant-plans] 配音时长查询失败（按占位段长选片）")
         voice_durations = [0.0] * request.count
 
     from app.services.edit_plan_service import EditPlanService
@@ -143,7 +143,7 @@ def create_variant_plans(
     except HTTPException:
         raise
     except Exception as e:
-        logger.error("[variant-plans] 选片异常: %s", e, exc_info=True)
+        logger.exception("[variant-plans] 选片异常")
         raise HTTPException(status_code=500, detail="选片失败，请稍后重试") from e
 
     # 组装 clips 响应

apps/api/app/api/routes/lipsync.py

@@ -1,11 +1,12 @@
-"""对口型 API 路由 — #1796 MediaKit 对口型, #1809 参数调整.
+"""对口型 API 路由 — #1796 MediaKit 对口型, #1809 参数调整, #1845 配音前置.
 
 接口：
-  POST   /api/v1/lipsync/jobs          提交对口型任务
+  POST   /api/v1/lipsync/jobs          提交对口型任务（支持 TTS/直传/预合成 三种模式）
   GET    /api/v1/lipsync/jobs          任务列表
   GET    /api/v1/lipsync/jobs/{id}     任务详情
   POST   /api/v1/lipsync/jobs/{id}/refresh  刷新任务状态
   POST   /api/v1/lipsync/jobs/{id}/cancel   取消任务
+  POST   /api/v1/lipsync/tts-preview   #1845 步骤1 TTS 预合成（同步 HTTP，~2-3s）
 """
 
 from __future__ import annotations
@@ -13,15 +14,21 @@ from __future__ import annotations
 import logging
 
 from app.auth import AuthenticatedUser, get_current_user
-from app.dependencies import get_cosyvoice_service, get_db_session, get_voice_clone_profile_repository
-from app.schemas.lipsync import CreateLipsyncJobRequest, LipsyncJobResponse
+from app.dependencies import (
+    get_db_session,
+    get_voice_clone_profile_repository,
+)
+from app.schemas.lipsync import (
+    AiAvatarTtsPreviewRequest,
+    AiAvatarTtsPreviewResponse,
+    CreateLipsyncJobRequest,
+    LipsyncJobResponse,
+)
 from app.services.lipsync_service import LipsyncService
 from app.services.mediakit_client import MediaKitError
-from fastapi import APIRouter, Depends, HTTPException, Query
+from fastapi import APIRouter, BackgroundTasks, Depends, HTTPException, Query
 from sqlalchemy.orm import Session
 
-from packages.application.cosyvoice_service import CosyVoiceError, CosyVoiceService
-
 logger = logging.getLogger(__name__)
 
 router = APIRouter()
@@ -29,35 +36,13 @@ router = APIRouter()
 
 def _get_service(
     db: Session = Depends(get_db_session),
-    cosyvoice_service: CosyVoiceService = Depends(get_cosyvoice_service),
+    voice_clone_repo=Depends(get_voice_clone_profile_repository),
 ) -> LipsyncService:
-    return LipsyncService(db, cosyvoice_service=cosyvoice_service)
-
-
-def _resolve_voice_id(
-    raw_voice_id: str,
-    user_id: str,
-    voice_clone_repo,
-) -> str:
-    """解析 voice_id：支持预设音色 ID 或克隆音色 profile UUID.
-
-    与 TTS 路由保持一致：命中 profile → 校验归属 → 取 CosyVoice voice_id。
-    """
-    try:
-        profile = voice_clone_repo.get(raw_voice_id)
-    except Exception as exc:
-        logger.error("查询克隆音色失败: voice_id=%s, error=%s", raw_voice_id, exc)
-        raise HTTPException(
-            status_code=400,
-            detail=f"voice_id 无效: {raw_voice_id}",
-        ) from exc
-    if profile is not None:
-        if profile.user_id != user_id:
-            raise HTTPException(status_code=403, detail="无权访问该音色")
-        if not profile.voice_id:
-            raise HTTPException(status_code=400, detail="音色克隆尚未完成，请稍后再试")
-        return profile.voice_id
-    return raw_voice_id
+    # voice_clone_repo 用于克隆音色 profile 解析
+    return LipsyncService(
+        db,
+        voice_clone_repo=voice_clone_repo,
+    )
 
 
 # ── POST /jobs — 提交对口型任务 ───────────────────────────────────────────
@@ -68,45 +53,47 @@ def create_lipsync_job(
     body: CreateLipsyncJobRequest,
     current_user: AuthenticatedUser = Depends(get_current_user),
     svc: LipsyncService = Depends(_get_service),
-    voice_clone_repo=Depends(get_voice_clone_profile_repository),
 ):
     """提交对口型任务.
 
-    #1809: 前端传 {voice_id, script_text, video_url}，
-    后端内部调 TTS 合成音频，再提交 MediaKit。
+    三种模式：
+    - TTS 直生（旧版/降级）：传 {video_url, voice_id, script_text, speed?, emotion?}，
+      后端 dispatch Celery 异步任务。
+    - 直接音频：传 {video_url, audio_url}，后端同步下载+算timings+提交MediaKit。
+    - 预合成音频（#1845 新主路径）：传 {video_url, audio_url, audio_duration, sentence_timings}，
+      后端同步ffprobe+写入timings+直接提交MediaKit（~2-3s）。
     """
-    # 解析 voice_id（支持克隆音色 profile UUID）
-    actual_voice_id = _resolve_voice_id(body.voice_id, current_user.id, voice_clone_repo)
-
     try:
         job = svc.create_job(
-            user_id=current_user.id,
+            user_id=current_user.user.id,
             video_url=body.video_url,
-            voice_id=actual_voice_id,
+            audio_url=body.audio_url,
+            audio_duration=body.audio_duration,
+            sentence_timings=body.sentence_timings,
+            voice_id=body.voice_id,
             script_text=body.script_text,
+            speed=body.speed,
+            emotion=body.emotion,
             enable_video_loop=body.enable_video_loop,
             project_id=body.project_id,
         )
     except ValueError as exc:
-        # 参数无效（如 voice_id 格式不对、文本过长等）
         raise HTTPException(status_code=400, detail=str(exc)) from exc
-    except CosyVoiceError as exc:
-        # TTS 合成基础设施失败（API/网络/认证）
-        raise HTTPException(
-            status_code=502,
-            detail={"code": "TTSSynthesisFailed", "message": str(exc)},
-        ) from exc
     except MediaKitError as exc:
+        status_code = 502
+        if exc.code in ("VoiceForbidden",):
+            status_code = 403
+        elif exc.code in ("InvalidInput", "TTSInvalidParam", "VoiceNotReady"):
+            status_code = 400
         raise HTTPException(
-            status_code=502,
+            status_code=status_code,
             detail={
                 "code": exc.code,
                 "message": str(exc),
-                "request_id": exc.request_id,
+                "request_id": getattr(exc, "request_id", ""),
             },
         ) from exc
     except Exception as exc:
-        # 兜底：任何未预期的错误返回 400 而非 500
         logger.error("创建对口型任务异常: %s", exc, exc_info=True)
         raise HTTPException(
             status_code=400,
@@ -116,6 +103,52 @@ def create_lipsync_job(
     return job
 
 
+# ── POST /tts-preview — #1845 步骤1 TTS 预合成 ──────────────────────────
+
+
+@router.post("/tts-preview", response_model=AiAvatarTtsPreviewResponse)
+def preview_tts(
+    body: AiAvatarTtsPreviewRequest,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    svc: LipsyncService = Depends(_get_service),
+):
+    """步骤1「生成配音」同步 TTS 预合成.
+
+    同步执行 TTS 合成 → 下载音频 → ffprobe 时长 → 句子时间戳计算，
+    不创建 LipsyncJob、不转存 OSS，直接返回 CosyVoice 临时 URL（~24h 有效）。
+    耗时约 2-3 秒。
+    """
+    try:
+        result = svc.preview_tts(
+            user_id=current_user.user.id,
+            voice_id=body.voice_id,
+            script_text=body.script_text,
+            speed=body.speed,
+            emotion=body.emotion,
+        )
+    except MediaKitError as exc:
+        status_code = 400
+        if exc.code in ("VoiceForbidden",):
+            status_code = 403
+        elif exc.code in ("TTSNoAudio",):
+            status_code = 502
+        raise HTTPException(
+            status_code=status_code,
+            detail={
+                "code": exc.code,
+                "message": str(exc),
+            },
+        ) from exc
+    except Exception as exc:
+        logger.error("TTS 预合成异常: %s", exc, exc_info=True)
+        raise HTTPException(
+            status_code=400,
+            detail=f"TTS 合成失败: {exc}",
+        ) from exc
+
+    return result
+
+
 # ── GET /jobs — 任务列表 ─────────────────────────────────────────────────
 
 
@@ -130,7 +163,7 @@ def list_lipsync_jobs(
 ):
     """获取对口型任务列表."""
     items, total = svc.list_jobs(
-        user_id=current_user.id,
+        user_id=current_user.user.id,
         project_id=project_id,
         status=status,
         offset=offset,
@@ -150,13 +183,18 @@ def list_lipsync_jobs(
 @router.get("/jobs/{job_id}", response_model=LipsyncJobResponse)
 def get_lipsync_job(
     job_id: str,
+    background: BackgroundTasks,
     current_user: AuthenticatedUser = Depends(get_current_user),
     svc: LipsyncService = Depends(_get_service),
 ):
     """获取对口型任务详情."""
-    job = svc.get_job(job_id, current_user.id)
+    job = svc.get_job(job_id, current_user.user.id)
     if job is None:
         raise HTTPException(status_code=404, detail="任务不存在")
+
+    if job.status not in ("completed", "failed"):
+        background.add_task(svc.refresh_job_status, job_id, current_user.user.id)
+
     return job
 
 
@@ -170,7 +208,7 @@ def refresh_lipsync_job(
     svc: LipsyncService = Depends(_get_service),
 ):
     """从 MediaKit 拉取最新状态并更新."""
-    job = svc.refresh_job_status(job_id, current_user.id)
+    job = svc.refresh_job_status(job_id, current_user.user.id)
     if job is None:
         raise HTTPException(status_code=404, detail="任务不存在")
     return job
@@ -185,13 +223,13 @@ def cancel_lipsync_job(
     current_user: AuthenticatedUser = Depends(get_current_user),
     svc: LipsyncService = Depends(_get_service),
 ):
-    """取消对口型任务（仅 pending/submitted 状态可取消）."""
-    job = svc.cancel_job(job_id, current_user.id)
+    """取消对口型任务（仅 pending/tts_processing/submitted 状态可取消）."""
+    job = svc.cancel_job(job_id, current_user.user.id)
     if job is None:
         raise HTTPException(status_code=404, detail="任务不存在")
     if job.status != "cancelled":
         raise HTTPException(
             status_code=400,
-            detail=f"任务状态 {job.status} 不可取消，仅 pending/submitted 可取消",
+            detail=f"任务状态 {job.status} 不可取消，仅 pending/tts_processing/submitted 可取消",
         )
     return job

apps/api/app/api/routes/templates_editor/ai_features.py

@@ -65,8 +65,8 @@ def _build_asset_analyses(
                 if url:
                     video_urls.append(url)
                     valid_asset_ids.append(aid)
-            except Exception as e:
-                logger.warning("获取素材URL失败: asset_id=%s error=%s", aid, str(e))
+            except Exception:
+                logger.exception("获取素材URL失败: asset_id=%s", aid)
 
         if not video_urls:
             logger.info("无可用视频素材，跳过视频理解分析")
@@ -108,7 +108,7 @@ def _build_asset_analyses(
         return analyses
 
     except Exception as e:
-        logger.warning("MediaKit 视频理解异常，将降级到无分析模式: %s", str(e))
+        logger.exception("MediaKit 视频理解异常，将降级到无分析模式: %s", e)
         return {}
 
 
@@ -177,7 +177,7 @@ def editor_ai_recommend(
         try:
             db.rollback()
         except Exception:
-            pass
+            logger.exception("db rollback failed in ai_recommend")
         raise HTTPException(
             status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
             detail="AI推荐结果保存失败，请稍后重试",

apps/api/app/api/routes/templates_editor/clips.py

@@ -132,8 +132,8 @@ def _build_asset_url_map(
     result: dict[str, str | None] = {}
     try:
         storage = get_storage_service()
-    except Exception:
-        logger.warning("获取存储服务失败，跳过asset_url生成")
+    except Exception as e:
+        logger.exception("获取存储服务失败，跳过asset_url生成: %s", e)
         return {aid: None for aid in asset_ids}
 
     # 批量查询所有 Asset（单次 SQL IN 查询，避免 N+1）
@@ -141,7 +141,7 @@ def _build_asset_url_map(
         assets = asset_repo.find_by_ids(unique_ids)
         asset_map = {a.id: a for a in assets}
     except Exception:
-        logger.warning("批量查询素材失败: asset_ids=%s", asset_ids, exc_info=True)
+        logger.exception("批量查询素材失败: asset_ids=%s", asset_ids)
         return {aid: None for aid in asset_ids if aid}
 
     for aid in unique_ids:
@@ -156,7 +156,7 @@ def _build_asset_url_map(
                 continue
             result[aid] = storage.get_download_url(storage_key, expires_seconds=3600)
         except Exception:
-            logger.warning("生成素材签名URL失败: asset_id=%s", aid, exc_info=True)
+            logger.exception("生成素材签名URL失败: asset_id=%s", aid)
             result[aid] = None
 
     return result
@@ -486,8 +486,8 @@ def _get_mediakit_recommendations(
                 if url:
                     video_urls.append(url)
                     valid_asset_ids.append(asset_id)
-            except Exception as e:
-                logger.warning("获取素材URL失败: asset_id=%s error=%s", asset_id, e)
+            except Exception:
+                logger.exception("获取素材URL失败: asset_id=%s", asset_id)
 
         if not video_urls:
             return {}
@@ -563,7 +563,7 @@ def _get_mediakit_recommendations(
         return recommendations
 
     except Exception as e:
-        logger.warning("MediaKit 智能选片异常，降级为随机选择: %s", e)
+        logger.exception("MediaKit 智能选片异常，降级为随机选择: %s", e)
         return {}
 
 
@@ -860,7 +860,7 @@ def create_clips_from_assets_editor(
     duplicate_warning = None
     if dup_rate > 50:
         duplicate_warning = f"查重率 {dup_rate:.1f}% 超过50%，建议更换素材或模板"
-        logger.warning(
+        logger.exception(
             "from-assets 成片查重率超标: plan_id=%s dup_rate=%.1f%%",
             plan_id,
             dup_rate,
@@ -960,8 +960,8 @@ def _update_mediakit_recommendations_async(  # pragma: no cover
         # 尝试获取存储服务（用于生成视频 URL）
         try:
             storage = get_storage_service()
-        except Exception:
-            logger.warning("后台任务: 获取存储服务失败，跳过 SceneChange 更新")
+        except Exception as e:
+            logger.exception("后台任务: 获取存储服务失败，跳过 SceneChange 更新: %s", e)
             return
 
         # 获取 MediaKit 客户端
@@ -987,8 +987,8 @@ def _update_mediakit_recommendations_async(  # pragma: no cover
             if storage_key and mime.startswith("video/"):
                 try:
                     video_url = storage.get_download_url(storage_key)
-                except Exception as e:
-                    logger.warning("后台任务: 获取素材URL失败: asset_id=%s error=%s", asset_id, e)
+                except Exception:
+                    logger.exception("后台任务: 获取素材URL失败: asset_id=%s", asset_id)
 
             # 构建该素材的占用区间列表（排除已更新片段）
             def _get_other_segments(asset_id_inner, clip_id_inner):
@@ -1039,12 +1039,11 @@ def _update_mediakit_recommendations_async(  # pragma: no cover
                             asset_id,
                             len(scene_changes),
                         )
-                    except Exception as cache_err:
+                    except Exception:
                         # 缓存写入失败不影响本次片段更新
-                        logger.warning(
-                            "后台任务: 场景点缓存写入失败: asset_id=%s error=%s",
+                        logger.exception(
+                            "后台任务: 场景点缓存写入失败: asset_id=%s",
                             asset_id,
-                            cache_err,
                         )
 
             # SceneChange 未获得有效结果 → 尝试 analyze_videos 作为 fallback
@@ -1124,11 +1123,10 @@ def _update_mediakit_recommendations_async(  # pragma: no cover
                                 recommended_start + clip_duration,
                                 plan_id,
                             )
-                    except Exception as me:
-                        logger.warning(
-                            "后台任务: 同步素材区间记录失败，回滚本次片段更新: clip_id=%s error=%s",
+                    except Exception:
+                        logger.exception(
+                            "后台任务: 同步素材区间记录失败，回滚本次片段更新: clip_id=%s",
                             clip.id,
-                            me,
                         )
                         db.rollback()
                         continue
@@ -1144,8 +1142,8 @@ def _update_mediakit_recommendations_async(  # pragma: no cover
                         asset_id,
                         recommended_start,
                     )
-                except Exception as ue:
-                    logger.warning("后台任务: 单个片段更新失败: clip_id=%s error=%s", clip.id, ue)
+                except Exception:
+                    logger.exception("后台任务: 单个片段更新失败: clip_id=%s", clip.id)
                     try:
                         db.rollback()
                     except Exception:
@@ -1154,9 +1152,9 @@ def _update_mediakit_recommendations_async(  # pragma: no cover
 
         logger.info("后台任务完成: plan_id=%s 成功更新 %d 个片段", plan_id, updated_count)
 
-    except Exception as e:
+    except Exception:
         # 后台任务失败不影响已创建的片段，静默处理
-        logger.warning("后台任务异常: plan_id=%s error=%s", plan_id, e, exc_info=True)
+        logger.exception("后台任务异常: plan_id=%s", plan_id)
         if db:
             try:
                 db.rollback()

apps/api/app/api/routes/tts.py

@@ -173,6 +173,14 @@ def synthesize(
         # job.voice_id 统一存解析后的 CosyVoice voice_id
         actual_voice_id = resolved_profile.voice_id
 
+    # 语速/情绪等合成参数随 metadata 落库，workflow 提交 CosyVoice 时读取透传
+    synthesis_meta = {
+        "speed": request.speed,
+        "emotion": request.emotion or "",
+    }
+    if request.metadata_:
+        synthesis_meta.update(request.metadata_)
+
     use_case = CreateTTSJobUseCase(repository)
     job = use_case.execute(
         user_id=user_id,
@@ -180,7 +188,7 @@ def synthesize(
         voice_id=actual_voice_id,
         voice_model=request.voice_model,
         voice_clone_profile_id=voice_clone_profile_id,
-        metadata=request.metadata_,
+        metadata=synthesis_meta,
     )
 
     # 提交 CosyVoice 合成任务
@@ -567,6 +575,7 @@ def preview_tts(
             text=request.text,
             voice_id=actual_voice_id,
             speed=request.speed,
+            emotion=request.emotion,
         )
     except CosyVoiceError as e:
         raise HTTPException(

apps/api/app/api/routes/voice_clones.py

@@ -163,12 +163,12 @@ def create_voice_clone(
             celery_app.send_task("worker.process_voice_clone", args=[profile.id])
             logger.info(f"Celery task dispatched for voice clone {profile.id}")
         except Exception as e:
-            logger.error(f"Failed to dispatch Celery task: {e}")
+            logger.exception("Failed to dispatch Celery task")
             # P2-3: Celery 调度失败时标记 profile 为 failed，避免永久卡在 processing
             try:
                 workflow.process_clone_failure(profile.id, f"Celery 任务调度失败: {e}")
-            except Exception as inner_e:
-                logger.error(f"Failed to mark profile as failed after dispatch error: {inner_e}")
+            except Exception:
+                logger.exception("Failed to mark profile as failed after dispatch error")
 
     return _to_response(profile)
 
@@ -277,12 +277,12 @@ def retry_voice_clone(
             celery_app.send_task("worker.process_voice_clone", args=[profile.id])
             logger.info(f"Celery task dispatched for voice clone retry {profile.id}")
         except Exception as e:
-            logger.error(f"Failed to dispatch Celery task: {e}")
+            logger.exception("Failed to dispatch Celery task")
             # P2-3: Celery 调度失败时标记 profile 为 failed，避免永久卡在 processing
             try:
                 workflow.process_clone_failure(profile.id, f"Celery 任务调度失败: {e}")
-            except Exception as inner_e:
-                logger.error(f"Failed to mark profile as failed after dispatch error: {inner_e}")
+            except Exception:
+                logger.exception("Failed to mark profile as failed after dispatch error")
 
     return _to_response(profile)
 

apps/api/app/api/routes/voices.py

@@ -105,8 +105,8 @@ def _resolve_preset_preview_url(
         _preset_preview_cache[voice_id] = (audio_url, time.time())
         logger.info("Preset voice preview generated: %s", voice_id)
         return audio_url
-    except Exception as e:
-        logger.warning("Failed to generate preview for %s, using fallback: %s", voice_id, e)
+    except Exception:
+        logger.exception("Failed to generate preset voice preview: voice_id=%s", voice_id)
         return fallback_url
 
 
@@ -127,6 +127,7 @@ def _resolve_all_preset_preview_urls(
         try:
             result_map[p.voice_id] = _resolve_preset_preview_url(p.voice_id, p.preview_url, cosyvoice)
         except Exception:
+            logger.exception("Failed to resolve preset preview URL: voice_id=%s", p.voice_id)
             result_map[p.voice_id] = p.preview_url
     return result_map
 
@@ -732,7 +733,7 @@ def _find_or_create_voice_library_for_extract(*, user_id, project_repository, as
             try:
                 session.rollback()
             except Exception:
-                pass
+                logger.exception("session rollback failed in _find_or_create_voice_library")
         for lib in asset_library_repository.find_by_project(project.id):
             kind = lib.kind.value if hasattr(lib.kind, "value") else lib.kind
             if kind == AssetLibraryKind.VOICE.value:

apps/api/app/schemas/ai_avatar_render.py

@@ -50,9 +50,11 @@ class CreateAiAvatarRenderRequest(BaseModel):
     """创建渲染任务请求."""
 
     lipsync_job_id: str = Field(..., description="对口型任务 ID")
-    script_id: str = Field(..., description="文案 ID")
+    script_id: str = Field("", description="文案 ID（选自文案库时传；手动输入文案直生场景可留空）")
     b_roll_segments: list[BRollSegment] = Field(default_factory=list, description="B-roll 片段列表")
-    title_config: dict[str, Any] = Field(default_factory=dict, description="标题配置")
+    title_config: dict[str, Any] = Field(
+        default_factory=dict, description="标题配置（可含 title_image_dataurl：前端 Canvas 渲染的标题 PNG dataURL）"
+    )
     cover_config: dict[str, Any] = Field(default_factory=dict, description="封面配置")
     project_id: str = Field("", description="项目 ID")
 
@@ -67,10 +69,7 @@ class CreateAiAvatarRenderRequest(BaseModel):
     @field_validator("script_id")
     @classmethod
     def validate_script_id(cls, v: str) -> str:
-        v = v.strip()
-        if not v:
-            raise ValueError("script_id 不能为空")
-        return v
+        return (v or "").strip()
 
 
 class AiAvatarRenderJobResponse(BaseModel):
@@ -80,7 +79,7 @@ class AiAvatarRenderJobResponse(BaseModel):
     user_id: str
     project_id: str
     lipsync_job_id: str
-    script_id: str
+    script_id: str = ""
     b_roll_segments: list[dict[str, Any]]
     title_config: dict[str, Any]
     cover_config: dict[str, Any]
@@ -109,3 +108,11 @@ class AiAvatarRenderProgressResponse(BaseModel):
     output_cover_url: str
     output_duration: float
     error_message: str
+
+
+class SmartCoverResponse(BaseModel):
+    """智能封面响应（封面从最终成片抽帧，不再叠加标题）."""
+
+    cover_url: str = Field("", description="封面图公网 URL（OSS，非临时）；失败为空")
+    status: str = Field("completed", description="completed / fallback_failed")
+    message: str = Field("", description="失败原因（如有）")

apps/api/app/schemas/lipsync.py

@@ -1,11 +1,20 @@
-"""对口型 API Schema 定义 — #1796, #1809 参数调整."""
+"""对口型 API Schema 定义 — #1796 / #1809 / #1822 / #1845（配音前置）.
+
+支持三种输入模式：
+1. TTS 直生模式（兼容旧版前端）：传 voice_id + script_text（+ speed/emotion），
+   后端 Celery 异步做 TTS 合成 + MediaKit 提交。
+2. 直接音频模式：传 video_url + audio_url（音频已由调用方准备好）。
+3. 预合成音频模式（#1845 配音前置新主路径）：前端先调 POST /lipsync/tts-preview
+   拿到 audio_url + sentence_timings，再在 create_job 时传 audio_url + audio_duration
+   + sentence_timings，后端跳过 TTS 和时间戳计算，直接 ffprobe 校验后提交 MediaKit。
+"""
 
 from __future__ import annotations
 
 from datetime import datetime
 from typing import Optional
 
-from pydantic import BaseModel, Field, field_validator
+from pydantic import BaseModel, Field, model_validator
 
 
 class LipsyncJobResponse(BaseModel):
@@ -17,12 +26,17 @@ class LipsyncJobResponse(BaseModel):
     video_url: str
     audio_url: str
     enable_video_loop: bool
+    voice_id: str = ""
+    script_text: str = ""
+    speed: float = 1.0
+    emotion: str = ""
     mediakit_task_id: str
     status: str
     output_video_url: str
     output_duration: float
     error_message: str
     error_code: str
+    sentence_timings: Optional[list] = None
     submitted_at: Optional[datetime] = None
     completed_at: Optional[datetime] = None
     created_at: datetime
@@ -33,45 +47,84 @@ class LipsyncJobResponse(BaseModel):
 
 
 class CreateLipsyncJobRequest(BaseModel):
-    """创建对口型任务请求 — #1809.
+    """创建对口型任务请求.
 
-    前端传 {voice_id, script_text, video_url}，
-    后端内部调 TTS 生成 audio_url 再提交 MediaKit。
+    三种模式（三选一）：
+    - TTS 直生（旧版/降级）：voice_id + script_text 必填；audio_url 留空。
+    - 直接音频：video_url + audio_url 必填。
+    - 预合成音频（#1845 新主路径）：audio_url 必填 + 可选 audio_duration/sentence_timings；
+      后端同步 ffprobe 校验时长、写入 timings，直接提交 MediaKit。
     """
 
     video_url: str = Field(..., description="人物视频 URL（MP4，≤30min，单人真人）")
-    voice_id: str = Field(..., description="音色 ID（预设音色或克隆音色 profile ID）")
-    script_text: str = Field(..., description="要合成的脚本文本")
-    enable_video_loop: bool = Field(False, description="音频长于视频时是否循环画面")
+
+    # 模式 2/3：直接/预合成音频
+    audio_url: str = Field("", description="驱动音频 URL（mp3/aac/wav/m4a/flac）；直生模式留空")
+    audio_duration: Optional[float] = Field(None, ge=0, description="预合成音频时长（秒），可选；后端会 ffprobe 校验")
+    sentence_timings: Optional[list] = Field(None, description="预合成接口返回的句子时间戳，可选；若传入则直接写入 job")
+
+    # 模式 1：TTS 直生
+    voice_id: str = Field("", description="音色 ID（预置音色或克隆音色 profile UUID）")
+    script_text: str = Field("", description="要合成的文案（直生模式必填，最长 5000 字符）")
+    speed: float = Field(1.0, ge=0.5, le=2.0, description="语速（0.5-2.0），默认 1.0")
+    emotion: str = Field("", description="情绪（natural/excited/calm/friendly 或中文 自然/兴奋/沉稳/亲切）")
+
+    enable_video_loop: bool = Field(
+        True, description="音频长于视频时是否循环画面（AI数字人默认开启，防止音频长于视频被截断）"
+    )
     project_id: str = Field("", description="项目 ID（可选）")
 
-    @field_validator("video_url")
-    @classmethod
-    def validate_video_url(cls, v: str) -> str:
-        v = v.strip()
-        if not v:
+    @model_validator(mode="after")
+    def _validate_input_mode(self) -> "CreateLipsyncJobRequest":
+        video = (self.video_url or "").strip()
+        if not video:
             raise ValueError("video_url 不能为空")
-        if not v.startswith(("http://", "https://")):
+        if not video.startswith(("http://", "https://")):
             raise ValueError("video_url 必须是 HTTP/HTTPS URL")
-        lower = v.lower().split("?")[0]
+        lower = video.lower().split("?")[0]
         if not lower.endswith(".mp4"):
             raise ValueError("video_url 仅支持 MP4 格式")
-        return v
 
-    @field_validator("voice_id")
-    @classmethod
-    def validate_voice_id(cls, v: str) -> str:
-        v = v.strip()
-        if not v:
-            raise ValueError("voice_id 不能为空")
-        return v
+        has_audio = bool((self.audio_url or "").strip())
+        has_tts = bool((self.voice_id or "").strip()) and bool((self.script_text or "").strip())
 
-    @field_validator("script_text")
-    @classmethod
-    def validate_script_text(cls, v: str) -> str:
-        v = v.strip()
-        if not v:
-            raise ValueError("script_text 不能为空")
-        if len(v) > 5000:
+        if not has_audio and not has_tts:
+            raise ValueError(
+                "必须提供驱动音频：要么传 audio_url（直接/预合成音频模式），"
+                "要么同时传 voice_id + script_text（TTS 直生模式）"
+            )
+
+        if has_tts and len(self.script_text) > 5000:
             raise ValueError("script_text 最长 5000 字符")
-        return v
+
+        if has_audio:
+            au = self.audio_url.strip()
+            if not au.startswith(("http://", "https://")):
+                raise ValueError("audio_url 必须是 HTTP/HTTPS URL")
+            au_lower = au.lower().split("?")[0]
+            allowed = (".mp3", ".aac", ".wav", ".m4a", ".flac")
+            if not any(au_lower.endswith(ext) for ext in allowed):
+                raise ValueError(f"audio_url 格式不支持，仅支持: {', '.join(allowed)}")
+            self.audio_url = au
+
+        return self
+
+
+# ── #1845 TTS 预合成接口 ────────────────────────────────────────────────
+
+
+class AiAvatarTtsPreviewRequest(BaseModel):
+    """步骤1「生成配音」预合成请求（同步 HTTP，~2-3s）."""
+
+    voice_id: str = Field(..., min_length=1, max_length=128, description="音色 ID")
+    script_text: str = Field(..., min_length=1, max_length=5000, description="要合成的文案")
+    speed: float = Field(1.0, ge=0.5, le=2.0, description="语速（0.5-2.0），默认 1.0")
+    emotion: str = Field("natural", max_length=32, description="情绪")
+
+
+class AiAvatarTtsPreviewResponse(BaseModel):
+    """TTS 预合成响应（临时 URL，24h 内有效，足够当前会话使用）."""
+
+    audio_url: str = Field(..., description="CosyVoice 临时音频 URL")
+    duration: float = Field(..., ge=0, description="音频总时长（秒），ffprobe 测得")
+    sentence_timings: list[dict] = Field(..., description="句子级精确时间戳")

apps/api/app/schemas/tts.py

@@ -16,6 +16,7 @@ class TTSSynthesizeRequest(BaseModel):
     output_name: str = Field("", description="输出文件名")
     language: str = Field("zh-CN", description="语言")
     speed: float = Field(1.0, ge=0.5, le=2.0, description="语速")
+    emotion: str = Field("", description="情绪（natural/excited/calm/friendly，或中文 自然/兴奋/沉稳/亲切）")
     voice_model: str = Field("", description="语音模型名称")
     voice_clone_profile_id: str = Field("", description="关联的音色克隆档案 ID")
     format: str = Field("mp3", description="输出格式（mp3/wav/pcm）")
@@ -109,6 +110,7 @@ class TTSPreviewRequest(BaseModel):
     text: str = Field(..., min_length=1, max_length=200, description="合成文本，限制 200 字")
     voice_id: str = Field(..., min_length=1, description="音色 ID")
     speed: float = Field(1.0, ge=0.5, le=2.0, description="语速")
+    emotion: str = Field("", description="情绪（natural/excited/calm/friendly，或中文）")
     pitch: float = Field(1.0, ge=0.5, le=2.0, description="音调（预留，当前未使用）")
 
 

apps/api/app/services/ai_avatar_cover_service.py

@@ -0,0 +1,215 @@
+"""AI 数字人封面服务 — MediaKit 抽帧 + 质量评分选最佳帧 + 转存 OSS.
+
+与 generation_cover.py 的智能选帧能力对齐（不再用 FFmpeg 简单截帧）：
+1. MediaKit extract_frames 抽取多帧（默认 5 帧，SpecifiedFrames 策略）
+2. cover_frame_scorer.score_frames 按清晰度/亮度/色彩评分选最佳
+3. 下载最佳帧并转存 OSS，返回公网封面 URL
+
+设计原则：封面一律从最终成片（已叠加标题/B-roll）抽帧，帧本身已含标题，
+本服务**不再叠加标题**。对口型阶段的裸视频封面入口已删除（废弃）。
+
+降级：MediaKit 不可用或抽帧失败时返回空字符串，由调用方决定回退策略。
+"""
+
+from __future__ import annotations
+
+import logging
+import tempfile
+import uuid
+from pathlib import Path
+from typing import Optional
+from urllib.parse import urlparse
+
+logger = logging.getLogger(__name__)
+
+# MediaKit 抽帧轮询参数：poll_interval=1s × max_poll=15 → 最长 15s，配合前端 120s 超时足够
+COVER_POLL_INTERVAL = 1.0
+COVER_MAX_POLL_ATTEMPTS = 15
+
+# 帧图片下载超时（秒）
+FRAME_DOWNLOAD_TIMEOUT = 20
+# 最佳帧下载超时（用于 persist）
+BEST_FRAME_DOWNLOAD_TIMEOUT = 30
+
+# 自家 OSS 私有桶 URL 重签有效期（供 MediaKit GPU worker 拉取）
+MEDIAKIT_URL_TTL_SECONDS = 7 * 24 * 3600
+
+
+def _sign_video_url_for_mediakit(video_url: str) -> str:
+    """如果 video_url 是自家 OSS 私有桶 URL，重新签名为长有效期预签名 URL。
+
+    MediaKit GPU worker 需要能公网访问 video_url，裸 public_url 在私有桶下会 403。
+    """
+    if not video_url:
+        return video_url
+    try:
+        from packages.shared.storage import get_shared_storage_service
+
+        storage = get_shared_storage_service()
+        public_base = getattr(storage, "public_url", "")
+        if not isinstance(public_base, str) or not public_base:
+            return video_url
+        own_host = urlparse(public_base).netloc.lower()
+        url_host = urlparse(video_url).netloc.lower()
+        if own_host and url_host == own_host:
+            signed = storage.get_download_url(video_url, expires_seconds=MEDIAKIT_URL_TTL_SECONDS)
+            if signed:
+                logger.info("[数字人封面] video_url 已重签（自家 OSS 私有桶）")
+                return signed
+    except Exception:
+        logger.warning("[数字人封面] video_url 重签失败，使用原始 URL", exc_info=True)
+    return video_url
+
+
+def select_best_cover_frame(video_url: str, *, max_frames: int = 5) -> str:
+    """从视频抽取多帧并评分选最佳帧，返回最佳帧的临时 URL."""
+    if not video_url:
+        return ""
+
+    video_url = _sign_video_url_for_mediakit(video_url)
+
+    try:
+        from packages.shared.cover_frame_scorer import score_frames
+        from packages.shared.mediakit_client import get_mediakit_client
+
+        mk = get_mediakit_client()
+        if not mk.is_available:
+            logger.warning("[数字人封面] MediaKit 未配置，无法智能抽帧")
+            return ""
+
+        logger.info(
+            "[数字人封面] 开始抽帧: video_url=%s max_frames=%d",
+            video_url[:80],
+            max_frames,
+        )
+
+        snapshots = mk.extract_frames(
+            video_url=video_url,
+            strategy="SpecifiedFrames",
+            max_frames=max_frames,
+            poll_interval=COVER_POLL_INTERVAL,
+            max_poll_attempts=COVER_MAX_POLL_ATTEMPTS,
+            max_retries=1,
+        )
+        if not snapshots:
+            logger.warning("[数字人封面] MediaKit 未返回帧: %s", video_url[:80])
+            return ""
+
+        if len(snapshots) == 1:
+            return snapshots[0].get("image_url") or snapshots[0].get("url") or ""
+
+        import httpx
+
+        candidates = []
+        with httpx.Client(timeout=FRAME_DOWNLOAD_TIMEOUT, follow_redirects=True) as client:
+            for snap in snapshots:
+                url = snap.get("image_url") or snap.get("url") or ""
+                if not url:
+                    continue
+                tmp_path: Optional[str] = None
+                try:
+                    resp = client.get(url)
+                    resp.raise_for_status()
+                    with tempfile.NamedTemporaryFile(suffix=".jpg", delete=False) as tmp:
+                        tmp.write(resp.content)
+                        tmp_path = tmp.name
+                    candidates.append({"image_path": tmp_path, "url": url})
+                except Exception as e:
+                    logger.warning("[数字人封面] 帧下载失败，跳过: url=%s err=%s", url[:80], e)
+                    candidates.append({"image_path": None, "url": url, "score": 0.0})
+
+        if not candidates:
+            return snapshots[0].get("image_url") or snapshots[0].get("url") or ""
+
+        scored = score_frames(candidates)
+        best = scored[0] if scored else None
+        best_url = best.get("url", "") if best else ""
+
+        for c in candidates:
+            p = c.get("image_path")
+            if p:
+                try:
+                    Path(p).unlink(missing_ok=True)
+                except Exception:
+                    pass
+
+        logger.info(
+            "[数字人封面] 智能选帧完成: candidates=%d best_score=%s",
+            len(candidates),
+            best.get("score") if best else "n/a",
+        )
+        return best_url
+
+    except Exception:
+        logger.warning("[数字人封面] 智能选帧失败", exc_info=True)
+        return ""
+
+
+def persist_cover_to_oss(
+    frame_url: str,
+    *,
+    job_id: str = "",
+    prefix: str = "ai-avatar/covers",
+) -> str:
+    """下载最佳帧图并转存到 OSS，返回公网封面 URL（预签名）.
+
+    封面来自最终成片抽帧，帧本身已含标题，本函数不再做任何文字/图片叠加。
+    """
+    if not frame_url:
+        return ""
+    tmp_path: Optional[str] = None
+    try:
+        import httpx
+
+        with httpx.Client(timeout=BEST_FRAME_DOWNLOAD_TIMEOUT, follow_redirects=True) as client:
+            resp = client.get(frame_url)
+            resp.raise_for_status()
+            if not resp.content:
+                logger.warning("[数字人封面] 帧图内容为空: %s", frame_url[:80])
+                return frame_url
+
+            with tempfile.NamedTemporaryFile(suffix=".jpg", delete=False) as tmp:
+                tmp.write(resp.content)
+                tmp_path = tmp.name
+
+        from packages.shared.storage import get_shared_storage_service
+
+        storage = get_shared_storage_service()
+        token = job_id or uuid.uuid4().hex[:12]
+        cover_key = f"{prefix}/{token}/cover_{uuid.uuid4().hex[:8]}.jpg"
+
+        public_url = storage.upload_file(
+            file_or_path=tmp_path,
+            storage_key=cover_key,
+            content_type="image/jpeg",
+        )
+        logger.info("[数字人封面] 封面已转存 OSS: key=%s", cover_key)
+        if public_url:
+            signed = storage.get_download_url(cover_key, expires_seconds=86400)
+            return signed
+        return frame_url
+    except Exception:
+        logger.warning("[数字人封面] 封面转存 OSS 失败，返回原始 URL", exc_info=True)
+        return frame_url
+    finally:
+        if tmp_path:
+            try:
+                Path(tmp_path).unlink(missing_ok=True)
+            except Exception:
+                pass
+
+
+def generate_smart_cover(
+    video_url: str,
+    *,
+    job_id: str = "",
+    max_frames: int = 5,
+) -> str:
+    """一站式：MediaKit 智能抽帧选最佳 → 转存 OSS。失败返回空字符串。
+
+    封面从最终成片抽帧，不再叠加任何标题（帧本身已含）。
+    """
+    best_frame = select_best_cover_frame(video_url, max_frames=max_frames)
+    if not best_frame:
+        return ""
+    return persist_cover_to_oss(best_frame, job_id=job_id)

apps/api/app/services/ai_avatar_render_service.py

@@ -9,8 +9,11 @@
 
 from __future__ import annotations
 
+import base64
+import binascii
 import logging
 import os
+import subprocess
 import tempfile
 import uuid
 from datetime import datetime, timezone
@@ -24,8 +27,9 @@ from packages.adapters.sqlalchemy_impl.models import (
     ScriptModel,
 )
 from packages.domain.video_filter_builder import (
-    build_cover_extract_command,
+    build_broll_overlay_filter,
     build_title_drawtext_filter,
+    build_title_overlay_filter,
 )
 from packages.shared.storage import get_shared_storage_service
 
@@ -53,8 +57,8 @@ class AiAvatarRenderService:
         *,
         user_id: str,
         lipsync_job_id: str,
-        script_id: str,
-        b_roll_segments: list[dict[str, Any]],
+        script_id: str = "",
+        b_roll_segments: list[dict[str, Any]] | None = None,
         title_config: dict[str, Any],
         cover_config: dict[str, Any],
         project_id: str = "",
@@ -83,17 +87,19 @@ class AiAvatarRenderService:
         if not lipsync_job.output_video_url:
             raise AiAvatarRenderError("对口型任务输出视频 URL 为空", code="LipsyncJobNoOutput")
 
-        # 2. 验证文案归属
-        script = (
-            self.db.query(ScriptModel)
-            .filter(
-                ScriptModel.id == script_id,
-                ScriptModel.user_id == user_id,
+        # 2. 验证文案归属（仅当选了文案库条目时；手动输入文案直生场景 script_id 可空）
+        script_id = (script_id or "").strip()
+        if script_id:
+            script = (
+                self.db.query(ScriptModel)
+                .filter(
+                    ScriptModel.id == script_id,
+                    ScriptModel.user_id == user_id,
+                )
+                .first()
             )
-            .first()
-        )
-        if script is None:
-            raise AiAvatarRenderError("文案不存在或无权访问", code="ScriptNotFound")
+            if script is None:
+                raise AiAvatarRenderError("文案不存在或无权访问", code="ScriptNotFound")
 
         # 3. 创建渲染任务
         job_id = str(uuid.uuid4())
@@ -103,7 +109,7 @@ class AiAvatarRenderService:
             project_id=project_id,
             lipsync_job_id=lipsync_job_id,
             script_id=script_id,
-            b_roll_segments=[s if isinstance(s, dict) else s.model_dump() for s in b_roll_segments],
+            b_roll_segments=[s if isinstance(s, dict) else s.model_dump() for s in (b_roll_segments or [])],
             title_config=title_config,
             cover_config=cover_config,
             status="pending",
@@ -194,9 +200,8 @@ class AiAvatarRenderService:
         1. 下载对口型输出视频 (20%)
         2. 构建 FFmpeg 滤镜链 (40%)
         3. 执行 FFmpeg 渲染 (80%)
-        4. 提取封面 (90%)
-        5. 上传到 OSS (95%)
-        6. 更新任务状态 (100%)
+        4. 上传到 OSS (95%) — 封面不再自动生成，改由前端主动抽帧
+        5. 更新任务状态 (100%)
         """
         job = self.db.query(AiAvatarRenderJob).filter(AiAvatarRenderJob.id == job_id).first()
         if job is None:
@@ -226,27 +231,32 @@ class AiAvatarRenderService:
             self.db.commit()
 
             # 2. 构建 FFmpeg 滤镜链 (40%)
-            from packages.domain.video_filter_builder import build_broll_overlay_filter
+            # 用 ffprobe 探测输入视频分辨率，确保 B-roll 缩放与标题位置与实际输出一致。
+            # AI 数字人对口型输出为 9:16 竖屏，默认兜底 720x1280；探测失败时使用默认值不阻断渲染。
+            output_width, output_height = self._probe_video_resolution(input_video_path)
+            if output_width <= 0 or output_height <= 0:
+                output_width, output_height = 720, 1280
+                logger.info(
+                    "[数字人渲染] ffprobe 探测分辨率失败或无效，使用默认竖屏尺寸 %sx%s",
+                    output_width,
+                    output_height,
+                )
+            else:
+                logger.info("[数字人渲染] 探测输入视频分辨率: %sx%s", output_width, output_height)
 
-            filter_complex = build_broll_overlay_filter(
+            broll_filter, broll_label = build_broll_overlay_filter(
                 b_roll_segments=job.b_roll_segments,
                 video_duration=lipsync_job.output_duration,
+                output_width=output_width,
+                output_height=output_height,
             )
 
-            # 标题叠加
-            title_filter = build_title_drawtext_filter(job.title_config)
-            if title_filter:
-                if filter_complex:
-                    filter_complex += f"[vout]{title_filter}[vout_titled];"
-                else:
-                    filter_complex = f"[0:v]{title_filter}[vout_titled];"
-
-            # 清理末尾分号
-            if filter_complex.endswith(";"):
-                filter_complex = filter_complex[:-1]
-
-            # 最终输出标签
-            final_label = "vout_titled" if title_filter else ("vout" if filter_complex else None)
+            # 标题叠加路径：优先前端 Canvas 渲染的 PNG 图层（所见即所得），
+            # 无 title_image_dataurl 时降级到 drawtext 重画文字。
+            title_cfg = job.title_config if isinstance(job.title_config, dict) else {}
+            title_dataurl = (title_cfg or {}).get("title_image_dataurl") if title_cfg else None
+            use_title_png = isinstance(title_dataurl, str) and title_dataurl.startswith("data:image/")
+            title_input_index = 1 + len(job.b_roll_segments or []) if use_title_png else None
 
             job.progress = 40
             self.db.commit()
@@ -255,42 +265,126 @@ class AiAvatarRenderService:
             with tempfile.TemporaryDirectory() as tmpdir:
                 output_video_path = os.path.join(tmpdir, "output.mp4")
 
-                cmd = self._build_ffmpeg_command(
+                # 在临时目录里解码保存标题 PNG（with 退出自动清理）
+                title_png_path: Optional[str] = None
+                extra_inputs: list[str] = []
+                title_filter = None
+                if use_title_png:
+                    try:
+                        title_png_path = os.path.join(tmpdir, f"title_{job.id}.png")
+                        self._save_title_dataurl_to_file(title_dataurl, dst_path=title_png_path)
+                        extra_inputs.append(title_png_path)
+                        logger.info(
+                            "[数字人渲染] 标题 PNG 已保存: %s (input index %d)", title_png_path, title_input_index
+                        )
+                    except Exception as exc:
+                        logger.warning("[数字人渲染] 标题 PNG 解码/保存失败，降级 drawtext: %s", exc)
+                        title_png_path = None
+                        extra_inputs = []
+
+                # 构建标题滤镜
+                final_label = None
+                if title_png_path and title_input_index is not None:
+                    title_input_label = f"[{title_input_index}:v]"
+                    base_label = f"[{broll_label}]" if broll_label else "[0:v]"
+                    title_filter = build_title_overlay_filter(
+                        title_cfg,
+                        output_width=output_width,
+                        output_height=output_height,
+                        title_png_path=title_png_path,
+                        title_input_label=title_input_label,
+                        base_label=base_label,
+                        output_label="vout_titled",
+                    )
+                    if not title_filter:
+                        # build 返回 None → 文件不存在（极端并发情况），降级 drawtext
+                        title_png_path = None
+                        extra_inputs = []
+
+                if title_png_path:
+                    # overlay 路径
+                    if broll_filter and title_filter:
+                        filter_complex = broll_filter + f";{title_filter}"
+                    elif broll_filter:
+                        filter_complex = broll_filter
+                        final_label = broll_label
+                    elif title_filter:
+                        filter_complex = title_filter
+                    else:
+                        filter_complex = ""
+                    if title_filter:
+                        final_label = "vout_titled"
+                    elif not final_label:
+                        final_label = None
+                else:
+                    # 降级：drawtext 重画文字
+                    title_filter = build_title_drawtext_filter(
+                        title_cfg,
+                        output_width=output_width,
+                        output_height=output_height,
+                    )
+                    if broll_filter and title_filter:
+                        filter_complex = broll_filter + f";[{broll_label}]{title_filter}[vout_titled]"
+                        final_label = "vout_titled"
+                    elif broll_filter:
+                        filter_complex = broll_filter
+                        final_label = broll_label
+                    elif title_filter:
+                        filter_complex = f"[0:v]{title_filter}[vout_titled]"
+                        final_label = "vout_titled"
+                    else:
+                        filter_complex = ""
+                        final_label = None
+
+                cmd_list = self._build_ffmpeg_command(
                     input_video=input_video_path,
                     b_roll_segments=job.b_roll_segments,
+                    extra_inputs=extra_inputs,
                     filter_complex=filter_complex,
                     final_label=final_label,
                     output_path=output_video_path,
                 )
 
-                exit_code = os.system(cmd)
-                if exit_code != 0:
-                    raise AiAvatarRenderError(f"FFmpeg 渲染失败，退出码: {exit_code}", code="FFmpegFailed")
+                try:
+                    render_result = subprocess.run(
+                        cmd_list,
+                        capture_output=True,
+                        text=True,
+                        timeout=600,
+                    )
+                except subprocess.TimeoutExpired as exc:
+                    raise AiAvatarRenderError(
+                        "FFmpeg 渲染超时（600s）",
+                        code="FFmpegTimeout",
+                    ) from exc
+
+                if render_result.returncode != 0:
+                    stderr_tail = (render_result.stderr or "").strip()[-800:]
+                    raise AiAvatarRenderError(
+                        f"FFmpeg 渲染失败，退出码: {render_result.returncode}, stderr: {stderr_tail}",
+                        code="FFmpegFailed",
+                    )
 
                 job.progress = 80
                 self.db.commit()
 
-                # 4. 提取封面 (90%)
-                cover_path = ""
-                if job.cover_config:
-                    cover_path = os.path.join(tmpdir, "cover.jpg")
-                    cover_cmd = build_cover_extract_command(job.cover_config, cover_path)
-                    cover_cmd = cover_cmd.replace("INPUT_VIDEO", output_video_path)
-                    cover_exit = os.system(cover_cmd)
-                    if cover_exit != 0:
-                        logger.warning("封面提取失败，跳过: %s", cover_cmd)
-                        cover_path = ""
-
-                job.progress = 90
-                self.db.commit()
-
-                # 5. 上传到 OSS (95%)
+                # 4/5. 上传成片到 OSS (95%) —— 已砍掉自动抽封面逻辑（步骤⑤）；
+                # 封面由前端在渲染完成后通过 /smart-cover 接口主动从成片抽帧，不阻塞渲染链路。
                 output_video_url = self._upload_to_oss(output_video_path, f"ai-avatar/{job_id}/output.mp4")
                 job.output_video_url = output_video_url
 
-                if cover_path:
-                    output_cover_url = self._upload_to_oss(cover_path, f"ai-avatar/{job_id}/cover.jpg")
-                    job.output_cover_url = output_cover_url
+                # 封面透传：如果用户已在 cover_config 中选定封面 URL（mode=upload 的自定义上传 或
+                # mode=auto_frame 已有的智能封面结果），直接透传到 output_cover_url，不再重新截帧。
+                if isinstance(job.cover_config, dict):
+                    _pre_cover_url = (
+                        job.cover_config.get("url")
+                        or job.cover_config.get("imageUrl")
+                        or job.cover_config.get("cover_url")
+                        or ""
+                    )
+                    if _pre_cover_url:
+                        job.output_cover_url = _pre_cover_url
+                        logger.info("[数字人渲染] 使用用户已选定封面 URL: job_id=%s", job_id)
 
                 # 获取输出视频时长
                 job.output_duration = lipsync_job.output_duration
@@ -305,18 +399,56 @@ class AiAvatarRenderService:
             self.db.commit()
             logger.info("渲染任务完成: %s", job_id)
 
+            # 7. 自动保存成片记录到成片库
+            if job.output_video_url:
+                try:
+                    from packages.adapters.sqlalchemy_impl.generated_video_repository import (
+                        SQLAlchemyGeneratedVideoRepository,
+                    )
+                    from packages.domain.generated_video import GeneratedVideo
+
+                    clip_name = f"AI数字人_{job_id[:8]}"
+                    # AI数字人入口是独立页面，前端可能不传 project_id（无项目概念），
+                    # 兜底为 "ai_avatar" 避免 DB 非空约束/查询问题；generation_task_id 同样兜底用 render_job_id
+                    clip_project_id = (job.project_id or "").strip() or "ai_avatar"
+                    clip_generation_task_id = (job.lipsync_job_id or "").strip() or job_id
+                    clip = GeneratedVideo.create(
+                        project_id=clip_project_id,
+                        generation_task_id=clip_generation_task_id,
+                        name=clip_name,
+                        file_url=job.output_video_url,
+                        user_id=job.user_id,
+                        duration=job.output_duration or 0.0,
+                        thumbnail_url=job.output_cover_url or None,
+                        generation_params={
+                            "source": "ai_avatar_render",
+                            "render_job_id": job.id,
+                        },
+                    )
+                    video_repo = SQLAlchemyGeneratedVideoRepository(self.db)
+                    video_repo.create(clip)
+                    logger.info("成片记录已保存到成片库: clip_id=%s, render_job=%s", clip.id, job_id)
+                except Exception:
+                    logger.error(
+                        "自动保存成片记录失败（不影响渲染任务状态）: render_job=%s",
+                        job_id,
+                        exc_info=True,
+                    )
+
         except AiAvatarRenderError as exc:
             job.status = "failed"
             job.error_message = str(exc)
             job.updated_at = datetime.now(timezone.utc)
             self.db.commit()
             logger.error("渲染任务失败 [%s]: %s", job_id, exc)
+            raise
         except Exception as exc:
             job.status = "failed"
             job.error_message = f"渲染异常: {str(exc)}"
             job.updated_at = datetime.now(timezone.utc)
             self.db.commit()
             logger.exception("渲染任务异常 [%s]", job_id)
+            raise
 
     def _download_video(self, url: str) -> str:
         """下载视频到临时文件."""
@@ -334,32 +466,132 @@ class AiAvatarRenderService:
                 os.unlink(tmp.name)
             raise
 
+    @staticmethod
+    def _save_title_dataurl_to_file(dataurl: str, *, dst_path: str | None = None, job_id: str = "") -> str:
+        """解码前端传来的 data:image/png;base64,... 并保存为本地 PNG 文件。
+
+        Args:
+            dataurl: 完整 dataURL 字符串
+            dst_path: 指定输出路径；为 None 时创建临时文件并返回路径
+            job_id: 仅在 dst_path 为空时用于临时文件命名
+
+        Returns:
+            保存后的本地文件路径
+        """
+        if not isinstance(dataurl, str) or not dataurl.startswith("data:image/"):
+            raise ValueError("title_image_dataurl 不是合法的 data:image URL")
+        # 拆分 data:image/png;base64,<payload>
+        try:
+            header, b64 = dataurl.split(",", 1)
+        except ValueError as exc:
+            raise ValueError("title_image_dataurl 缺少 base64 payload") from exc
+        if "base64" not in header:
+            raise ValueError("title_image_dataurl 不是 base64 编码")
+        try:
+            png_bytes = base64.b64decode(b64, validate=True)
+        except (binascii.Error, ValueError) as exc:
+            raise ValueError(f"title_image_dataurl base64 解码失败: {exc}") from exc
+        if not png_bytes:
+            raise ValueError("title_image_dataurl 解码后为空")
+
+        if dst_path:
+            out_path = dst_path
+            with open(out_path, "wb") as f:
+                f.write(png_bytes)
+            return out_path
+        suffix = f"_title_{job_id}.png" if job_id else "_title.png"
+        with tempfile.NamedTemporaryFile(suffix=suffix, delete=False) as tmp:
+            tmp.write(png_bytes)
+            return tmp.name
+
+    @staticmethod
+    def _probe_video_resolution(video_path: str) -> tuple[int, int]:
+        """用 ffprobe 探测视频分辨率，返回 (width, height)；失败返回 (0, 0)。"""
+        try:
+            result = subprocess.run(
+                [
+                    "ffprobe",
+                    "-v",
+                    "error",
+                    "-select_streams",
+                    "v:0",
+                    "-show_entries",
+                    "stream=width,height",
+                    "-of",
+                    "csv=p=0:s=x",
+                    video_path,
+                ],
+                capture_output=True,
+                text=True,
+                timeout=15,
+            )
+            if result.returncode == 0 and result.stdout.strip():
+                parts = result.stdout.strip().split("x")
+                if len(parts) == 2:
+                    w, h = int(parts[0]), int(parts[1])
+                    if w > 0 and h > 0:
+                        return w, h
+        except Exception as exc:
+            logger.warning("[数字人渲染] ffprobe 探测分辨率失败: %s", exc)
+        return 0, 0
+
     def _build_ffmpeg_command(
         self,
         *,
         input_video: str,
         b_roll_segments: list[dict[str, Any]],
+        extra_inputs: list[str] | None = None,
         filter_complex: str,
         final_label: Optional[str],
         output_path: str,
-    ) -> str:
-        """构建 FFmpeg 命令."""
-        # 输入文件
-        inputs = f"-i {input_video}"
+    ) -> list[str]:
+        """构建 FFmpeg 命令（list 形式，shell=False）.
+
+        根因修复 #1798 P0：OSS 预签名 URL 含 `&Expires=...&Signature=...` 特殊字符，
+        os.system(shell=True) 会把 `&` 解释为后台命令分隔符，导致 -filter_complex 被
+        当成独立命令报 sh: -filter_complex: not found（exit 127 → Python 32512）。
+        list + shell=False 彻底规避 shell 转义问题。
+        """
+        cmd: list[str] = ["ffmpeg", "-i", input_video]
         for seg in b_roll_segments:
             asset_url = seg.get("asset_url", "")
             if asset_url:
-                inputs += f" -i {asset_url}"
+                cmd.extend(["-i", asset_url])
+        # 额外输入（例如前端 Canvas 渲染的标题 PNG）
+        for extra in extra_inputs or []:
+            cmd.extend(["-i", extra])
 
-        # 滤镜
         if filter_complex and final_label:
-            filter_arg = f'-filter_complex "{filter_complex}" -map "[{final_label}]"'
+            cmd.extend(
+                [
+                    "-filter_complex",
+                    filter_complex,
+                    "-map",
+                    f"[{final_label}]",
+                    "-map",
+                    "0:a?",
+                ]
+            )
         elif filter_complex:
-            filter_arg = f'-filter_complex "{filter_complex}"'
-        else:
-            filter_arg = ""
+            cmd.extend(["-filter_complex", filter_complex])
 
-        return f"ffmpeg {inputs} {filter_arg} -c:v libx264 -preset fast -crf 23 -y {output_path}"
+        cmd.extend(
+            [
+                "-c:v",
+                "libx264",
+                "-preset",
+                "veryfast",
+                "-crf",
+                "23",
+                "-c:a",
+                "aac",
+                "-b:a",
+                "128k",
+                "-y",
+                output_path,
+            ]
+        )
+        return cmd
 
     def _upload_to_oss(self, local_path: str, oss_key: str) -> str:
         """上传文件到 OSS，返回 URL.

apps/api/app/services/lipsync_service.py

@@ -1,19 +1,25 @@
-"""对口型 Service — #1796 MediaKit 对口型业务逻辑, #1809 参数调整.
+"""对口型 Service — #1796 MediaKit 对口型业务逻辑, #1809 参数调整, #1845 配音前置.
 
 职责：
-- 创建/查询/取消对口型任务
-- 调用 TTS 合成音频（#1809：前端不再传 audio_url）
+- 创建/查询对口型任务
+- 三输入模式：
+  1. TTS 直生（voice_id + script_text）→ 走 Celery 异步（降级路径）
+  2. 直接音频（audio_url，前端未传 timings）→ 同步下载 + 算 timings + 提交 MediaKit
+  3. 预合成音频（audio_url + sentence_timings，#1845 新主路径）→ 同步 ffprobe 校验时长 +
+     写入前端传来的 timings → 直接提交 MediaKit（~2-3s）
 - 调用 MediaKit 客户端提交异步任务
-- 轮询更新任务状态
+- 轮询更新任务状态（中间状态同步 DB，成片转存自家 OSS）
 - 用户隔离（每个用户只能操作自己的任务）
 """
 
 from __future__ import annotations
 
+import io
 import logging
 import uuid
 from datetime import datetime, timezone
 from typing import Optional
+from urllib.parse import urlparse
 
 from app.services.mediakit_client import (
     STATUS_COMPLETED,
@@ -23,13 +29,25 @@ from app.services.mediakit_client import (
     MediaKitError,
     get_mediakit_client,
 )
+
+# Celery 异步任务：TTS 合成 + MediaKit 提交（降级路径）
+from app.tasks.lipsync_tts import tts_synthesize_and_submit
 from sqlalchemy.orm import Session
 
 from packages.adapters.sqlalchemy_impl.models import LipsyncJobModel
-from packages.application.cosyvoice_service import CosyVoiceError, CosyVoiceService
+from packages.application.cosyvoice_service import CosyVoiceError, normalize_emotion
+from packages.domain.sentence_timings import (
+    compute_sentence_timings,
+    probe_audio_duration,
+)
+from packages.shared.storage import get_shared_storage_service
+from packages.shared.url_security import ALLOWED_AUDIO_MIME_TYPES, safe_download_bytes
 
 logger = logging.getLogger(__name__)
 
+# 传给 MediaKit GPU worker / 回给前端播放的 OSS 预签名有效期：7 天。
+MEDIAKIT_URL_TTL_SECONDS = 7 * 24 * 3600
+
 
 class LipsyncService:
     """对口型任务 Service."""
@@ -38,19 +56,192 @@ class LipsyncService:
         self,
         db: Session,
         client: Optional[MediaKitClient] = None,
-        cosyvoice_service: Optional[CosyVoiceService] = None,
+        cosyvoice_service=None,
+        voice_clone_repo=None,
     ):
         self.db = db
         self.client = client or get_mediakit_client()
-        self._cosyvoice_service = cosyvoice_service
+        self._cosyvoice = cosyvoice_service
+        self._voice_clone_repo = voice_clone_repo
 
-    @property
-    def cosyvoice_service(self) -> CosyVoiceService:
-        if self._cosyvoice_service is None:
+    def _get_cosyvoice(self):
+        """延迟获取 CosyVoiceService（与 tts 路由一致，含 OSS 预签名配置）."""
+        if self._cosyvoice is None:
             from app.dependencies import get_cosyvoice_service
 
-            self._cosyvoice_service = get_cosyvoice_service()
-        return self._cosyvoice_service
+            self._cosyvoice = get_cosyvoice_service()
+        return self._cosyvoice
+
+    def _resolve_voice_id(self, voice_id: str, user_id: str) -> str:
+        """将克隆音色 profile UUID 解析为 CosyVoice voice_id。
+
+        与 /tts/synthesize 保持一致：命中 profile → 校验归属 → 返回其 voice_id；
+        未命中（预置音色 ID 或克隆 CosyVoice voice_id）原样返回。
+        """
+        if not voice_id:
+            return ""
+        if self._voice_clone_repo is None:
+            try:
+                from app.dependencies import get_voice_clone_profile_repository
+
+                self._voice_clone_repo = get_voice_clone_profile_repository(self.db)
+            except Exception:
+                return voice_id
+        try:
+            profile = self._voice_clone_repo.get(voice_id)
+        except Exception:
+            return voice_id
+        if profile is None:
+            return voice_id
+        if getattr(profile, "user_id", "") != user_id:
+            raise MediaKitError("无权访问该音色", code="VoiceForbidden")
+        if not getattr(profile, "voice_id", ""):
+            raise MediaKitError("音色克隆尚未完成，请稍后再试", code="VoiceNotReady")
+        return profile.voice_id
+
+    def _synthesize_and_persist_audio(
+        self,
+        *,
+        user_id: str,
+        job_id: str,
+        voice_id: str,
+        script_text: str,
+        speed: float,
+        emotion: str,
+    ) -> str:
+        """TTS 直生：调 CosyVoice 合成音频并转存 OSS，返回可公网访问的音频 URL.
+
+        Raises:
+            MediaKitError: 合成失败
+        """
+        actual_voice_id = self._resolve_voice_id(voice_id, user_id)
+        cosyvoice = self._get_cosyvoice()
+        try:
+            result = cosyvoice.submit_synthesize_task(
+                text=script_text,
+                voice_id=actual_voice_id,
+                speed=speed,
+                emotion=normalize_emotion(emotion),
+            )
+        except CosyVoiceError as exc:
+            raise MediaKitError(f"TTS 合成失败: {exc}", code="TTSSynthesisFailed") from exc
+        except ValueError as exc:
+            raise MediaKitError(f"TTS 参数错误: {exc}", code="TTSInvalidParam") from exc
+
+        temp_url = result.get("audio_url", "")
+        if not temp_url:
+            raise MediaKitError("TTS 未返回音频 URL", code="TTSNoAudio")
+
+        # 转存到自家 OSS，避免临时 URL 过期导致 MediaKit 拉取失败
+        try:
+            audio_data = safe_download_bytes(
+                temp_url,
+                purpose="lipsync_tts_audio",
+                allowed_mime_types=ALLOWED_AUDIO_MIME_TYPES,
+                timeout=60.0,
+            )
+            storage = get_shared_storage_service()
+            storage_key = f"lipsync-tts/{user_id}/{job_id}.mp3"
+            permanent_url = storage.upload_file(io.BytesIO(audio_data), storage_key, content_type="audio/mpeg")
+            logger.info("对口型 TTS 音频已转存 OSS: job_id=%s key=%s", job_id, storage_key)
+            return permanent_url
+        except Exception as exc:
+            logger.warning("TTS 音频转存 OSS 失败，回退临时 URL: job_id=%s err=%s", job_id, exc)
+            return temp_url
+
+    def _submit_audio_direct(
+        self,
+        *,
+        job: LipsyncJobModel,
+        supplied_timings: Optional[list] = None,
+        supplied_duration: Optional[float] = None,
+    ) -> None:
+        """音频直传模式（包含 #1845 预合成路径）：同步下载 → ffprobe → timings → 提交 MediaKit.
+
+        直接在 HTTP 请求内完成，不走 Celery。job.status 成功后置为 submitted。
+        失败时把 job 标成 failed 并 commit，然后抛 MediaKitError。
+
+        Args:
+            job: 已 commit 的 LipsyncJobModel（audio_url / video_url 已写入）
+            supplied_timings: 前端传来的预合成 timings（可选，可信时直接用）
+            supplied_duration: 前端传来的预合成时长（可选，用于优先避免重复探测）
+        """
+        # 1. 下载音频
+        audio_data: bytes | None = None
+        try:
+            audio_data = safe_download_bytes(
+                job.audio_url,
+                purpose="lipsync_direct_audio",
+                allowed_mime_types=ALLOWED_AUDIO_MIME_TYPES,
+                timeout=60.0,
+            )
+            logger.info(
+                "[lipsync] 直传音频下载完成: job_id=%s size=%d",
+                job.id,
+                len(audio_data) if audio_data else 0,
+            )
+        except Exception as exc:
+            logger.warning("[lipsync] 直传音频下载失败，跳过 timings 计算: job_id=%s err=%s", job.id, exc)
+
+        # 2. ffprobe 探测时长（优先用前端传入的预合成时长，但以 ffprobe 为准做兜底校验）
+        audio_duration = 0.0
+        if audio_data:
+            audio_duration = probe_audio_duration(audio_data)
+        if audio_duration <= 0 and supplied_duration and supplied_duration > 0:
+            audio_duration = supplied_duration
+            logger.info(
+                "[lipsync] ffprobe 失败，使用前端传入的预合成时长: job_id=%s duration=%.2f", job.id, audio_duration
+            )
+
+        # 3. 句子时间戳：优先用前端预合成传入的 timings（后端预合成接口已经算过，可信）；
+        #    否则若音频下载成功则重算；否则不设置（不阻塞主流程）
+        timings: Optional[list] = None
+        if supplied_timings:
+            timings = supplied_timings
+            logger.info("[lipsync] 使用前端预合成句子时间戳: job_id=%s sentences=%d", job.id, len(timings))
+        elif audio_data and audio_duration > 0 and job.script_text:
+            try:
+                timings = compute_sentence_timings(audio_data, job.script_text, audio_duration)
+                logger.info(
+                    "[lipsync] 后端重算句子时间戳: job_id=%s sentences=%d duration=%.2f",
+                    job.id,
+                    len(timings) if timings else 0,
+                    audio_duration,
+                )
+            except Exception as exc:
+                logger.warning("[lipsync] 句子时间戳计算失败（不阻塞）: job_id=%s err=%s", job.id, exc)
+
+        if timings:
+            job.sentence_timings = timings
+
+        # 4. 签名 URL 并提交 MediaKit
+        video_url = self._sign_media_url(job.video_url)
+        signed_audio_url = self._sign_media_url(job.audio_url)
+        job.audio_url = signed_audio_url
+
+        try:
+            result = self.client.submit_lipsync(
+                video_url=video_url,
+                audio_url=signed_audio_url,
+                enable_video_loop=job.enable_video_loop,
+                client_token=job.id,
+            )
+            job.mediakit_task_id = result["task_id"]
+            job.status = "submitted"
+            job.submitted_at = datetime.now(timezone.utc)
+            self.db.commit()
+            logger.info(
+                "[lipsync] 直传音频已提交 MediaKit: job_id=%s task_id=%s",
+                job.id,
+                result["task_id"],
+            )
+        except MediaKitError as exc:
+            job.status = "failed"
+            job.error_message = str(exc)
+            job.error_code = exc.code
+            logger.error("[lipsync] 直传音频提交 MediaKit 失败: job_id=%s err=%s", job.id, exc)
+            self.db.commit()
+            raise
 
     # ── 创建任务 ──────────────────────────────────────────────────────────
 
@@ -59,47 +250,49 @@ class LipsyncService:
         *,
         user_id: str,
         video_url: str,
-        voice_id: str,
-        script_text: str,
-        enable_video_loop: bool = False,
+        audio_url: str = "",
+        audio_duration: Optional[float] = None,
+        sentence_timings: Optional[list] = None,
+        voice_id: str = "",
+        script_text: str = "",
+        speed: float = 1.0,
+        emotion: str = "",
+        enable_video_loop: bool = True,
         project_id: str = "",
     ) -> LipsyncJobModel:
-        """创建对口型任务并提交到 MediaKit.
+        """创建对口型任务.
 
-        #1809: 内部调 TTS 合成音频，不再由前端传 audio_url。
+        三种输入模式：
+        - TTS 直生：voice_id + script_text（audio_url 留空）
+          → 创建 DB 记录（状态 tts_processing），dispatch Celery 异步任务（降级路径）。
+          API 响应 <1s。
+        - 直接音频：audio_url 非空 + 无 sentence_timings
+          → 同步下载音频 + 重算 timings + 提交 MediaKit（几秒完成）。
+        - 预合成音频（#1845 新主路径）：audio_url 非空 + 传 sentence_timings
+          → 同步 ffprobe 校验时长 + 写入 timings + 提交 MediaKit（~2-3s）。
 
         Raises:
-            CosyVoiceError: TTS 合成失败
-            MediaKitError: API 调用失败
+            MediaKitError: 参数校验失败或 MediaKit 提交失败
         """
-        # 1. 调 TTS 合成音频
-        try:
-            tts_result = self.cosyvoice_service.synthesize_speech(
-                text=script_text,
-                voice_id=voice_id,
-            )
-            audio_url = tts_result.audio_url
-        except CosyVoiceError as exc:
-            logger.error("TTS 合成失败: voice_id=%s, error=%s", voice_id, exc)
-            # 创建失败记录
-            job_id = str(uuid.uuid4())
-            job = LipsyncJobModel(
-                id=job_id,
-                user_id=user_id,
-                project_id=project_id,
-                video_url=video_url,
-                audio_url="",
-                enable_video_loop=enable_video_loop,
-                status="failed",
-                error_message=f"TTS 合成失败: {exc}",
-                error_code="TTSSynthesisFailed",
-            )
-            self.db.add(job)
-            self.db.commit()
-            self.db.refresh(job)
-            raise
+        # 0. 输入校验
+        is_pre_synth = bool(audio_url) and bool(sentence_timings)
+        bool(audio_url) and not is_pre_synth
+        is_tts_mode = not bool(audio_url)
 
-        # 2. 创建数据库记录
+        if is_tts_mode:
+            if not (voice_id and script_text):
+                raise MediaKitError(
+                    "必须提供 audio_url 或 voice_id+script_text",
+                    code="InvalidInput",
+                )
+            # TTS 模式：在 HTTP 请求中同步校验音色归属，快速失败
+            self._resolve_voice_id(voice_id, user_id)
+        elif is_pre_synth:
+            # 预合成模式：script_text 可空（因为 timings 已自带句子文本），但仍建议传
+            if not isinstance(sentence_timings, list) or len(sentence_timings) == 0:
+                raise MediaKitError("预合成模式 sentence_timings 不能为空", code="InvalidInput")
+
+        # 1. 创建数据库记录
         job_id = str(uuid.uuid4())
         job = LipsyncJobModel(
             id=job_id,
@@ -108,33 +301,139 @@ class LipsyncService:
             video_url=video_url,
             audio_url=audio_url,
             enable_video_loop=enable_video_loop,
-            status="pending",
+            voice_id=voice_id or "",
+            script_text=script_text or "",
+            speed=speed,
+            emotion=normalize_emotion(emotion) if is_tts_mode else (emotion or ""),
+            # 音频直传（含预合成）直接进入 pending（后续同步改为 submitted）；TTS 模式进入 tts_processing
+            status="tts_processing" if is_tts_mode else "pending",
         )
         self.db.add(job)
         self.db.flush()
 
-        # 3. 提交到 MediaKit
-        try:
-            result = self.client.submit_lipsync(
-                video_url=video_url,
-                audio_url=audio_url,
-                enable_video_loop=enable_video_loop,
-                client_token=job_id,  # 幂等控制
-            )
-            job.mediakit_task_id = result["task_id"]
-            job.status = "submitted"
-            job.submitted_at = datetime.now(timezone.utc)
-        except MediaKitError as exc:
-            job.status = "failed"
-            job.error_message = str(exc)
-            job.error_code = exc.code
-            logger.error("提交对口型任务失败: %s", exc)
-            raise
-
+        # ⚠️ 必须先 commit 再发 Celery 任务 / 后续同步操作，避免事务竞态
         self.db.commit()
         self.db.refresh(job)
+
+        if is_tts_mode:
+            # 2a. TTS 模式：dispatch Celery 异步任务处理 TTS 合成 + MediaKit 提交（降级路径）
+            try:
+                tts_synthesize_and_submit.apply_async(
+                    args=(
+                        job_id,
+                        user_id,
+                        voice_id,
+                        script_text,
+                        speed,
+                        normalize_emotion(emotion),
+                    )
+                )
+            except Exception as exc:
+                logger.exception(
+                    "Celery 任务提交失败，TTS 任务已创建但未触发执行: job_id=%s err=%s",
+                    job_id,
+                    exc,
+                )
+                job.status = "failed"
+                job.error_message = f"Celery 任务投递失败: {exc}"
+                job.error_code = "AsyncDispatchFailed"
+                job.updated_at = datetime.now(timezone.utc)
+                self.db.commit()
+        else:
+            # 2b/2c. 直接音频 / 预合成音频：同步路径
+            self._submit_audio_direct(
+                job=job,
+                supplied_timings=sentence_timings,
+                supplied_duration=audio_duration,
+            )
+            self.db.refresh(job)
+
         return job
 
+    # ── TTS 预合成（#1845 步骤1「生成配音」同步接口使用） ──────────────────
+
+    def preview_tts(
+        self,
+        *,
+        user_id: str,
+        voice_id: str,
+        script_text: str,
+        speed: float = 1.0,
+        emotion: str = "natural",
+    ) -> dict:
+        """同步做 TTS 合成 + 下载 + ffprobe + 句子时间戳计算.
+
+        不创建 LipsyncJob、不转存 OSS，直接返回 CosyVoice 临时 URL（~24h 有效期）。
+        耗时约 2-3 秒，由前端在步骤1点「生成配音」时同步等待。
+
+        Returns:
+            {"audio_url": str, "duration": float, "sentence_timings": list[dict]}
+
+        Raises:
+            MediaKitError: TTS 合成失败 / 下载失败 / ffprobe 失败
+        """
+        # 1. 音色解析（校验克隆音色归属）
+        actual_voice_id = self._resolve_voice_id(voice_id, user_id)
+        cosyvoice = self._get_cosyvoice()
+
+        # 2. TTS 合成（同步，~2-3s）
+        try:
+            result = cosyvoice.submit_synthesize_task(
+                text=script_text,
+                voice_id=actual_voice_id,
+                speed=speed,
+                emotion=normalize_emotion(emotion),
+            )
+        except CosyVoiceError as exc:
+            raise MediaKitError(f"TTS 合成失败: {exc}", code="TTSSynthesisFailed") from exc
+        except ValueError as exc:
+            raise MediaKitError(f"TTS 参数错误: {exc}", code="TTSInvalidParam") from exc
+
+        temp_url = result.get("audio_url", "")
+        if not temp_url:
+            raise MediaKitError("TTS 未返回音频 URL", code="TTSNoAudio")
+
+        # 3. 下载音频到内存（用于 ffprobe + 静音检测）
+        try:
+            audio_data = safe_download_bytes(
+                temp_url,
+                purpose="tts_preview_audio",
+                allowed_mime_types=ALLOWED_AUDIO_MIME_TYPES,
+                timeout=60.0,
+            )
+        except Exception as exc:
+            logger.warning("[tts-preview] TTS 音频下载失败，仍返回 audio_url: user_id=%s err=%s", user_id, exc)
+            return {
+                "audio_url": temp_url,
+                "duration": 0.0,
+                "sentence_timings": [],
+            }
+
+        # 4. ffprobe 时长
+        duration = probe_audio_duration(audio_data)
+        if duration <= 0:
+            logger.warning("[tts-preview] ffprobe 未返回有效时长，timings 留空: user_id=%s", user_id)
+            return {
+                "audio_url": temp_url,
+                "duration": 0.0,
+                "sentence_timings": [],
+            }
+
+        # 5. 句子时间戳
+        timings = compute_sentence_timings(audio_data, script_text, duration)
+
+        logger.info(
+            "[tts-preview] TTS 预合成完成: user_id=%s duration=%.2f sentences=%d",
+            user_id,
+            duration,
+            len(timings),
+        )
+        return {
+            "audio_url": temp_url,
+            "duration": round(duration, 2),
+            "sentence_timings": timings,
+        }
+
     # ── 查询任务 ──────────────────────────────────────────────────────────
 
     def get_job(self, job_id: str, user_id: str) -> Optional[LipsyncJobModel]:
@@ -168,11 +467,7 @@ class LipsyncService:
     # ── 更新任务状态（轮询） ──────────────────────────────────────────────
 
     def refresh_job_status(self, job_id: str, user_id: str) -> Optional[LipsyncJobModel]:
-        """从 MediaKit 拉取最新状态并更新本地记录.
-
-        Returns:
-            更新后的 Job，或 None（任务不存在/不属于该用户）
-        """
+        """从 MediaKit 拉取最新状态并更新本地记录."""
         job = self.get_job(job_id, user_id)
         if job is None:
             return None
@@ -192,34 +487,92 @@ class LipsyncService:
             return job
 
         mk_status = status_data.get("status", STATUS_RUNNING)
+        logger.info("MediaKit 对口型状态 [%s]: %s", job_id, mk_status)
 
         if mk_status == STATUS_COMPLETED:
             result = status_data.get("result", {})
             job.status = STATUS_COMPLETED
-            job.output_video_url = result.get("video_url", "")
+            temp_url = result.get("video_url", "")
+            job.output_video_url = temp_url
             job.output_duration = result.get("duration", 0.0)
             job.completed_at = datetime.now(timezone.utc)
+            job.updated_at = datetime.now(timezone.utc)
+            self.db.commit()
+            # 异步转存自家 OSS
+            try:
+                from app.tasks.lipsync_tts import persist_output_video_task
+
+                persist_output_video_task.apply_async(args=(job_id, user_id, temp_url))
+            except Exception as exc:
+                logger.warning(
+                    "提交输出视频异步转存任务失败，保留临时 URL: job_id=%s err=%s",
+                    job_id,
+                    exc,
+                )
+            self.db.refresh(job)
+            return job
         elif mk_status == STATUS_FAILED:
             error = status_data.get("error", {})
             job.status = "failed"
             job.error_message = error.get("message", "任务执行失败")
             job.error_code = error.get("code", "TaskFailed")
             job.completed_at = datetime.now(timezone.utc)
-        # running 状态只更新时间戳
+        else:
+            if isinstance(mk_status, str) and mk_status:
+                job.status = mk_status
         job.updated_at = datetime.now(timezone.utc)
         self.db.commit()
         self.db.refresh(job)
         return job
 
+    def _persist_output_video(self, temp_url: str, job_id: str, user_id: str) -> str:
+        """将 MediaKit 输出的临时视频 URL 转存到自家 OSS. 失败时回退返回原始临时 URL."""
+        if not temp_url:
+            return ""
+        try:
+            import httpx
+
+            with httpx.Client(timeout=180.0, follow_redirects=True) as client:
+                resp = client.get(temp_url)
+                resp.raise_for_status()
+                data = resp.content
+            storage = get_shared_storage_service()
+            storage_key = f"lipsync-outputs/{user_id}/{job_id}.mp4"
+            permanent_url = storage.upload_file(io.BytesIO(data), storage_key, content_type="video/mp4")
+            logger.info("对口型输出视频已转存 OSS: job_id=%s key=%s", job_id, storage_key)
+            return self._sign_media_url(permanent_url) or temp_url
+        except Exception as exc:
+            logger.warning("对口型输出视频转存 OSS 失败，回退临时 URL: job_id=%s err=%s", job_id, exc)
+            return temp_url
+
+    def _sign_media_url(self, url: str) -> str:
+        """对自家 OSS 私有桶 URL 重签长有效期预签名."""
+        if not url:
+            return url
+        try:
+            storage = get_shared_storage_service()
+            public_base = getattr(storage, "public_url", "")
+            if not isinstance(public_base, str) or not public_base:
+                return url
+            own_host = urlparse(public_base).netloc.lower()
+            host = urlparse(url).netloc.lower()
+            if not own_host or host != own_host:
+                return url  # 外部临时链接原样透传
+            signed = storage.get_download_url(url, expires_seconds=MEDIAKIT_URL_TTL_SECONDS)
+            return signed or url
+        except Exception as exc:
+            logger.warning("对口型 URL 重签失败，原样返回: url_prefix=%s err=%s", url[:80], exc)
+            return url
+
     # ── 取消任务 ──────────────────────────────────────────────────────────
 
     def cancel_job(self, job_id: str, user_id: str) -> Optional[LipsyncJobModel]:
-        """取消任务（仅 pending/submitted 状态可取消）."""
+        """取消任务（仅 pending/tts_processing/submitted 状态可取消）."""
         job = self.get_job(job_id, user_id)
         if job is None:
             return None
 
-        if job.status in ("pending", "submitted"):
+        if job.status in ("pending", "tts_processing", "submitted"):
             job.status = "cancelled"
             job.updated_at = datetime.now(timezone.utc)
             self.db.commit()

apps/api/app/services/mediakit_client.py

@@ -75,7 +75,7 @@ class MediaKitClient:
         *,
         video_url: str,
         audio_url: str,
-        enable_video_loop: bool = False,
+        enable_video_loop: bool = True,
         callback_url: Optional[str] = None,
         callback_args: Optional[str] = None,
         client_token: Optional[str] = None,
@@ -103,8 +103,7 @@ class MediaKitClient:
             "video_url": video_url,
             "audio_url": audio_url,
         }
-        if enable_video_loop:
-            payload["enable_video_loop"] = True
+        payload["enable_video_loop"] = bool(enable_video_loop)
         if callback_url:
             payload["callback_url"] = callback_url
         if callback_args:

apps/api/app/tasks/lipsync_tts.py

@@ -0,0 +1,355 @@
+"""AI 数字人对口型 TTS 异步任务 — 将 TTS 合成从 HTTP 请求移至 Celery 后台执行.
+
+优化目标：将 create_job 的 API 响应时间从 6~35s 降到 <1s。
+任务流程：
+1. 创建新 DB session，加载 job 记录
+2. 调用 CosyVoice 合成音频
+3. 下载音频并转存到自家 OSS
+4. 更新 job 的 audio_url
+5. 签名 URL 并提交到 MediaKit
+6. 更新 job 状态为 submitted
+7. 异常时标记 job 为 failed
+
+注意：使用 @shared_task 而非绑定到某个 celery_app 实例，
+确保任务能被 Worker 侧 celery_app 正确注册，同时 API 侧 send_task/apply_async 仍可正常调用。
+
+#1845：句子时间戳计算已提取至 packages/domain/sentence_timings.py，本模块保留
+_ 开头别名兼容历史导入，但 _compute_sentence_timings/_split_script_into_sentences/
+_estimate_sentence_timings_by_chars 等内部函数已复用共享实现，避免重复代码。
+"""
+
+import io
+import logging
+from datetime import datetime, timezone
+from urllib.parse import urlparse
+
+from celery import shared_task
+
+# 复用共享的句子时间戳工具（#1845 配音前置）
+from packages.domain.sentence_timings import compute_sentence_timings as _compute_sentence_timings
+from packages.domain.sentence_timings import (
+    probe_audio_duration,
+)
+
+logger = logging.getLogger(__name__)
+
+# MediaKit 预签名 URL 有效期（7天，秒），与 LipsyncService._sign_media_url 保持一致
+_MEDIAKIT_URL_TTL_SECONDS = 7 * 24 * 3600
+
+
+def _sign_media_url(url: str) -> str:
+    """对自家 OSS 私有桶 URL 重签长有效期预签名.
+
+    - 自家 OSS URL → 重签 7 天有效期
+    - 外部临时 URL → 原样透传
+    - 任何异常降级原样返回，不阻断主流程
+    """
+    if not url:
+        return url
+    try:
+        from packages.shared.storage import get_shared_storage_service
+
+        storage = get_shared_storage_service()
+        public_base = getattr(storage, "public_url", "")
+        if not isinstance(public_base, str) or not public_base:
+            return url
+        own_host = urlparse(public_base).netloc.lower()
+        host = urlparse(url).netloc.lower()
+        if not own_host or host != own_host:
+            return url
+        signed = storage.get_download_url(url, expires_seconds=_MEDIAKIT_URL_TTL_SECONDS)
+        return signed or url
+    except Exception as exc:  # noqa: BLE001
+        logger.warning("[lipsync_tts] URL 重签失败，原样返回: url_prefix=%s err=%s", url[:80], exc)
+        return url
+
+
+@shared_task(
+    bind=True,
+    name="lipsync_tts.synthesize_and_submit",
+    max_retries=5,  # 事务竞态重试3次（job not found）+ TTS偶发错误2次
+    default_retry_delay=30,
+    autoretry_for=(OSError, ConnectionError),  # 网络/连接错误自动重试
+    retry_backoff=True,
+    retry_backoff_max=30,
+    soft_time_limit=180,
+    time_limit=200,
+)
+def tts_synthesize_and_submit(
+    self,
+    job_id: str,
+    user_id: str,
+    voice_id: str,
+    script_text: str,
+    speed: float,
+    emotion: str,
+):
+    """异步执行 TTS 合成 + OSS 转存 + MediaKit 提交.
+
+    在 Celery worker 中运行，不阻塞 HTTP 请求。保留作为降级路径
+    （预合成失败 / 旧版前端未传 audio_url 时走此路径）。
+    """
+    from app.services.mediakit_client import MediaKitError, get_mediakit_client
+    from sqlalchemy.orm import Session as DBSession
+
+    from packages.adapters.sqlalchemy_impl.models import LipsyncJobModel
+    from packages.application.cosyvoice_service import CosyVoiceError, CosyVoiceService
+    from packages.shared.url_security import safe_download_bytes
+
+    # SessionLocal 获取：
+    # - API 容器：app.db.SessionLocal（环境变量完整，导入即建引擎）
+    # - Worker 容器：worker_app.db.SessionLocal（Worker 自己的 settings 初始化引擎）
+    # API 侧没有 worker_app 模块 → ImportError 直接回退；
+    # Worker 侧 app.db 会因缺少 API 专有环境变量抛 pydantic ValidationError，
+    # 此时也要回退到 worker_app.db。
+    try:
+        from worker_app.db import SessionLocal  # type: ignore
+    except Exception:  # noqa: BLE001
+        from app.db import SessionLocal  # type: ignore
+
+    db: DBSession = SessionLocal()
+    try:
+        job = (
+            db.query(LipsyncJobModel)
+            .filter(
+                LipsyncJobModel.id == job_id,
+                LipsyncJobModel.user_id == user_id,
+            )
+            .first()
+        )
+
+        if job is None:
+            # 事务竞态防御：API 在 commit 前投递了任务，worker 消费时事务尚未提交。
+            retries = getattr(self.request, "retries", 0)
+            max_retries = 3
+            if retries < max_retries:
+                backoff = (2**retries) + (retries * 1)  # 1s, 3s, 7s
+                logger.warning(
+                    "[lipsync_tts] Job not found yet (retry %d/%d, backoff %ds): job_id=%s",
+                    retries + 1,
+                    max_retries,
+                    backoff,
+                    job_id,
+                )
+                self.db.close()
+                raise self.retry(countdown=backoff, max_retries=max_retries)
+            logger.error(
+                "[lipsync_tts] Job not found after %d retries, giving up: job_id=%s",
+                max_retries,
+                job_id,
+            )
+            return
+
+        # 已取消的任务不再处理
+        if job.status == "cancelled":
+            logger.info("[lipsync_tts] Job already cancelled, skipping: job_id=%s", job_id)
+            return
+
+        # 1. TTS 合成
+        logger.info(
+            "[lipsync_tts] 开始 TTS 合成: job_id=%s voice_id=%s text_len=%d speed=%.2f",
+            job_id,
+            voice_id,
+            len(script_text),
+            speed,
+        )
+        try:
+            cosyvoice = CosyVoiceService()
+            result = cosyvoice.submit_synthesize_task(
+                text=script_text,
+                voice_id=voice_id,
+                speed=speed,
+                emotion=emotion,
+            )
+        except CosyVoiceError as exc:
+            logger.error("[lipsync_tts] TTS 合成失败: job_id=%s err=%s", job_id, exc)
+            job.status = "failed"
+            job.error_message = f"TTS 合成失败: {exc}"
+            job.error_code = "TTSSynthesisFailed"
+            job.updated_at = datetime.now(timezone.utc)
+            db.commit()
+            return
+        except ValueError as exc:
+            logger.error("[lipsync_tts] TTS 参数错误: job_id=%s err=%s", job_id, exc)
+            job.status = "failed"
+            job.error_message = f"TTS 参数错误: {exc}"
+            job.error_code = "TTSInvalidParam"
+            job.updated_at = datetime.now(timezone.utc)
+            db.commit()
+            return
+
+        temp_url = result.get("audio_url", "")
+        if not temp_url:
+            logger.error("[lipsync_tts] TTS 未返回音频 URL: job_id=%s", job_id)
+            job.status = "failed"
+            job.error_message = "TTS 未返回音频 URL"
+            job.error_code = "TTSNoAudio"
+            job.updated_at = datetime.now(timezone.utc)
+            db.commit()
+            return
+
+        # 2. 下载 TTS 音频到内存（用于 2.5 静音检测；不转存自家 OSS，直接使用 CosyVoice 临时 URL）
+        audio_data: bytes | None = None
+        try:
+            audio_data = safe_download_bytes(
+                temp_url,
+                purpose="lipsync_tts_audio",
+                allowed_mime_types={
+                    "audio/mpeg",
+                    "audio/mp3",
+                    "audio/wav",
+                    "audio/x-wav",  # CosyVoice 部分接口返回 audio/x-wav
+                    "audio/mp4",
+                    "audio/x-m4a",
+                },
+                timeout=60.0,
+            )
+            logger.info(
+                "[lipsync_tts] TTS 音频已下载到内存: job_id=%s size=%d",
+                job_id,
+                len(audio_data) if audio_data else 0,
+            )
+        except Exception as exc:
+            logger.warning(
+                "[lipsync_tts] TTS 音频下载失败，跳过静音检测，直接使用临时 URL 提交: job_id=%s err=%s",
+                job_id,
+                exc,
+            )
+        # TTS 音频使用 CosyVoice 临时 URL，跳过自家 OSS 转存（加速，步骤⑥）
+        job.audio_url = temp_url
+        logger.info("[lipsync_tts] TTS 音频使用 CosyVoice 临时 URL（跳过 OSS 转存）: job_id=%s", job_id)
+
+        db.commit()
+
+        # 2.5 计算精确句子时间戳（基于 TTS 音频静音检测）—— 复用共享工具
+        try:
+            if not audio_data:
+                logger.warning("[lipsync_tts] 无音频数据，跳过句子时间戳计算: job_id=%s", job_id)
+            else:
+                _audio_duration = probe_audio_duration(audio_data)
+                logger.info(
+                    "[lipsync_tts] 音频时长探测: job_id=%s duration=%.2f",
+                    job_id,
+                    _audio_duration,
+                )
+
+                if _audio_duration > 0:
+                    _timings = _compute_sentence_timings(audio_data, script_text, _audio_duration)
+                    if _timings:
+                        job.sentence_timings = _timings
+                        logger.info(
+                            "[lipsync_tts] 句子时间戳已计算: job_id=%s sentences=%d duration=%.1f",
+                            job_id,
+                            len(_timings),
+                            _audio_duration,
+                        )
+                    else:
+                        logger.warning("[lipsync_tts] 句子时间戳计算返回空结果: job_id=%s", job_id)
+                else:
+                    logger.warning(
+                        "[lipsync_tts] ffprobe 未获取到有效时长，跳过句子时间戳: job_id=%s",
+                        job_id,
+                    )
+                db.commit()
+        except Exception as _st_err:
+            logger.warning(
+                "[lipsync_tts] 句子时间戳计算失败（不影响主流程）: job_id=%s err=%s", job_id, _st_err, exc_info=True
+            )
+
+        # 3. 签名 URL 并提交到 MediaKit（复用模块内 _sign_media_url，避免对 LipsyncService 的耦合）
+        audio_url = _sign_media_url(job.audio_url)
+        video_url = _sign_media_url(job.video_url)
+
+        client = get_mediakit_client()
+        try:
+            mk_result = client.submit_lipsync(
+                video_url=video_url,
+                audio_url=audio_url,
+                enable_video_loop=job.enable_video_loop,
+                client_token=job_id,
+            )
+            job.mediakit_task_id = mk_result["task_id"]
+            job.status = "submitted"
+            job.submitted_at = datetime.now(timezone.utc)
+            logger.info(
+                "[lipsync_tts] 已提交 MediaKit: job_id=%s task_id=%s",
+                job_id,
+                mk_result["task_id"],
+            )
+        except MediaKitError as exc:
+            job.status = "failed"
+            job.error_message = str(exc)
+            job.error_code = exc.code
+            logger.error("[lipsync_tts] 提交 MediaKit 失败: job_id=%s err=%s", job_id, exc)
+
+        db.commit()
+
+    except Exception:
+        logger.exception("[lipsync_tts] 未预期的异常: job_id=%s", job_id)
+        try:
+            job = db.query(LipsyncJobModel).filter(LipsyncJobModel.id == job_id).first()
+            if job and job.status not in ("cancelled", "failed", "completed"):
+                job.status = "failed"
+                job.error_message = "TTS 异步任务执行异常"
+                job.error_code = "AsyncTaskError"
+                job.updated_at = datetime.now(timezone.utc)
+                db.commit()
+        except Exception:
+            logger.exception("[lipsync_tts] 回写失败状态时异常: job_id=%s", job_id)
+    finally:
+        db.close()
+
+
+@shared_task(
+    name="lipsync_tts.persist_output_video",
+    max_retries=2,
+    default_retry_delay=30,
+)
+def persist_output_video_task(job_id: str, user_id: str, temp_url: str):
+    """异步转存对口型输出视频到自家 OSS（步骤⑦ — 将同步阻塞挪到后台，加速前端响应）."""
+
+    try:
+        from worker_app.db import SessionLocal  # type: ignore
+    except Exception:  # noqa: BLE001
+        from app.db import SessionLocal  # type: ignore
+
+    from packages.adapters.sqlalchemy_impl.models import LipsyncJobModel
+    from packages.shared.storage import get_shared_storage_service
+
+    db = SessionLocal()
+    try:
+        job = db.query(LipsyncJobModel).filter(LipsyncJobModel.id == job_id, LipsyncJobModel.user_id == user_id).first()
+        if job is None:
+            logger.error("[lipsync_tts.persist] Job not found: job_id=%s", job_id)
+            return
+
+        if not temp_url:
+            logger.warning("[lipsync_tts.persist] temp_url 为空，跳过转存: job_id=%s", job_id)
+            return
+
+        try:
+            import httpx
+
+            with httpx.Client(timeout=180.0, follow_redirects=True) as client:
+                resp = client.get(temp_url)
+                resp.raise_for_status()
+                data = resp.content
+
+            storage = get_shared_storage_service()
+            storage_key = f"lipsync-outputs/{user_id}/{job_id}.mp4"
+            permanent_url = storage.upload_file(io.BytesIO(data), storage_key, content_type="video/mp4")
+            final_url = _sign_media_url(permanent_url) if permanent_url else temp_url
+            job.output_video_url = final_url
+            job.updated_at = datetime.now(timezone.utc)
+            db.commit()
+            logger.info("[lipsync_tts.persist] 输出视频已转存 OSS: job_id=%s key=%s", job_id, storage_key)
+        except Exception as exc:
+            logger.warning(
+                "[lipsync_tts.persist] 输出视频转存失败，保留临时 URL: job_id=%s err=%s",
+                job_id,
+                exc,
+            )
+    except Exception:
+        logger.exception("[lipsync_tts.persist] 未预期异常: job_id=%s", job_id)
+    finally:
+        db.close()

apps/web/src/api/products/products.ts

@@ -1,6 +1,6 @@
 /**
  * 成品 / 视频相关 API 函数
- * 后端实际接口：/videos
+ * 后端实际接口：/videos（分页：page/page_size，返回 {items, total, page, page_size}）
  */
 import apiClient from "../client"
 import type {
@@ -12,16 +12,39 @@ import type {
 } from "./types"
 import { mapVideoToProductItem } from "./utils"
 
-/** 获取成品列表（支持分页和筛选） */
-export const getProducts = async (params?: ProductListParams): Promise<ProductItem[]> => {
-  const response = await apiClient.get("/videos", { params })
-  const data = response.data
-  const videos: VideoItem[] = Array.isArray(data?.items)
-    ? data.items
-    : Array.isArray(data)
-      ? data
-      : []
-  return videos.map(mapVideoToProductItem)
+/** 分页列表响应（前端消费用） */
+export interface ProductListResult {
+  items: ProductItem[]
+  total: number
+  page: number
+  page_size: number
+}
+
+/**
+ * 获取成品列表（分页）
+ * @param params 分页与筛选参数：page 默认 1，page_size 默认 20
+ */
+export const getProducts = async (params?: ProductListParams): Promise<ProductListResult> => {
+  const response = await apiClient.get("/videos", {
+    params: {
+      page: 1,
+      page_size: 20,
+      ...params,
+    },
+  })
+  const data = response.data as {
+    items?: VideoItem[]
+    total?: number
+    page?: number
+    page_size?: number
+  }
+  const items: VideoItem[] = Array.isArray(data?.items) ? data.items : []
+  return {
+    items: items.map(mapVideoToProductItem),
+    total: data.total ?? items.length,
+    page: data.page ?? params?.page ?? 1,
+    page_size: data.page_size ?? params?.page_size ?? 20,
+  }
 }
 
 /** 获取单个成品详情 */

apps/web/src/api/tts/types.ts

@@ -103,6 +103,7 @@ export interface TTSPreviewRequest {
   voice_id: string
   speed?: number
   pitch?: number
+  emotion?: string // 情绪参数：natural/excited/calm/friendly
 }
 
 /** TTS 试听响应 */

apps/web/src/pages/ai-avatar/AiAvatar.css

@@ -486,14 +486,16 @@
 
 .aa-lipsync-preview {
   width: 100%;
+  max-width: 240px;
   aspect-ratio: 9/16;
   background: #0f0f1a;
   border-radius: 10px;
   overflow: hidden;
-  margin-bottom: 12px;
+  margin: 0 auto 12px auto;
   display: flex;
   align-items: center;
   justify-content: center;
+  position: relative;
 }
 
 .aa-lipsync-preview video {
@@ -547,21 +549,25 @@
 /* ── 封面 & 生成 ── */
 .aa-cover-preview {
   width: 100%;
+  max-width: 240px;
   aspect-ratio: 9/16;
-  max-height: 160px;
   background: #f0f0f5;
-  border-radius: 10px;
+  border-radius: 12px;
   overflow: hidden;
   display: flex;
   align-items: center;
   justify-content: center;
-  margin-bottom: 12px;
+  margin: 0 auto 12px auto;
+  position: relative;
 }
 
 .aa-cover-preview img {
   width: 100%;
   height: 100%;
+  aspect-ratio: 9/16;
   object-fit: cover;
+  display: block;
+  border-radius: 12px;
 }
 
 .aa-cover-preview__placeholder {
@@ -569,6 +575,19 @@
   color: #8c8ca1;
 }
 
+.aa-cover-preview__loading {
+  position: absolute;
+  inset: 0;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  background: rgba(0, 0, 0, 0.45);
+  color: #fff;
+  font-size: 13px;
+  backdrop-filter: blur(4px);
+  -webkit-backdrop-filter: blur(4px);
+}
+
 .aa-cover-actions {
   display: flex;
   gap: 8px;
@@ -1150,3 +1169,99 @@
   flex-direction: column;
   gap: 4px;
 }
+
+/* ─ 对口型生成弹窗 Spinner ── */
+.aa-lipsync-spinner {
+  width: 48px;
+  height: 48px;
+  border: 4px solid #f0f0f5;
+  border-top-color: #6366f1;
+  border-radius: 50%;
+  animation: aa-spin 0.8s linear infinite;
+}
+
+@keyframes aa-spin {
+  to {
+    transform: rotate(360deg);
+  }
+}
+
+.aa-btn--danger {
+  background: #ff4d4f;
+  color: #fff;
+  border: none;
+}
+
+.aa-btn--danger:hover {
+  background: #ff7875;
+}
+
+/* ── v3.1 两步骤导航（仅追加，不改动上方任何原有样式） ── */
+.aa-step-nav {
+  display: flex;
+  align-items: center;
+  gap: 24px;
+  padding: 10px 24px;
+  background: #fff;
+  border-bottom: 1px solid #e8e8ec;
+  flex-shrink: 0;
+}
+
+.aa-step-nav__item {
+  font-size: 13px;
+  font-weight: 500;
+  color: #8c8ca1;
+  position: relative;
+  padding-bottom: 6px;
+}
+
+.aa-step-nav__item.active {
+  color: #4f46e5;
+  font-weight: 600;
+}
+
+.aa-step-nav__item.active::after {
+  content: "";
+  position: absolute;
+  left: 0;
+  right: 0;
+  bottom: -11px;
+  height: 2px;
+  background: #4f46e5;
+  border-radius: 1px;
+}
+
+/* 步骤切换按钮行 */
+.aa-step-btn-row {
+  display: flex;
+  justify-content: flex-end;
+  gap: 8px;
+  margin-top: 16px;
+  padding-top: 12px;
+  border-top: 1px dashed #f0f0f5;
+}
+
+.aa-step-btn-row .aa-btn {
+  min-width: 120px;
+}
+
+/* 分步布局：每个步骤 3 个横向面板等宽撑满（沿用原 .aa-panel 外观，仅改宽度） */
+.aa-page-body > .aa-panel--s1,
+.aa-page-body > .aa-panel--s2 {
+  flex: 1 1 0;
+  width: auto;
+  min-width: 220px;
+}
+
+/* 文案 / 对口型预览面板内容较多，给更宽的弹性比例 */
+.aa-page-body > .aa-panel--s1-wide {
+  flex: 1.4 1 0;
+  width: auto;
+  min-width: 280px;
+}
+
+.aa-page-body > .aa-panel--s2-wide {
+  flex: 1.4 1 0;
+  width: auto;
+  min-width: 300px;
+}

apps/web/src/pages/ai-avatar/api/aiAvatar.ts

@@ -1,8 +1,8 @@
 /**
- * AI数字人 — API 封装
+ * AI数字人 — API 封装（#1822 契约对齐）
  */
 import apiClient from "@/api/client"
-import type { Script, LipsyncJob, RenderJob, BRollSegment } from "../types"
+import type { Script, LipsyncJob, RenderJob, BRollSegment, SentenceTiming } from "../types"
 
 /* ── 文案库 ── */
 export const getScripts = async (): Promise<Script[]> => {
@@ -28,24 +28,63 @@ export const deleteScript = async (id: string): Promise<void> => {
   await apiClient.delete(`/scripts/${id}`)
 }
 
-/* ── 素材单查（用于拿到 file_url 传给对口型等新接口） ── */
+/* ── 素材单查（拿到 file_url 作为对口型的 video_url） ── */
 export const getAssetById = async (id: string): Promise<{ file_url?: string; id: string }> => {
   const response = await apiClient.get<{ file_url?: string; id: string }>(`/assets/${id}`)
   return response.data
 }
 
-/* ── 对口型 ── */
+/* ── 对口型（支持三种模式） ──
+ * 1. TTS 直生（降级/旧版）：传 voice_id + script_text（+speed/emotion），后端 Celery 异步合成
+ * 2. 直接音频：传 video_url + audio_url，后端同步下载+算timings+提交MediaKit
+ * 3. 预合成音频（#1845 新主路径）：先调 previewTts 拿 audio_url+sentence_timings，
+ *    再把 audio_url + audio_duration + sentence_timings 一起传过来，后端直接提交 MediaKit
+ */
 export const createLipsyncJob = async (data: {
-  voice_id: string
-  script_text: string
+  /** 人物视频 URL（MP4）；由素材 id 经 getAssetById 拿 file_url */
   video_url: string
+  /** 预合成/直接音频模式：音频 URL（#1845 步骤1 预合成的 CosyVoice 临时 URL，或外部音频 URL） */
+  audio_url?: string
+  /** 预合成音频时长（秒），由 previewTts 返回 */
+  audio_duration?: number
+  /** 预合成接口返回的句子时间戳（精确），后端直接写入 job */
+  sentence_timings?: SentenceTiming[]
+  /** 音色 ID（TTS 直生模式用） */
+  voice_id?: string
+  /** 要合成的文案（TTS 直生模式用） */
+  script_text?: string
+  /** 语速 0.5~2.0，默认 1.0（TTS 直生模式用） */
+  speed?: number
+  /** 情绪英文枚举：natural/excited/calm/friendly（TTS 直生模式用） */
+  emotion?: string
+  enable_video_loop?: boolean
+  project_id?: string
 }): Promise<LipsyncJob> => {
   const response = await apiClient.post<LipsyncJob>("/lipsync/jobs", data)
   return response.data
 }
 
+/* ── #1845 TTS 预合成（步骤1「生成配音」同步接口，~2-3s） ── */
+export const previewTts = async (data: {
+  voice_id: string
+  script_text: string
+  speed?: number
+  emotion?: string
+}): Promise<{
+  audio_url: string
+  duration: number
+  sentence_timings: SentenceTiming[]
+}> => {
+  const response = await apiClient.post<{
+    audio_url: string
+    duration: number
+    sentence_timings: SentenceTiming[]
+  }>("/lipsync/tts-preview", data, { timeout: 30000 })
+  return response.data
+}
+
 export const getLipsyncJob = async (id: string): Promise<LipsyncJob> => {
-  const response = await apiClient.get<LipsyncJob>(`/lipsync/jobs/${id}`)
+  const response = await apiClient.get<LipsyncJob>(`/lipsync/jobs/${id}`, { timeout: 60000 })
   return response.data
 }
 
@@ -57,17 +96,30 @@ export const submitRender = async (data: {
   title_config?: Record<string, unknown>
   cover_config?: Record<string, unknown>
   project_id?: string
-  resolution?: string
 }): Promise<RenderJob> => {
+  // title_config 内可含 title_image_dataurl（前端 Canvas 渲染的 PNG dataURL）
   const response = await apiClient.post<RenderJob>("/ai-avatar/render", data)
   return response.data
 }
 
 export const getRenderJob = async (jobId: string): Promise<RenderJob> => {
-  const response = await apiClient.get<RenderJob>(`/ai-avatar/render/${jobId}`)
+  const response = await apiClient.get<RenderJob>(`/ai-avatar/render/${jobId}`, { timeout: 60000 })
   return response.data
 }
 
 export const cancelRenderJob = async (jobId: string): Promise<void> => {
   await apiClient.post(`/ai-avatar/render/${jobId}/cancel`)
 }
+
+/* ── 从最终渲染成片智能抽封面（POST /ai-avatar/renders/{job_id}/smart-cover） ── */
+export const generateRenderSmartCover = async (
+  jobId: string,
+): Promise<{ cover_url: string; status: string; message: string }> => {
+  const response = await apiClient.post<{ cover_url: string; status: string; message: string }>(
+    `/ai-avatar/render/${jobId}/smart-cover`,
+    {},
+    // 抽帧+评分+转存 OSS 链路较长，120s 超时
+    { timeout: 120000 },
+  )
+  return response.data
+}

apps/web/src/pages/ai-avatar/components/ModalBRollEditor.tsx

@@ -5,12 +5,12 @@
  * - 左侧：先选素材库（video 库）→ 再选该库视频素材（已被其他 segment 使用的素材
  *   标灰 + "已选择" 遮罩，pointer-events:none 防重复选择）
  * - 右侧：文案句子列表（点选对应段落，替代原数字索引框）/ 全屏 or 画中画 / 四角位置+大小
- *   （开始/结束时间已删除，按句子字数占比 × 口播总时长自动估算）
+ *   （开始/结束时间来自后端精确句子时间戳，基于 TTS 音频静音检测）
  * - 底部：已配置的画面插入列表（可删除）
  */
 import React, { useEffect, useMemo, useState } from "react"
 import { getAssets, getAssetLibraries, type AssetItem, type AssetLibraryItem } from "@/api/assets"
-import type { BRollSegment, BRollInsertMode, PipPosition } from "../types"
+import type { BRollSegment, BRollInsertMode, PipPosition, SentenceTiming } from "../types"
 import { splitScriptIntoSentences, type ScriptSentence } from "../utils/sentences"
 
 interface ModalBRollEditorProps {
@@ -18,10 +18,12 @@ interface ModalBRollEditorProps {
   onClose: () => void
   /** 当前已有的 B-roll segments（用于标灰已选素材） */
   existingSegments: BRollSegment[]
-  /** 当前文案全文（用于分句） */
+  /** 文案全文（优先使用对口型时锁定的 scriptText） */
   scriptText: string
-  /** 对口型成片总时长（秒），用于时间自动估算 */
+  /** 对口型成片总时长（秒） */
   outputDuration: number
+  /** 后端精确句子时间戳（来自 lipsyncJob.sentence_timings） */
+  sentenceTimings?: SentenceTiming[] | null
   onConfirm: (segment: BRollSegment) => void
   onRemove: (id: string) => void
 }
@@ -43,7 +45,8 @@ const ModalBRollEditor: React.FC<ModalBRollEditorProps> = ({
   onClose,
   existingSegments,
   scriptText,
-  outputDuration,
+  outputDuration: _outputDuration,
+  sentenceTimings,
   onConfirm,
   onRemove,
 }) => {
@@ -62,10 +65,10 @@ const ModalBRollEditor: React.FC<ModalBRollEditorProps> = ({
   const [pipPosition, setPipPosition] = useState<PipPosition>("top-right")
   const [pipScale, setPipScale] = useState(0.3)
 
-  /** 文案分句（⑤） */
+  /** 文案分句（优先使用后端精确时间戳，降级为字数比例估算） */
   const sentences = useMemo(
-    () => splitScriptIntoSentences(scriptText, outputDuration),
-    [scriptText, outputDuration],
+    () => splitScriptIntoSentences(scriptText, sentenceTimings, _outputDuration),
+    [scriptText, sentenceTimings, _outputDuration],
   )
 
   /** 已被现有 segments 占用的素材 id 集合（标灰、禁止重复选择） */
@@ -142,7 +145,7 @@ const ModalBRollEditor: React.FC<ModalBRollEditorProps> = ({
     setSelectedAsset(asset)
   }
 
-  /** 确认添加一段 B-roll（⑥ 时间取所选句子的估算起止） */
+  /** 确认添加一段 B-roll（⑥ 时间取所选句子的精确起止，后端静音检测 / 前端字数比例降级） */
   const handleConfirm = () => {
     if (!selectedAsset || !selectedSentence) return
     const startTime = selectedSentence.startTime
@@ -264,11 +267,9 @@ const ModalBRollEditor: React.FC<ModalBRollEditorProps> = ({
                           >
                             <span className="aa-sentence-item__idx">{sent.index + 1}</span>
                             <span className="aa-sentence-item__text">{sent.text}</span>
-                            {outputDuration > 0 && (
-                              <span className="aa-sentence-item__time">
-                                {sent.startTime.toFixed(1)}-{sent.endTime.toFixed(1)}s
-                              </span>
-                            )}
+                            <span className="aa-sentence-item__time">
+                              {sent.startTime.toFixed(1)}-{sent.endTime.toFixed(1)}s
+                            </span>
                           </button>
                         )
                       })}
@@ -349,7 +350,7 @@ const ModalBRollEditor: React.FC<ModalBRollEditorProps> = ({
                           selectedSentence.endTime,
                           selectedSentence.startTime + 0.5,
                         ).toFixed(1)}
-                        s （按字数自动估算）
+                        s
                       </div>
                     </>
                   ) : (

apps/web/src/pages/ai-avatar/components/PanelCoverAndGenerate.tsx

@@ -1,14 +1,15 @@
 /**
- * AI数字人 — 面板5：封面 & 生成
- * - 竖屏 9:16 封面预览（从视频截取 / 自定义上传）
- * - 分辨率选择（720p / 1080p / 4K）
- * - 配置汇总卡片（出镜视频/音色/文案/对口型/B-roll/标题/封面）
- * - 渐变紫色生成按钮
+ * AI数字人 — 面板5：分辨率/配置摘要/生成按钮/封面
+ * v3 调整（步骤③④）：
+ * - 布局顺序：分辨率 → 配置摘要卡片 → 🔘「开始生成视频」按钮 → （渲染完成后）封面区域
+ * - 渲染未完成时封面区域显示占位态，按钮 disabled
+ * - 「智能获取封面」从最终成片抽帧（调用 POST /renders/{id}/smart-cover），不再依赖 lipsync 状态
+ * - 修复点 2 次 bug：内部维护 smartCoverLoading，不依赖外层异步 state 更新
  *
  * 注意：v3 已删除"画面插入模式"，本面板不包含该选项。
  */
-import React, { useRef } from "react"
-import type { AiAvatarCoverConfig } from "../types"
+import React, { useRef, useState } from "react"
+import type { AiAvatarCoverConfig, RenderJob } from "../types"
 
 interface PanelCoverAndGenerateProps {
   coverConfig: AiAvatarCoverConfig
@@ -17,6 +18,12 @@ interface PanelCoverAndGenerateProps {
   onResolutionChange: (r: string) => void
   isGenerating: boolean
   onGenerate: () => void
+  /** 当前渲染任务（渲染完成后才有 output_video_url，才能抽封面） */
+  renderJob: RenderJob | null
+  /** 从最终成片智能抽帧（参数 renderId），返回 { cover_url } */
+  onGenerateRenderSmartCover: (renderId: string) => Promise<{ cover_url: string; message?: string }>
+  /** 自定义上传封面（选择本地文件后由父组件处理实际上传） */
+  onUploadCover?: (file: File) => void
   /** 配置汇总信息 */
   summary: {
     videoName: string | null
@@ -25,7 +32,8 @@ interface PanelCoverAndGenerateProps {
     lipsyncStatus: string | null
     brollCount: number
     hasTitle: boolean
-    hasCover: boolean
+    /** 封面状态：'not_ready'(视频未生成) / 'pending'(视频生成了但未选) / 'selected'(已选) */
+    coverStatus: "not_ready" | "pending" | "selected"
   }
 }
 
@@ -50,9 +58,14 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
   onResolutionChange,
   isGenerating,
   onGenerate,
+  renderJob,
+  onGenerateRenderSmartCover,
+  onUploadCover,
   summary,
 }) => {
   const uploadInputRef = useRef<HTMLInputElement>(null)
+  // 内部维护智能封面加载态（修复点 2 次 bug：不依赖外层异步 setState 顺序）
+  const [smartCoverLoading, setSmartCoverLoading] = useState(false)
 
   /** 自定义上传封面 */
   const handleUploadClick = () => {
@@ -62,57 +75,66 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
   const handleFileChange = (e: React.ChangeEvent<HTMLInputElement>) => {
     const file = e.target.files?.[0]
     if (!file) return
-    // 本地预览：生成 object URL（实际上传由父级/后端链路处理）
-    const url = URL.createObjectURL(file)
-    onCoverConfigChange({ mode: "upload", upload_url: url, thumbnail_url: url })
-    // 允许重复选择同一文件
+    if (onUploadCover) {
+      onUploadCover(file)
+    } else {
+      // 本地预览兜底（实际上传由父级处理；blob URL 仅作本地展示）
+      const url = URL.createObjectURL(file)
+      onCoverConfigChange({ mode: "upload", upload_url: url, thumbnail_url: url })
+    }
     e.target.value = ""
   }
 
-  /** 从视频截取（使用配置的帧时间，默认首帧） */
-  const handleCaptureFromVideo = () => {
-    onCoverConfigChange({ mode: "auto_frame" })
+  /** 智能获取封面（从最终成片抽帧；必须等 render 完成） */
+  const handleSmartCover = async () => {
+    if (!renderJob || renderJob.status !== "completed" || !renderJob.id) return
+    setSmartCoverLoading(true)
+    try {
+      const res = await onGenerateRenderSmartCover(renderJob.id)
+      if (res.cover_url) {
+        onCoverConfigChange({
+          mode: "auto_frame",
+          smart_cover_url: res.cover_url,
+          thumbnail_url: res.cover_url,
+        })
+      } else {
+        // 失败由父组件 message 提示，这里不重复弹窗
+        console.warn("[智能封面] 返回空 cover_url:", res.message)
+      }
+    } catch (err) {
+      console.error("[智能封面] 调用失败:", err)
+    } finally {
+      setSmartCoverLoading(false)
+    }
   }
 
   const lipsync = summary.lipsyncStatus ? LIPSYNC_STATUS_LABEL[summary.lipsyncStatus] : null
-
   const canGenerate = summary.lipsyncStatus === "completed" && !isGenerating
+  // 渲染已完成 → 封面区可用
+  const isRenderCompleted = renderJob?.status === "completed"
+  const canSmartCover = isRenderCompleted && !smartCoverLoading
+
+  /** 封面图实际展示的 url：智能封面 > 自定义上传 > 空 */
+  const coverUrl =
+    coverConfig.smart_cover_url || coverConfig.thumbnail_url || coverConfig.upload_url
+  const hasCoverImage = Boolean(coverUrl)
+
+  /** 封面区占位文字 */
+  const coverPlaceholder = isRenderCompleted ? "暂无封面" : "视频生成后可选择封面"
+
+  /** 封面摘要状态文本 */
+  const coverSummaryNode = (() => {
+    if (summary.coverStatus === "selected") {
+      return <span className="aa-config-summary__value">已选择</span>
+    }
+    if (summary.coverStatus === "pending") {
+      return <span className="aa-config-summary__value">待选择</span>
+    }
+    return <span className="aa-config-summary__empty">生成视频后可选</span>
+  })()
 
   return (
     <div className="aa-cover-generate">
-      {/* 封面预览（竖屏 9:16） */}
-      <div className="aa-cover-preview">
-        {coverConfig.thumbnail_url ? (
-          <img src={coverConfig.thumbnail_url} alt="封面预览" />
-        ) : (
-          <span className="aa-cover-preview__placeholder">暂无封面</span>
-        )}
-      </div>
-
-      <div className="aa-cover-actions">
-        <button
-          type="button"
-          className={`aa-btn aa-btn--ghost${coverConfig.mode === "auto_frame" ? " active" : ""}`}
-          onClick={handleCaptureFromVideo}
-        >
-          🎬 从视频截取
-        </button>
-        <button
-          type="button"
-          className={`aa-btn aa-btn--ghost${coverConfig.mode === "upload" ? " active" : ""}`}
-          onClick={handleUploadClick}
-        >
-          📷 自定义上传
-        </button>
-        <input
-          ref={uploadInputRef}
-          type="file"
-          accept="image/*"
-          style={{ display: "none" }}
-          onChange={handleFileChange}
-        />
-      </div>
-
       {/* 分辨率选择 */}
       <div className="aa-form-field">
         <label className="aa-label">分辨率</label>
@@ -120,6 +142,7 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
           className="aa-select"
           value={resolution}
           onChange={(e) => onResolutionChange(e.target.value)}
+          disabled={isGenerating}
         >
           {RESOLUTION_OPTIONS.map((opt) => (
             <option key={opt.value} value={opt.value}>
@@ -180,11 +203,7 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
           </div>
           <div className="aa-config-summary__row">
             <span>封面</span>
-            {summary.hasCover ? (
-              <span className="aa-config-summary__value">已开启</span>
-            ) : (
-              <span className="aa-config-summary__empty">未配置</span>
-            )}
+            {coverSummaryNode}
           </div>
         </div>
 
@@ -202,6 +221,55 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
             请先完成对口型生成
           </div>
         )}
+        {isGenerating && (
+          <div style={{ marginTop: 8, fontSize: 11, color: "#8c8ca1", textAlign: "center" }}>
+            视频生成中，请稍候…
+          </div>
+        )}
+      </div>
+
+      {/* 封面区域（视频生成后才激活；步骤③④要求：按钮在封面上方，完成后再显示封面区） */}
+      <div className="aa-cover-section" style={{ marginTop: 16 }}>
+        <div className="aa-label" style={{ marginBottom: 8 }}>
+          封面
+        </div>
+        {/* 封面预览（竖屏 9:16）——成片帧已经通过 Canvas PNG overlay 带有标题，直接展示原图即可 */}
+        <div className="aa-cover-preview" style={{ opacity: isRenderCompleted ? 1 : 0.5 }}>
+          {hasCoverImage ? (
+            <img src={coverUrl!} alt="封面预览" draggable={false} />
+          ) : (
+            <span className="aa-cover-preview__placeholder">{coverPlaceholder}</span>
+          )}
+          {smartCoverLoading && <div className="aa-cover-preview__loading">⏳ 智能选帧中…</div>}
+        </div>
+
+        <div className="aa-cover-actions">
+          <button
+            type="button"
+            className={`aa-btn aa-btn--ghost${coverConfig.mode === "auto_frame" ? " active" : ""}`}
+            onClick={handleSmartCover}
+            disabled={!canSmartCover}
+            title={isRenderCompleted ? "从成片智能选帧" : "请先生成视频"}
+          >
+            {smartCoverLoading ? "⏳ 智能选帧中…" : "🎬 智能获取封面"}
+          </button>
+          <button
+            type="button"
+            className={`aa-btn aa-btn--ghost${coverConfig.mode === "upload" ? " active" : ""}`}
+            onClick={handleUploadClick}
+            disabled={!isRenderCompleted || smartCoverLoading}
+            title={isRenderCompleted ? "自定义上传封面" : "请先生成视频"}
+          >
+            📷 自定义上传
+          </button>
+          <input
+            ref={uploadInputRef}
+            type="file"
+            accept="image/*"
+            style={{ display: "none" }}
+            onChange={handleFileChange}
+          />
+        </div>
       </div>
     </div>
   )

apps/web/src/pages/ai-avatar/components/PanelLipsyncPreview.tsx

@@ -0,0 +1,338 @@
+/**
+ * AI数字人 — 对口型预览面板（步骤2用）
+ * B-roll 画面插入 + 对口型视频预览 + 生成/重新生成按钮
+ * v3.1: 标题字号按预览容器实际宽度动态计算 previewScale（基准 720p），与成片一致
+ */
+import React, { useCallback, useEffect, useRef, useState } from "react"
+import type { LipsyncJob, BRollSegment, AiAvatarTitleConfig } from "../types"
+
+interface PanelLipsyncPreviewProps {
+  lipsyncJob: LipsyncJob | null
+  onGenerateLipsync: () => void
+  bRollSegments: BRollSegment[]
+  onOpenBRollModal: () => void
+  onRemoveBRoll: (id: string) => void
+  /** 标题配置（实时叠加预览用） */
+  titleConfig?: AiAvatarTitleConfig
+  /** 标题位置变更回调（拖拽结束时调用，发送百分比坐标 + position:"custom"） */
+  onTitlePositionChange?: (pos: { pos_x: number; pos_y: number; position: string }) => void
+}
+
+const BROLL_MODE_LABEL: Record<BRollSegment["mode"], string> = {
+  fullscreen: "全屏",
+  pip: "画中画",
+}
+
+function formatTime(seconds: number): string {
+  const m = Math.floor(seconds / 60)
+  const s = Math.round(seconds % 60)
+  return `${m}:${s.toString().padStart(2, "0")}`
+}
+
+/** 字体名 → CSS font-family 映射（与 titleCanvas 字体链对齐） */
+const FONT_FAMILY_MAP: Record<string, string> = {
+  思源黑体: "'Noto Sans CJK SC', 'Source Han Sans CN', 'PingFang SC', 'Microsoft YaHei', sans-serif",
+  思源宋体: "'Noto Serif SC', 'Source Han Serif SC', 'SimSun', serif",
+  楷体: "KaiTi, 'STKaiti', serif",
+  黑体: "'Heiti SC', 'SimHei', 'Microsoft YaHei', sans-serif",
+}
+const getFontFamily = (font: string): string =>
+  FONT_FAMILY_MAP[font] || FONT_FAMILY_MAP["思源黑体"]
+
+export function PanelLipsyncPreview({
+  lipsyncJob,
+  onGenerateLipsync,
+  bRollSegments,
+  onOpenBRollModal,
+  onRemoveBRoll,
+  titleConfig,
+  onTitlePositionChange,
+}: PanelLipsyncPreviewProps) {
+  const titleDragRef = useRef<HTMLDivElement>(null)
+  const draggingTitleRef = useRef(false)
+  const previewContainerRef = useRef<HTMLDivElement>(null)
+  // 预览容器实际宽度（通过 ResizeObserver 监听），用于动态计算 previewScale
+  const [containerWidth, setContainerWidth] = useState(0)
+  const isGenerating = lipsyncJob?.status === "pending" || lipsyncJob?.status === "processing"
+  const isDone = lipsyncJob?.status === "completed"
+  const isFailed = lipsyncJob?.status === "failed"
+
+  const statusText =
+    lipsyncJob?.status === "processing"
+      ? "对口型生成中…"
+      : lipsyncJob?.status === "pending"
+        ? "排队中…"
+        : "对口型生成中…"
+
+  // 监听预览容器尺寸变化，动态测量宽度以计算 previewScale（基准 720p）
+  useEffect(() => {
+    const el = previewContainerRef.current
+    if (!el) return
+    const update = () => setContainerWidth(el.clientWidth || 0)
+    update()
+    if (typeof ResizeObserver !== "undefined") {
+      const ro = new ResizeObserver(update)
+      ro.observe(el)
+      return () => ro.disconnect()
+    }
+    window.addEventListener("resize", update)
+    return () => window.removeEventListener("resize", update)
+  }, [])
+
+  // 预览缩放比：预览宽度 / 720（基准宽度）
+  const previewScale = containerWidth > 0 ? containerWidth / 720 : 0.35
+  const ps = useCallback((v: number) => Math.round(v * previewScale * 100) / 100, [previewScale])
+
+  /** 标题叠加样式（字号/padding/描边/阴影均按 previewScale 缩放，保持与成片视觉一致） */
+  const titleOverlayStyle: React.CSSProperties | null =
+    titleConfig?.title && containerWidth > 0
+      ? (() => {
+          const baseSize = titleConfig.size || 48
+          const fontSize = ps(baseSize)
+          // 描边宽度基准 ≈ size * 0.06，最小 1.5px @720p
+          const strokeW = Math.max(ps(1.5), +(baseSize * 0.06 * previewScale).toFixed(2))
+          // 阴影按比例缩放
+          const shadowBlur = ps(4)
+          const shadowOffsetY = ps(2)
+          // padding / top 边距按比例（基准 8px 对应预览小窗，成片基准 16px，这里 8px 对应约 0.33 缩放）
+          const padV = ps(16) * 0.5 // ≈ 8px in ~240px container
+          const padH = ps(24) * 0.5
+
+          const style: React.CSSProperties = {
+            position: "absolute",
+            color: titleConfig.color || "#ffffff",
+            fontFamily: getFontFamily(titleConfig.font || "思源黑体"),
+            fontSize: `${fontSize}px`,
+            fontWeight: titleConfig.bold ? 700 : 400,
+            fontStyle: titleConfig.italic ? "italic" : "normal",
+            textAlign: "center",
+            width: "90%",
+            lineHeight: 1.2,
+            padding: `${ps(4)}px ${padH}px`,
+            textShadow: titleConfig.shadow
+              ? `0 ${shadowOffsetY}px ${shadowBlur}px rgba(0,0,0,0.8), 0 0 ${ps(2)}px rgba(0,0,0,0.5)`
+              : undefined,
+            WebkitTextStroke: titleConfig.stroke ? `${strokeW}px #000` : undefined,
+            boxSizing: "border-box",
+            wordBreak: "break-word",
+            whiteSpace: "pre-wrap",
+          }
+
+          if (
+            titleConfig.position === "custom" &&
+            titleConfig.pos_x != null &&
+            titleConfig.pos_y != null
+          ) {
+            style.left = `${titleConfig.pos_x}%`
+            style.top = `${titleConfig.pos_y}%`
+            style.transform = "translateX(-50%) translateY(-50%)"
+          } else if (titleConfig.position === "top") {
+            style.left = "50%"
+            style.top = padV
+            style.transform = "translateX(-50%)"
+          } else if (titleConfig.position === "bottom") {
+            style.left = "50%"
+            style.bottom = padV
+            style.transform = "translateX(-50%)"
+          } else {
+            style.left = "50%"
+            style.top = "50%"
+            style.transform = "translateX(-50%) translateY(-50%)"
+          }
+          return style
+        })()
+      : null
+
+  const handleTitlePointerDown = (e: React.PointerEvent<HTMLDivElement>) => {
+    if (!onTitlePositionChange || !previewContainerRef.current) return
+    e.preventDefault()
+    e.stopPropagation()
+    ;(e.target as Element).setPointerCapture(e.pointerId)
+    draggingTitleRef.current = true
+    ;(e.currentTarget as HTMLDivElement).style.cursor = "grabbing"
+  }
+  const handleTitlePointerMove = (e: React.PointerEvent<HTMLDivElement>) => {
+    if (!draggingTitleRef.current || !previewContainerRef.current) return
+    e.preventDefault()
+    e.stopPropagation()
+    if (titleDragRef.current) {
+      const rect = previewContainerRef.current.getBoundingClientRect()
+      const relX = Math.max(0, Math.min(rect.width, e.clientX - rect.left))
+      const relY = Math.max(0, Math.min(rect.height, e.clientY - rect.top))
+      const xpct = (relX / rect.width) * 100
+      const ypct = (relY / rect.height) * 100
+      titleDragRef.current.style.left = `${xpct}%`
+      titleDragRef.current.style.top = `${ypct}%`
+    }
+  }
+  const handleTitlePointerUp = (e: React.PointerEvent<HTMLDivElement>) => {
+    if (!draggingTitleRef.current) return
+    draggingTitleRef.current = false
+    if (onTitlePositionChange && previewContainerRef.current) {
+      const rect = previewContainerRef.current.getBoundingClientRect()
+      const relX = Math.max(0, Math.min(rect.width, e.clientX - rect.left))
+      const relY = Math.max(0, Math.min(rect.height, e.clientY - rect.top))
+      // 发送百分比坐标（0-100），与后端 drawtext 百分比表达式对齐
+      const xpct = Math.round((relX / rect.width) * 1000) / 10
+      const ypct = Math.round((relY / rect.height) * 1000) / 10
+      onTitlePositionChange({ pos_x: xpct, pos_y: ypct, position: "custom" })
+    }
+    ;(e.currentTarget as HTMLDivElement).style.cursor = "grab"
+  }
+
+  return (
+    <div className="aa-script-lipsync">
+      {/* ── B-roll 画面 ── */}
+      <div className="aa-lipsync-section">
+        <div className="aa-lipsync-section__title">
+          <span style={{ marginRight: 8 }}>🎞️ 插入画面</span>
+          {bRollSegments.length > 0 && (
+            <span className="aa-broll-badge">🎬 {bRollSegments.length} 个画面</span>
+          )}
+        </div>
+        <div className="aa-lipsync-actions">
+          <button
+            type="button"
+            className="aa-btn aa-btn--primary aa-btn--full"
+            onClick={onOpenBRollModal}
+          >
+            🎬 插入画面
+          </button>
+        </div>
+
+        {bRollSegments.length > 0 && (
+          <div className="aa-broll-list">
+            {bRollSegments.map((seg) => (
+              <div key={seg.id} className="aa-broll-item">
+                {seg.asset.thumbnail_url || seg.asset.file_url ? (
+                  <img
+                    className="aa-broll-item__thumb"
+                    src={seg.asset.thumbnail_url || seg.asset.file_url}
+                    alt={seg.asset.name}
+                  />
+                ) : (
+                  <span className="aa-broll-item__thumb" style={{ padding: "6px 4px" }}>
+                    🎬
+                  </span>
+                )}
+                <div className="aa-broll-item__info">
+                  <div
+                    style={{
+                      overflow: "hidden",
+                      textOverflow: "ellipsis",
+                      whiteSpace: "nowrap",
+                    }}
+                  >
+                    {seg.asset.name}
+                  </div>
+                  <div style={{ fontSize: 11, color: "#8c8ca1", marginTop: 2 }}>
+                    {BROLL_MODE_LABEL[seg.mode]} · {formatTime(seg.start_time)}-
+                    {formatTime(seg.end_time)}
+                  </div>
+                </div>
+                <button
+                  type="button"
+                  className="aa-broll-item__remove"
+                  title="删除"
+                  onClick={() => onRemoveBRoll(seg.id)}
+                >
+                  ✕
+                </button>
+              </div>
+            ))}
+          </div>
+        )}
+      </div>
+
+      {/* ── 对口型预览（标题字号按 previewScale 动态缩放） ─ */}
+      <div className="aa-lipsync-section">
+        <div className="aa-lipsync-section__title">对口型预览</div>
+
+        <div className="aa-lipsync-preview" ref={previewContainerRef}>
+          {isDone && lipsyncJob?.output_video_url ? (
+            <div style={{ position: "relative", width: "100%", height: "100%" }}>
+              <video src={lipsyncJob.output_video_url} controls />
+              {titleOverlayStyle && (
+                <div
+                  ref={titleDragRef}
+                  style={{
+                    ...titleOverlayStyle,
+                    cursor: onTitlePositionChange ? "grab" : "default",
+                    pointerEvents: onTitlePositionChange ? "auto" : "none",
+                  }}
+                  onPointerDown={handleTitlePointerDown}
+                  onPointerMove={handleTitlePointerMove}
+                  onPointerUp={handleTitlePointerUp}
+                  onPointerCancel={handleTitlePointerUp}
+                >
+                  {titleConfig!.title.split(/[/／]/).map((part, i) => (
+                    <span key={i}>
+                      {i > 0 && <br />}
+                      {part}
+                    </span>
+                  ))}
+                </div>
+              )}
+            </div>
+          ) : isGenerating ? (
+            <div style={{ width: "80%", textAlign: "center", color: "#fff" }}>
+              <div style={{ fontSize: 13, marginBottom: 8 }}>
+                {statusText} {Math.round(lipsyncJob?.progress ?? 0)}%
+              </div>
+              <div className="aa-progress">
+                <div
+                  className="aa-progress__bar"
+                  style={{ width: `${lipsyncJob?.progress ?? 0}%` }}
+                />
+              </div>
+            </div>
+          ) : (
+            <div className="aa-video-preview__placeholder">
+              {isFailed ? (
+                <>
+                  <div style={{ fontSize: 28, marginBottom: 8 }}>❌</div>
+                  <div>对口型生成失败</div>
+                  {lipsyncJob?.error_message && (
+                    <div
+                      style={{
+                        fontSize: 11,
+                        marginTop: 4,
+                        color: "#fca5a5",
+                      }}
+                    >
+                      {lipsyncJob.error_message}
+                    </div>
+                  )}
+                </>
+              ) : (
+                "生成对口型视频后在此预览"
+              )}
+            </div>
+          )}
+        </div>
+
+        <div className="aa-lipsync-actions">
+          {isDone ? (
+            <button type="button" className="aa-btn aa-btn--full" onClick={onGenerateLipsync}>
+              🔄 重新生成对口型
+            </button>
+          ) : isGenerating ? (
+            <button type="button" className="aa-btn aa-btn--full" disabled>
+              对口型生成中…
+            </button>
+          ) : (
+            <button
+              type="button"
+              className="aa-btn aa-btn--primary aa-btn--full"
+              onClick={onGenerateLipsync}
+            >
+              🎬 生成对口型视频
+            </button>
+          )}
+        </div>
+      </div>
+    </div>
+  )
+}
+
+export default PanelLipsyncPreview

apps/web/src/pages/ai-avatar/components/PanelScript.tsx

@@ -0,0 +1,67 @@
+/**
+ * AI数字人 — 文案面板（步骤1用）
+ * 文案库选择 / 手动输入 + 字数统计
+ */
+import { useState } from "react"
+
+interface PanelScriptProps {
+  scriptText: string
+  onScriptTextChange: (text: string) => void
+  onOpenScriptModal: () => void
+}
+
+type ScriptTab = "library" | "manual"
+
+export function PanelScript({
+  scriptText,
+  onScriptTextChange,
+  onOpenScriptModal,
+}: PanelScriptProps) {
+  const [scriptTab, setScriptTab] = useState<ScriptTab>("library")
+
+  return (
+    <div className="aa-script-lipsync">
+      {/* ── Tab 切换 ── */}
+      <div className="aa-script-tabs">
+        <button
+          type="button"
+          className={`aa-script-tab${scriptTab === "library" ? " active" : ""}`}
+          onClick={() => setScriptTab("library")}
+        >
+          从文案库选择
+        </button>
+        <button
+          type="button"
+          className={`aa-script-tab${scriptTab === "manual" ? " active" : ""}`}
+          onClick={() => setScriptTab("manual")}
+        >
+          手动输入
+        </button>
+      </div>
+
+      {scriptTab === "library" && (
+        <button
+          type="button"
+          className="aa-btn aa-btn--ghost aa-btn--full"
+          style={{ marginBottom: 8 }}
+          onClick={onOpenScriptModal}
+        >
+          📚 从文案库选择文案
+        </button>
+      )}
+
+      <textarea
+        className="aa-textarea"
+        value={scriptText}
+        readOnly={scriptTab === "library"}
+        placeholder={
+          scriptTab === "library" ? "点击上方按钮，从文案库选择文案…" : "请输入数字人口播文案…"
+        }
+        onChange={(e) => onScriptTextChange(e.target.value)}
+      />
+      <div className="aa-char-count">{scriptText.length} 字</div>
+    </div>
+  )
+}
+
+export default PanelScript

apps/web/src/pages/ai-avatar/components/PanelScriptAndLipsync.tsx

@@ -1,222 +0,0 @@
-/**
- * AI数字人 — 文案 & 对口型面板（面板4）
- * 上半区：文案（文案库选择 / 手动输入）；下半区：对口型视频预览（9:16）+ B-roll 画面
- */
-import { useState } from "react"
-import type { LipsyncJob, BRollSegment } from "../types"
-
-interface PanelScriptAndLipsyncProps {
-  scriptText: string
-  onScriptTextChange: (text: string) => void
-  onOpenScriptModal: () => void
-  lipsyncJob: LipsyncJob | null
-  onGenerateLipsync: () => void
-  bRollSegments: BRollSegment[]
-  onOpenBRollModal: () => void
-  onRemoveBRoll: (id: string) => void
-}
-
-type ScriptTab = "library" | "manual"
-
-const BROLL_MODE_LABEL: Record<BRollSegment["mode"], string> = {
-  fullscreen: "全屏",
-  pip: "画中画",
-}
-
-function formatTime(seconds: number): string {
-  const m = Math.floor(seconds / 60)
-  const s = Math.round(seconds % 60)
-  return `${m}:${s.toString().padStart(2, "0")}`
-}
-
-export function PanelScriptAndLipsync({
-  scriptText,
-  onScriptTextChange,
-  onOpenScriptModal,
-  lipsyncJob,
-  onGenerateLipsync,
-  bRollSegments,
-  onOpenBRollModal,
-  onRemoveBRoll,
-}: PanelScriptAndLipsyncProps) {
-  const [scriptTab, setScriptTab] = useState<ScriptTab>("library")
-
-  /* 对口型状态判断 */
-  const isGenerating = lipsyncJob?.status === "pending" || lipsyncJob?.status === "processing"
-  const isDone = lipsyncJob?.status === "completed"
-  const isFailed = lipsyncJob?.status === "failed"
-
-  const statusText =
-    lipsyncJob?.status === "processing"
-      ? "对口型生成中…"
-      : lipsyncJob?.status === "pending"
-        ? "排队中…"
-        : "对口型生成中…"
-
-  return (
-    <div className="aa-script-lipsync">
-      {/* ── 上半区：文案 ── */}
-      <div className="aa-script-tabs">
-        <button
-          type="button"
-          className={`aa-script-tab${scriptTab === "library" ? " active" : ""}`}
-          onClick={() => setScriptTab("library")}
-        >
-          从文案库选择
-        </button>
-        <button
-          type="button"
-          className={`aa-script-tab${scriptTab === "manual" ? " active" : ""}`}
-          onClick={() => setScriptTab("manual")}
-        >
-          手动输入
-        </button>
-      </div>
-
-      {scriptTab === "library" && (
-        <button
-          type="button"
-          className="aa-btn aa-btn--ghost aa-btn--full"
-          style={{ marginBottom: 8 }}
-          onClick={onOpenScriptModal}
-        >
-          📚 从文案库选择文案
-        </button>
-      )}
-
-      <textarea
-        className="aa-textarea"
-        value={scriptText}
-        readOnly={scriptTab === "library"}
-        placeholder={
-          scriptTab === "library" ? "点击上方按钮，从文案库选择文案…" : "请输入数字人口播文案…"
-        }
-        onChange={(e) => onScriptTextChange(e.target.value)}
-      />
-      <div className="aa-char-count">{scriptText.length} 字</div>
-
-      {/* ── B-roll 画面 ── */}
-      <div className="aa-lipsync-section">
-        <div className="aa-lipsync-section__title">
-          <span style={{ marginRight: 8 }}>🎞️ 插入画面</span>
-          {bRollSegments.length > 0 && (
-            <span className="aa-broll-badge">🎬 {bRollSegments.length} 个画面</span>
-          )}
-        </div>
-        <div className="aa-lipsync-actions">
-          <button
-            type="button"
-            className="aa-btn aa-btn--primary aa-btn--full"
-            onClick={onOpenBRollModal}
-          >
-            🎬 插入画面
-          </button>
-        </div>
-
-        {bRollSegments.length > 0 && (
-          <div className="aa-broll-list">
-            {bRollSegments.map((seg) => (
-              <div key={seg.id} className="aa-broll-item">
-                {seg.asset.thumbnail_url || seg.asset.file_url ? (
-                  <img
-                    className="aa-broll-item__thumb"
-                    src={seg.asset.thumbnail_url || seg.asset.file_url}
-                    alt={seg.asset.name}
-                  />
-                ) : (
-                  <span className="aa-broll-item__thumb" style={{ padding: "6px 4px" }}>
-                    🎬
-                  </span>
-                )}
-                <div className="aa-broll-item__info">
-                  <div
-                    style={{
-                      overflow: "hidden",
-                      textOverflow: "ellipsis",
-                      whiteSpace: "nowrap",
-                    }}
-                  >
-                    {seg.asset.name}
-                  </div>
-                  <div style={{ fontSize: 11, color: "#8c8ca1", marginTop: 2 }}>
-                    {BROLL_MODE_LABEL[seg.mode]} · {formatTime(seg.start_time)}-
-                    {formatTime(seg.end_time)}
-                  </div>
-                </div>
-                <button
-                  type="button"
-                  className="aa-broll-item__remove"
-                  title="删除"
-                  onClick={() => onRemoveBRoll(seg.id)}
-                >
-                  ✕
-                </button>
-              </div>
-            ))}
-          </div>
-        )}
-      </div>
-
-      {/* ── 下半区：对口型预览（竖屏 9:16） ── */}
-      <div className="aa-lipsync-section">
-        <div className="aa-lipsync-section__title">对口型预览</div>
-
-        <div className="aa-lipsync-preview">
-          {isDone && lipsyncJob?.output_video_url ? (
-            <video src={lipsyncJob.output_video_url} controls />
-          ) : isGenerating ? (
-            <div style={{ width: "80%", textAlign: "center", color: "#fff" }}>
-              <div style={{ fontSize: 13, marginBottom: 8 }}>
-                {statusText} {Math.round(lipsyncJob?.progress ?? 0)}%
-              </div>
-              <div className="aa-progress">
-                <div
-                  className="aa-progress__bar"
-                  style={{ width: `${lipsyncJob?.progress ?? 0}%` }}
-                />
-              </div>
-            </div>
-          ) : (
-            <div className="aa-video-preview__placeholder">
-              {isFailed ? (
-                <>
-                  <div style={{ fontSize: 28, marginBottom: 8 }}>❌</div>
-                  <div>对口型生成失败</div>
-                  {lipsyncJob?.error_message && (
-                    <div style={{ fontSize: 11, marginTop: 4, color: "#fca5a5" }}>
-                      {lipsyncJob.error_message}
-                    </div>
-                  )}
-                </>
-              ) : (
-                "生成对口型视频后在此预览"
-              )}
-            </div>
-          )}
-        </div>
-
-        <div className="aa-lipsync-actions">
-          {isDone ? (
-            <button type="button" className="aa-btn aa-btn--full" onClick={onGenerateLipsync}>
-              🔄 重新生成对口型
-            </button>
-          ) : isGenerating ? (
-            <button type="button" className="aa-btn aa-btn--full" disabled>
-              对口型生成中…
-            </button>
-          ) : (
-            <button
-              type="button"
-              className="aa-btn aa-btn--primary aa-btn--full"
-              onClick={onGenerateLipsync}
-            >
-              🎬 生成对口型视频
-            </button>
-          )}
-        </div>
-      </div>
-    </div>
-  )
-}
-
-export default PanelScriptAndLipsync

apps/web/src/pages/ai-avatar/components/PanelTitleConfig.tsx

@@ -7,16 +7,17 @@
  * - AiAvatarTitleConfig ↔ TitleSettings 的双向适配
  * - 自动生成字幕开关
  */
-import React, { useMemo, useState } from "react"
+import React, { useMemo, useState, useEffect } from "react"
+import { Input } from "antd"
 import TitleStylePanel from "@/pages/generate/components/title/TitleStylePanel"
+import TitleLibraryAutoComplete from "@/pages/generate/components/title/TitleLibraryAutoComplete"
+import type { TitleOption } from "@/pages/generate/components/title/TitleLibraryAutoComplete"
 import type { TitleSettings } from "@/pages/generate/types"
-import {
-  POSITION_OPTIONS,
-  FONT_OPTIONS,
-  TITLE_PRESETS,
-  getFontFamily,
-} from "@/pages/generate/constants"
+import { POSITION_OPTIONS, FONT_OPTIONS, TITLE_PRESETS } from "@/pages/generate/constants"
 import type { AiAvatarTitleConfig } from "../types"
+import { getTitles } from "@/api/titles"
+
+const { TextArea } = Input
 
 interface PanelTitleConfigProps {
   titleConfig: AiAvatarTitleConfig
@@ -27,6 +28,14 @@ const PanelTitleConfig: React.FC<PanelTitleConfigProps> = ({ titleConfig, onUpda
   /** TitleStylePanel 内部高亮的预设 key（面板本地状态） */
   const [activePreset, setActivePreset] = useState<string | null>(null)
 
+  /** 标题库选项（复用智能剪辑的标题库） */
+  const [titleOptions, setTitleOptions] = useState<TitleOption[]>([])
+  useEffect(() => {
+    getTitles()
+      .then((items) => setTitleOptions(items.map((t) => ({ label: t.content, value: t.content }))))
+      .catch(() => setTitleOptions([]))
+  }, [])
+
   /** AiAvatarTitleConfig → TitleSettings（补齐 aiAutoSelect / 自由坐标字段） */
   const titleSettings: TitleSettings = useMemo(
     () => ({
@@ -62,37 +71,32 @@ const PanelTitleConfig: React.FC<PanelTitleConfigProps> = ({ titleConfig, onUpda
 
   return (
     <div className="aa-title-config">
-      {/* 主标题输入 */}
+      {/* 主标题输入 — TextArea 多行 + 标题库选择 */}
       <div className="aa-form-field">
         <label className="aa-label">主标题</label>
-        <input
-          className="aa-input aa-title-input"
-          type="text"
-          placeholder="输入视频标题（留空则不显示标题）"
+        <TextArea
+          className="aa-title-input"
+          placeholder="输入视频标题（支持 / 分行）"
           value={titleConfig.title}
-          maxLength={30}
+          autoSize={{ minRows: 2, maxRows: 4 }}
+          maxLength={200}
           onChange={(e) => onUpdate({ title: e.target.value })}
+          style={{ fontSize: 15 }}
         />
-        {titleConfig.title && (
-          <div
-            style={{
-              fontSize: 13,
-              padding: "6px 8px",
-              background: "#f8f8fc",
-              borderRadius: 6,
-              fontFamily: getFontFamily(titleConfig.font),
-              fontWeight: titleConfig.bold ? 700 : 400,
-              fontStyle: titleConfig.italic ? "italic" : "normal",
-              color: titleConfig.color,
-              textShadow: titleConfig.shadow ? "1px 1px 3px rgba(0,0,0,0.6)" : undefined,
-              overflow: "hidden",
-              textOverflow: "ellipsis",
-              whiteSpace: "nowrap",
+        <div style={{ marginTop: 8, display: "flex", alignItems: "center", gap: 8 }}>
+          <span style={{ fontSize: 12, color: "#8c8ca1", whiteSpace: "nowrap" }}>📚 标题库</span>
+          <TitleLibraryAutoComplete
+            key={titleConfig.title}
+            placeholder="选择标题填入上方"
+            value=""
+            onChange={(val) => {
+              if (val) onUpdate({ title: val })
             }}
-          >
-            {titleConfig.title}
-          </div>
-        )}
+            options={titleOptions}
+            maxLength={200}
+            style={{ flex: 1 }}
+          />
+        </div>
       </div>
 
       {/* 标题样式：直接复用智能剪辑 TitleStylePanel（位置/字体/字号/样式/预设） */}

apps/web/src/pages/ai-avatar/components/PanelVideoSelector.tsx

@@ -2,6 +2,8 @@
  * AI数字人 — 出镜视频选择面板
  * - 未选视频：虚线上传区，点击打开素材库弹窗
  * - 已选视频：竖屏 9:16 预览播放器 + 视频信息卡片 + 移除按钮
+ *
+ * 注意：本面板只展示原始素材视频，不叠加标题（标题在对口型预览和最终成片上展示）
  */
 import type { AssetItem } from "@/api/assets"
 
@@ -53,7 +55,7 @@ export function PanelVideoSelector({
 
   return (
     <div>
-      {/* 竖屏 9:16 视频预览播放器 */}
+      {/* 竖屏 9:16 视频预览播放器（纯素材预览，不叠加标题） */}
       <div className="aa-video-preview">
         {fileUrl ? (
           <video src={fileUrl} poster={selectedVideo.thumbnail_url} controls playsInline />

apps/web/src/pages/ai-avatar/components/PanelVoiceSelector.tsx

@@ -6,6 +6,7 @@ import { useEffect, useRef, useState } from "react"
 import { message } from "antd"
 import { fetchVoices } from "@/api/voices/voices"
 import { previewTts } from "@/api/tts"
+import { normalizeEmotion } from "../utils/contract"
 import type { UnifiedVoiceItem } from "@/api/voices/types"
 import {
   type VoiceSource,
@@ -92,11 +93,14 @@ export function PanelVoiceSelector({
 
   /** 用指定 URL 真实播放（抽取公共） */
   const playAudioUrl = (voiceId: string, url: string) => {
+    // 临时兼容：后端 /tts/preview 返回 HTTP URL，staging 是 HTTPS，Mixed Content 会阻止加载
+    // OSS 同时支持 HTTP/HTTPS，直接替换协议即可
+    const safeUrl = url.startsWith("http://") ? url.replace("http://", "https://") : url
     if (audioRef.current) {
       audioRef.current.pause()
       audioRef.current = null
     }
-    const audio = new Audio(url)
+    const audio = new Audio(safeUrl)
     audioRef.current = audio
     setPreviewingId(voiceId)
     audio.onended = () => {
@@ -147,7 +151,8 @@ export function PanelVoiceSelector({
         const res = await previewTts({
           text: VOICE_PREVIEW_TEXT,
           voice_id: targetId,
-          speed: 1.0,
+          speed: speed, // 透传用户选择的语速（#1822）
+          emotion: normalizeEmotion(emotion), // 情绪中文→英文枚举
         })
         console.log("[AI数字人-克隆试听] previewTts 响应:", {
           audio_url: res.audio_url?.substring(0, 80),
@@ -247,7 +252,7 @@ export function PanelVoiceSelector({
                   type="button"
                   className="aa-voice-card__preview"
                   title={previewingId === voice.id ? "停止试听" : "试听"}
-                  disabled={!previewUrl}
+                  disabled={voice.type === "preset" && !previewUrl}
                   onClick={(e) => {
                     e.stopPropagation()
                     handlePreview(voice)

apps/web/src/pages/ai-avatar/components/TitleLibraryModal.tsx

@@ -0,0 +1,97 @@
+/**
+ * AI数字人 — 标题库选择弹窗
+ * 复用智能剪辑的标题库 API，选择标题后填入输入框
+ */
+import React, { useEffect, useState } from "react"
+import { getTitles } from "@/api/titles"
+import type { TitleItem } from "@/api/titles/types"
+
+interface TitleLibraryModalProps {
+  open: boolean
+  onClose: () => void
+  onSelect: (title: string) => void
+}
+
+const TitleLibraryModal: React.FC<TitleLibraryModalProps> = ({ open, onClose, onSelect }) => {
+  const [titles, setTitles] = useState<TitleItem[]>([])
+  const [loading, setLoading] = useState(false)
+  const [search, setSearch] = useState("")
+
+  useEffect(() => {
+    if (!open) return
+    setLoading(true)
+    getTitles()
+      .then((items) => setTitles(items))
+      .catch(() => setTitles([]))
+      .finally(() => setLoading(false))
+  }, [open])
+
+  const filtered = titles.filter(
+    (t) => !search || t.content.toLowerCase().includes(search.toLowerCase()),
+  )
+
+  if (!open) return null
+
+  return (
+    <div className="aa-modal-overlay" onClick={onClose}>
+      <div className="aa-modal" onClick={(e) => e.stopPropagation()} style={{ maxWidth: 600 }}>
+        <div className="aa-modal__header">
+          <span className="aa-modal__title">从标题库选择</span>
+          <button className="aa-modal__close" onClick={onClose}></button>
+        </div>
+        <div className="aa-modal__body">
+          <div style={{ marginBottom: 12 }}>
+            <input
+              className="aa-input"
+              placeholder="搜索标题..."
+              value={search}
+              onChange={(e) => setSearch(e.target.value)}
+            />
+          </div>
+          {loading ? (
+            <div style={{ textAlign: "center", padding: 40, color: "#8c8ca1" }}>加载中...</div>
+          ) : filtered.length === 0 ? (
+            <div style={{ textAlign: "center", padding: 40, color: "#8c8ca1" }}>
+              暂无标题，请先在标题库创建
+            </div>
+          ) : (
+            <div style={{ maxHeight: 400, overflowY: "auto" }}>
+              {filtered.map((t) => (
+                <div
+                  key={t.id}
+                  style={{
+                    padding: "12px 16px",
+                    marginBottom: 8,
+                    background: "#f8f8fc",
+                    borderRadius: 8,
+                    cursor: "pointer",
+                    transition: "background 0.2s",
+                  }}
+                  onMouseEnter={(e) => (e.currentTarget.style.background = "#eef0ff")}
+                  onMouseLeave={(e) => (e.currentTarget.style.background = "#f8f8fc")}
+                  onClick={() => {
+                    onSelect(t.content)
+                    onClose()
+                  }}
+                >
+                  <div style={{ fontSize: 14, color: "#1a1a2e", marginBottom: 4 }}>{t.content}</div>
+                  <div style={{ fontSize: 12, color: "#8c8ca1" }}>
+                    {t.word_count ?? t.content.length}字 ·{" "}
+                    {t.created_at ? new Date(t.created_at).toLocaleDateString() : ""}
+                  </div>
+                </div>
+              ))}
+            </div>
+          )}
+        </div>
+        <div className="aa-modal__footer">
+          <button className="aa-btn" onClick={onClose}>
+            取消
+          </button>
+        </div>
+      </div>
+    </div>
+  )
+}
+
+export default TitleLibraryModal

apps/web/src/pages/ai-avatar/hooks/useAiAvatar.ts

@@ -1,5 +1,5 @@
 /**
- * AI数字人 — 页面全局状态管理 hook（v3）
+ * AI数字人 — 页面全局状态管理 hook（v3 + #1845 配音前置）
  */
 import { useState, useCallback } from "react"
 import type { AssetItem } from "@/api/assets"
@@ -13,10 +13,19 @@ import {
   type BRollSegment,
   type AiAvatarTitleConfig,
   type AiAvatarCoverConfig,
+  type TtsPreviewResult,
   DEFAULT_TITLE_CONFIG,
   DEFAULT_COVER_CONFIG,
 } from "../types"
 
+const DEFAULT_TTS_PREVIEW: TtsPreviewResult = {
+  audioUrl: null,
+  duration: 0,
+  sentenceTimings: [],
+  status: "idle",
+  error: null,
+}
+
 export function useAiAvatar() {
   /* ── 面板1：出镜视频 ── */
   const [selectedVideo, setSelectedVideo] = useState<AssetItem | null>(null)
@@ -36,6 +45,9 @@ export function useAiAvatar() {
   const [showScriptModal, setShowScriptModal] = useState(false)
   const [showBRollModal, setShowBRollModal] = useState(false)
 
+  /* ── #1845 TTS 预合成（步骤1「生成配音」） ── */
+  const [ttsPreview, setTtsPreview] = useState<TtsPreviewResult>(DEFAULT_TTS_PREVIEW)
+
   /* ── 面板3.5：B-roll ── */
   const [bRollSegments, setBRollSegments] = useState<BRollSegment[]>([])
 
@@ -81,6 +93,7 @@ export function useAiAvatar() {
     setScript(null)
     setScriptText("")
     setLipsyncJob(null)
+    setTtsPreview(DEFAULT_TTS_PREVIEW)
     setBRollSegments([])
     setTitleConfig(DEFAULT_TITLE_CONFIG)
     setCoverConfig(DEFAULT_COVER_CONFIG)
@@ -118,6 +131,10 @@ export function useAiAvatar() {
     showBRollModal,
     setShowBRollModal,
     selectScript,
+    // #1845 TTS 预合成
+    ttsPreview,
+    setTtsPreview,
+    resetTtsPreview: useCallback(() => setTtsPreview(DEFAULT_TTS_PREVIEW), []),
     // B-roll
     bRollSegments,
     addBRollSegment,

apps/web/src/pages/ai-avatar/types.ts

@@ -28,6 +28,17 @@ export const VOICE_LANGUAGE_OPTIONS: { value: VoiceLanguage; label: string }[] =
 /* ── 对口型任务状态 ── */
 export type LipsyncStatus = "idle" | "pending" | "processing" | "completed" | "failed"
 
+/* ── TTS 预合成（#1845 配音前置：步骤1「生成配音」状态） ── */
+export type TtsPreviewStatus = "idle" | "generating" | "done" | "failed"
+
+export interface TtsPreviewResult {
+  audioUrl: string | null
+  duration: number
+  sentenceTimings: SentenceTiming[]
+  status: TtsPreviewStatus
+  error: string | null
+}
+
 /* ── 文案 ── */
 export interface Script {
   id: string
@@ -44,12 +55,23 @@ export interface LipsyncJob {
   status: LipsyncStatus
   progress: number
   output_video_url: string | null
-  /** 对口型成片总时长（秒），后端返回；用于 B-roll 时间自动估算（#1809 ⑥） */
+  /** 对口型成片总时长（秒），后端返回 */
+  script_text: string
   output_duration?: number
+  /** 精确句子时间戳（后端基于 TTS 音频静音检测计算） */
+  sentence_timings?: SentenceTiming[] | null
   error_message: string | null
   created_at: string
 }
 
+/* ── 句子时间戳（后端精确计算） ── */
+export interface SentenceTiming {
+  index: number
+  text: string
+  start_time: number
+  end_time: number
+}
+
 /* ── B-roll 画面插入 ── */
 export type BRollInsertMode = "fullscreen" | "pip"
 export type PipPosition = "top-left" | "top-right" | "bottom-left" | "bottom-right"
@@ -77,6 +99,9 @@ export interface AiAvatarTitleConfig {
   shadow: boolean
   color: string
   auto_subtitle: boolean
+  /** 自定义位置坐标（position=custom 时生效，百分比 0-100） */
+  pos_x?: number
+  pos_y?: number
 }
 
 /* ── 封面配置 ── */
@@ -86,6 +111,8 @@ export interface AiAvatarCoverConfig {
   frame_time: number
   upload_url: string | null
   thumbnail_url: string | null
+  /** 智能封面（MediaKit 选帧）返回的 OSS 非临时 URL（#1822） */
+  smart_cover_url: string | null
 }
 
 /* ── 渲染任务 ── */
@@ -96,6 +123,7 @@ export interface RenderJob {
   status: RenderStatus
   progress: number
   output_video_url: string | null
+  output_cover_url: string | null
   error_message: string | null
   created_at: string
 }
@@ -103,15 +131,17 @@ export interface RenderJob {
 /* ── 默认值 ── */
 export const DEFAULT_TITLE_CONFIG: AiAvatarTitleConfig = {
   title: "",
-  position: "top",
+  position: "bottom",
   font: "思源黑体",
-  size: 28,
+  size: 48,
   bold: true,
   italic: false,
   stroke: false,
   shadow: false,
   color: "#ffffff",
   auto_subtitle: true,
+  pos_x: undefined,
+  pos_y: undefined,
 }
 
 export const DEFAULT_COVER_CONFIG: AiAvatarCoverConfig = {
@@ -120,4 +150,5 @@ export const DEFAULT_COVER_CONFIG: AiAvatarCoverConfig = {
   frame_time: 0,
   upload_url: null,
   thumbnail_url: null,
+  smart_cover_url: null,
 }

apps/web/src/pages/ai-avatar/utils/contract.ts

@@ -0,0 +1,88 @@
+/**
+ * AI数字人 — 前后端接口契约转换工具（#1822）
+ *
+ * 以 packages/domain/video_filter_builder.py 的 build_title_drawtext_filter() 为唯一口径
+ * （契约文档第 5 节的 titles[]/fontSize/frame/start/end 为误写，后端不认，禁止使用）。
+ */
+import type { AiAvatarTitleConfig, AiAvatarCoverConfig, VoiceEmotion } from "../types"
+
+/* ── 情绪：中文 → 英文（防御性映射；state 默认已是英文） ── */
+const EMOTION_ZH_TO_EN: Record<string, VoiceEmotion> = {
+  自然: "natural",
+  兴奋: "excited",
+  沉稳: "calm",
+  亲切: "friendly",
+}
+const VALID_EMOTIONS: VoiceEmotion[] = ["natural", "excited", "calm", "friendly"]
+
+/** 归一化为后端英文枚举 natural/excited/calm/friendly；非法/空值回退 natural。 */
+export function normalizeEmotion(raw: string | undefined | null): VoiceEmotion {
+  if (!raw) return "natural"
+  const v = raw.trim()
+  if ((VALID_EMOTIONS as string[]).includes(v)) return v as VoiceEmotion
+  return EMOTION_ZH_TO_EN[v] ?? "natural"
+}
+
+/* ── 标题：前端 state → 后端 build_title_drawtext_filter 字段（单个 title_config dict） ── */
+/**
+ * 后端真实字段：text(或content)、font(或font_preset)、font_size(或size)、
+ * font_color(或color，可传 #RRGGBB)、position(top/center/bottom/custom)、
+ * enabled、bold、stroke{enabled,width,color}、shadow{enabled,color,offset_x,offset_y}、
+ * pos_x/pos_y(custom 时)、title_image_dataurl（前端 Canvas 渲染的 PNG dataURL，WYSIWYG 路径优先）。
+ * 口播标题默认 position=bottom（不传后端会默认 top 跑到画面顶部）。
+ */
+export function buildTitleConfigPayload(
+  cfg: AiAvatarTitleConfig,
+  titleImageDataUrl?: string | null,
+): Record<string, unknown> {
+  const text = (cfg.title || "").trim()
+  if (!text) return {}
+  const position = cfg.position || "bottom"
+  const payload: Record<string, unknown> = {
+    text,
+    enabled: true,
+    font: cfg.font || "思源黑体",
+    font_size: Math.round(cfg.size) || 48,
+    font_color: cfg.color || "#ffffff",
+    position,
+    bold: !!cfg.bold,
+    stroke: cfg.stroke ? { enabled: true, width: 2, color: "#000000" } : { enabled: false },
+    shadow: cfg.shadow
+      ? { enabled: true, color: "#000000", offset_x: 2, offset_y: 2 }
+      : { enabled: false },
+  }
+  // 自定义坐标（custom 位置）
+  if (position === "custom" && typeof cfg.pos_x === "number" && typeof cfg.pos_y === "number") {
+    payload.pos_x = cfg.pos_x
+    payload.pos_y = cfg.pos_y
+  }
+  // 前端 Canvas 渲染好的 PNG dataURL（所见即所得，后端优先 overlay 此图片图层）
+  if (titleImageDataUrl) {
+    payload.title_image_dataurl = titleImageDataUrl
+  }
+  return payload
+}
+
+/* ── 封面：前端 state → 后端 render cover_config ── */
+export function buildCoverConfigPayload(
+  cfg: AiAvatarCoverConfig,
+  smartCoverUrl: string | null,
+): Record<string, unknown> {
+  const payload: Record<string, unknown> = {
+    enabled: !!cfg.enabled,
+    mode: cfg.mode,
+    // build_cover_extract_command 读取 timestamp（截帧秒数）
+    timestamp: cfg.frame_time || 0,
+  }
+  // 智能封面 URL（后端字段名为 url/imageUrl/cover_url 都兼容，优先 url）
+  if (smartCoverUrl) {
+    payload.url = smartCoverUrl
+    payload.cover_url = smartCoverUrl
+  }
+  // 自定义上传：blob: 本地预览地址无法给后端，仅 OSS URL 可用
+  if (cfg.mode === "upload" && cfg.upload_url && !cfg.upload_url.startsWith("blob:")) {
+    payload.url = cfg.upload_url
+    payload.upload_url = cfg.upload_url
+  }
+  return payload
+}

apps/web/src/pages/ai-avatar/utils/sentences.ts

@@ -1,5 +1,10 @@
 /**
- * AI数字人 — 文案分句 & B-roll 时间自动估算（#1809 ⑤⑥）
+ * AI数字人 — 文案分句 & B-roll 时间计算
+ *
+ * 数据来源优先级：
+ *   1. 后端 sentence_timings（基于 TTS 音频静音检测，精确到句子边界）—— 直接使用，不重新分句
+ *   2. 后端 output_duration（最终渲染视频时长） + 本地分句 —— 按字数比例估算
+ *   3. 两者都没有（对口型还在生成中）—— 返回分句文本但 startTime/endTime 全部 0，等数据到位重算
  */
 
 export interface ScriptSentence {
@@ -11,25 +16,67 @@ export interface ScriptSentence {
   charCount: number
   /** 累计起始字数（用于时间估算） */
   startChar: number
-  /** 估算的对口型视频内起始时间（秒） */
+  /** 对口型视频内起始时间（秒）——后端精确值或前端估算 */
   startTime: number
-  /** 估算的对口型视频内结束时间（秒） */
+  /** 对口型视频内结束时间（秒）——后端精确值或前端估算 */
   endTime: number
 }
 
+/** 句子分隔符：中英文句号/问号/感叹号/分号/逗号/换行（覆盖中文短视频常用断句） */
+const SENTENCE_SPLIT_RE = /[。！？!?？！;；,，\n\r]+/
+
 /**
- * 按句号/问号/感叹号/分号/换行分句（兼容中英文标点）。
- * 空文案返回空数组。时间按「该句字数 ÷ 全文总字数 × 口播总时长」线性估算。
+ * 分句并计算每句的起止时间。
+ *
+ * @param sentenceTimings 后端返回的精确句子时间戳（来自 lipsync_job.sentence_timings）。
+ *   非空时直接按后端返回的句子列表渲染，不再本地分句（避免前后端分句不一致导致时间错位）。
+ * @param outputDuration  最终视频时长（秒）。对口型预览阶段可能为 0，此时降级估算只能给 0。
  */
 export function splitScriptIntoSentences(
   scriptText: string,
-  outputDuration: number,
+  sentenceTimings?:
+    { index?: number; text?: string; start_time: number; end_time: number }[] | null,
+  outputDuration: number = 0,
 ): ScriptSentence[] {
   const text = (scriptText || "").trim()
   if (!text) return []
 
+  // 1. 后端返回了 sentence_timings：校验通过就直接用，跳过本地分句
+  //    校验条件放宽：只要是数组、至少1条、每条 start_time/end_time 是数字即可
+  //    （不再强制要求条数相等——后端静音检测可能按停顿切出更多/更少边界，
+  //    比如文案用逗号连写时本地只分1句、后端按停顿切4句，后端的切法才是对的）
+  if (Array.isArray(sentenceTimings) && sentenceTimings.length > 0) {
+    const valid = sentenceTimings.every(
+      (t) =>
+        t &&
+        typeof t.start_time === "number" &&
+        typeof t.end_time === "number" &&
+        isFinite(t.start_time) &&
+        isFinite(t.end_time) &&
+        t.end_time >= t.start_time,
+    )
+    if (valid) {
+      let accChar = 0
+      return sentenceTimings.map((t, i) => {
+        const sentenceText = (t.text || "").trim() || `句子${i + 1}`
+        const charCount = sentenceText.replace(/\s/g, "").length
+        const sentence: ScriptSentence = {
+          index: typeof t.index === "number" ? t.index : i,
+          text: sentenceText,
+          charCount,
+          startChar: accChar,
+          startTime: round1(t.start_time),
+          endTime: round1(t.end_time),
+        }
+        accChar += charCount
+        return sentence
+      })
+    }
+  }
+
+  // 2. 本地分句 + 按字数比例估算（降级路径）
   const rawParts = text
-    .split(/[。！？!?；;\n\r]+/)
+    .split(SENTENCE_SPLIT_RE)
     .map((part) => part.trim())
     .filter((part) => part.length > 0)
 

apps/web/src/pages/ai-avatar/utils/titleCanvas.ts

@@ -0,0 +1,179 @@
+/**
+ * AI数字人 — 标题 Canvas 渲染工具
+ *
+ * 把标题按前端预览的 HTML/CSS 效果画到透明背景 PNG 上（与视频同分辨率），
+ * 以 dataURL 形式传给后端，后端用 FFmpeg overlay 直接叠加图层，
+ * 彻底解决前端 HTML/CSS 预览 ≠ FFmpeg drawtext 成片的 WYSIWYG 问题。
+ *
+ * 约定：titleConfig.size 的语义是"720p 基准宽度下的字号（px）"，
+ *       按 videoWidth / 720 得到 scale，所有长度类参数乘以 scale，
+ *       保证 1080p / 4K 成片里标题视觉大小与预览一致。
+ */
+import type { AiAvatarTitleConfig } from "../types"
+
+export interface RenderTitlePngOptions {
+  /** 标题配置 */
+  titleConfig: AiAvatarTitleConfig
+  /** 视频宽度（像素），默认 720 */
+  videoWidth?: number
+  /** 视频高度（像素），默认 1280 */
+  videoHeight?: number
+}
+
+/**
+ * 将标题渲染为透明背景 PNG 的 dataURL（data:image/png;base64,...）
+ * Canvas 尺寸与视频一致，保证叠加时 1:1 像素对齐。
+ *
+ * 标题为空时返回 null。
+ */
+export function renderTitleToPngDataUrl(opts: RenderTitlePngOptions): string | null {
+  const { titleConfig, videoWidth = 720, videoHeight = 1280 } = opts
+  if (!titleConfig) return null
+  const rawTitle = (titleConfig.title || "").trim()
+  if (!rawTitle) return null
+
+  // 按 / 或 ／ 分割为多行
+  const lines = rawTitle
+    .split(/[/／]/)
+    .map((l) => l.trim())
+    .filter((l) => l.length > 0)
+  if (lines.length === 0) return null
+
+  // 分辨率缩放系数：基准 720p，所有长度类参数乘以 scale
+  const scale = videoWidth / 720
+  const r = (v: number) => Math.round(v * scale)
+
+  const canvas = document.createElement("canvas")
+  canvas.width = videoWidth
+  canvas.height = videoHeight
+  const ctx = canvas.getContext("2d")
+  if (!ctx) return null
+
+  const baseSize = Math.max(12, Math.round(titleConfig.size || 48))
+  const size = r(baseSize)
+  const bold = !!titleConfig.bold
+  const italic = !!titleConfig.italic
+  const color = titleConfig.color || "#ffffff"
+  const stroke = !!titleConfig.stroke
+  const shadow = !!titleConfig.shadow
+
+  // 字体族 fallback 链：优先中文字体
+  const fontFamily =
+    '"Noto Sans CJK SC","Source Han Sans CN","PingFang SC","Microsoft YaHei",sans-serif'
+  const fontParts: string[] = []
+  if (italic) fontParts.push("italic")
+  if (bold) fontParts.push("bold")
+  fontParts.push(`${size}px`, fontFamily)
+  ctx.font = fontParts.join(" ")
+  ctx.fillStyle = color
+  ctx.textAlign = "center"
+  ctx.textBaseline = "middle"
+
+  // 阴影（shadow=true 时开启）——按 scale 缩放
+  if (shadow) {
+    ctx.shadowColor = "rgba(0,0,0,0.8)"
+    ctx.shadowBlur = r(4)
+    ctx.shadowOffsetX = 0
+    ctx.shadowOffsetY = r(2)
+  }
+
+  // 位置计算：与 PanelLipsyncPreview 的 CSS 对齐（按 scale 缩放 PAD）
+  const PAD = r(16)
+  let centerX = videoWidth / 2
+  const position = titleConfig.position || "bottom"
+  const lineGap = size * 1.2
+  const totalTextH = lines.length * lineGap - (lineGap - size) // 所有行的总高度
+  // 文本块顶部 y（textBaseline=middle 时首行基线）
+  let firstLineY: number
+  if (
+    position === "custom" &&
+    typeof titleConfig.pos_x === "number" &&
+    typeof titleConfig.pos_y === "number"
+  ) {
+    centerX = (Math.max(0, Math.min(100, titleConfig.pos_x)) / 100) * videoWidth
+    const centerY = (Math.max(0, Math.min(100, titleConfig.pos_y)) / 100) * videoHeight
+    firstLineY = centerY - totalTextH / 2 + size / 2
+  } else if (position === "top") {
+    // 顶部：y = size/2 + PAD
+    firstLineY = size / 2 + PAD
+  } else if (position === "center") {
+    firstLineY = videoHeight / 2 - totalTextH / 2 + size / 2
+  } else {
+    // bottom（默认）
+    firstLineY = videoHeight - totalTextH - PAD + size / 2
+  }
+
+  // 描边参数：描边 lineWidth 按 scale 缩放（基准 size * 0.06，最小 2px @720p）
+  const doStroke = stroke
+  const strokeWidth = Math.max(r(2), Math.round(size * 0.06))
+  // 逐行绘制
+  lines.forEach((line, idx) => {
+    const y = firstLineY + idx * lineGap
+    if (doStroke) {
+      const prevShadowColor = ctx.shadowColor
+      const prevShadowBlur = ctx.shadowBlur
+      // 描边不要带阴影（避免黑色描边发虚）
+      ctx.shadowColor = "rgba(0,0,0,0)"
+      ctx.shadowBlur = 0
+      ctx.lineWidth = strokeWidth
+      ctx.strokeStyle = "#000000"
+      ctx.lineJoin = "round"
+      ctx.strokeText(line, centerX, y)
+      // 恢复阴影
+      if (shadow) {
+        ctx.shadowColor = "rgba(0,0,0,0.8)"
+        ctx.shadowBlur = r(4)
+      } else {
+        ctx.shadowColor = prevShadowColor
+        ctx.shadowBlur = prevShadowBlur
+      }
+    }
+    ctx.fillText(line, centerX, y)
+  })
+
+  try {
+    return canvas.toDataURL("image/png")
+  } catch {
+    return null
+  }
+}
+
+/**
+ * 获取视频真实分辨率（HTMLVideoElement + loadedmetadata，超时 3 秒兜底 720×1280）。
+ */
+export function getVideoResolution(
+  videoUrl: string,
+  timeoutMs = 3000,
+): Promise<{ width: number; height: number }> {
+  return new Promise((resolve) => {
+    if (!videoUrl) {
+      resolve({ width: 720, height: 1280 })
+      return
+    }
+    const video = document.createElement("video")
+    video.preload = "metadata"
+    video.muted = true
+    video.playsInline = true
+    video.crossOrigin = "anonymous"
+    let settled = false
+    const done = (w: number, h: number) => {
+      if (settled) return
+      settled = true
+      video.removeAttribute("src")
+      video.load()
+      resolve({ width: w, height: h })
+    }
+    const timer = window.setTimeout(() => done(720, 1280), timeoutMs)
+    video.onloadedmetadata = () => {
+      window.clearTimeout(timer)
+      const w = video.videoWidth || 720
+      const h = video.videoHeight || 1280
+      done(w, h)
+    }
+    video.onerror = () => {
+      window.clearTimeout(timer)
+      done(720, 1280)
+    }
+    video.src = videoUrl
+  })
+}

apps/web/src/pages/assets/hooks/asset-operations/batch/index.ts

@@ -1,4 +0,0 @@
-export { useBatchDelete } from "./useBatchDelete"
-export { useBatchTag } from "./useBatchTag"
-export { useBatchClassify } from "./useBatchClassify"
-export { useBatchMark } from "./useBatchMark"

apps/web/src/pages/assets/hooks/asset-operations/batch/useBatchClassify.ts

@@ -1,58 +0,0 @@
-import { useState, useCallback } from "react"
-import { message } from "antd"
-import { batchClassifyAssets, type BatchOperationResult } from "@/api/assets"
-
-interface UseBatchClassifyOptions {
-  selectedIds: Set<string>
-  queryClient: ReturnType<typeof import("@tanstack/react-query").useQueryClient>
-  showResult: (result: BatchOperationResult, title: string, clear?: boolean) => void
-}
-
-export const useBatchClassify = ({
-  selectedIds,
-  queryClient,
-  showResult,
-}: UseBatchClassifyOptions) => {
-  const [classifyModalOpen, setClassifyModalOpen] = useState(false)
-  const [batchCategory, setBatchCategory] = useState("")
-  const [batchLoading, setBatchLoading] = useState(false)
-
-  const handleBatchClassify = useCallback(async () => {
-    if (!batchCategory) {
-      message.warning("请选择分类")
-      return
-    }
-    const ids = Array.from(selectedIds)
-    setBatchLoading(true)
-    try {
-      const result = await batchClassifyAssets({
-        asset_ids: ids,
-        category: batchCategory,
-      })
-      queryClient.invalidateQueries({ queryKey: ["assets"] })
-      showResult(result, "批量改分类")
-      setClassifyModalOpen(false)
-      setBatchCategory("")
-      if (result.failure_count === 0) {
-        message.success(`成功将 ${result.success_count} 个素材改为「${batchCategory}」`)
-      } else {
-        message.warning(
-          `改分类完成：成功 ${result.success_count} 个，失败 ${result.failure_count} 个`,
-        )
-      }
-    } catch {
-      message.error("批量改分类失败，请重试")
-    } finally {
-      setBatchLoading(false)
-    }
-  }, [batchCategory, selectedIds, queryClient, showResult])
-
-  return {
-    classifyModalOpen,
-    setClassifyModalOpen,
-    batchCategory,
-    setBatchCategory,
-    batchLoading,
-    handleBatchClassify,
-  }
-}

apps/web/src/pages/assets/hooks/asset-operations/batch/useBatchDelete.ts

@@ -1,40 +0,0 @@
-import { useState, useCallback } from "react"
-import { message } from "antd"
-import { batchDeleteAssets, type BatchOperationResult } from "@/api/assets"
-
-interface UseBatchDeleteOptions {
-  selectedIds: Set<string>
-  invalidateAssets: () => void
-  showResult: (result: BatchOperationResult, title: string, clear?: boolean) => void
-}
-
-export const useBatchDelete = ({
-  selectedIds,
-  invalidateAssets,
-  showResult,
-}: UseBatchDeleteOptions) => {
-  const [batchLoading, setBatchLoading] = useState(false)
-
-  const handleBatchDelete = useCallback(async () => {
-    const ids = Array.from(selectedIds)
-    setBatchLoading(true)
-    try {
-      const result = await batchDeleteAssets(ids)
-      invalidateAssets()
-      showResult(result, "批量删除")
-      if (result.failure_count === 0) {
-        message.success(`成功删除 ${result.success_count} 个素材`)
-      } else {
-        message.warning(
-          `删除完成：成功 ${result.success_count} 个，失败 ${result.failure_count} 个`,
-        )
-      }
-    } catch {
-      message.error("批量删除失败，请重试")
-    } finally {
-      setBatchLoading(false)
-    }
-  }, [selectedIds, invalidateAssets, showResult])
-
-  return { batchLoading, handleBatchDelete }
-}

apps/web/src/pages/assets/hooks/asset-operations/batch/useBatchMark.ts

@@ -1,53 +0,0 @@
-import { useState, useCallback } from "react"
-import { message } from "antd"
-import { batchMarkAssets, type BatchOperationResult } from "@/api/assets"
-import type { SmartViewType } from "../../../components/BatchMarkModal"
-import { SMART_VIEW_LABELS } from "../constants"
-
-interface UseBatchMarkOptions {
-  selectedIds: Set<string>
-  queryClient: ReturnType<typeof import("@tanstack/react-query").useQueryClient>
-  showResult: (result: BatchOperationResult, title: string, clear?: boolean) => void
-}
-
-export const useBatchMark = ({ selectedIds, queryClient, showResult }: UseBatchMarkOptions) => {
-  const [markModalOpen, setMarkModalOpen] = useState(false)
-  const [batchSmartView, setBatchSmartView] = useState<SmartViewType>("recommended")
-  const [batchLoading, setBatchLoading] = useState(false)
-
-  const handleBatchMark = useCallback(async () => {
-    const ids = Array.from(selectedIds)
-    setBatchLoading(true)
-    try {
-      const result = await batchMarkAssets({
-        asset_ids: ids,
-        smart_view: batchSmartView,
-      })
-      queryClient.invalidateQueries({ queryKey: ["assets"] })
-      showResult(result, "批量智能标记")
-      setMarkModalOpen(false)
-      if (result.failure_count === 0) {
-        message.success(
-          `成功将 ${result.success_count} 个素材标记为「${SMART_VIEW_LABELS[batchSmartView]}」`,
-        )
-      } else {
-        message.warning(
-          `智能标记完成：成功 ${result.success_count} 个，失败 ${result.failure_count} 个`,
-        )
-      }
-    } catch {
-      message.error("批量智能标记失败，请重试")
-    } finally {
-      setBatchLoading(false)
-    }
-  }, [batchSmartView, selectedIds, queryClient, showResult])
-
-  return {
-    markModalOpen,
-    setMarkModalOpen,
-    batchSmartView,
-    setBatchSmartView,
-    batchLoading,
-    handleBatchMark,
-  }
-}

apps/web/src/pages/assets/hooks/asset-operations/batch/useBatchTag.ts

@@ -1,86 +0,0 @@
-import { useState, useCallback } from "react"
-import { message } from "antd"
-import { batchTagAssets, type BatchOperationResult } from "@/api/assets"
-
-interface UseBatchTagOptions {
-  selectedIds: Set<string>
-  queryClient: ReturnType<typeof import("@tanstack/react-query").useQueryClient>
-  showResult: (result: BatchOperationResult, title: string, clear?: boolean) => void
-}
-
-export const useBatchTag = ({ selectedIds, queryClient, showResult }: UseBatchTagOptions) => {
-  const [tagModalOpen, setTagModalOpen] = useState(false)
-  const [batchTagInput, setBatchTagInput] = useState("")
-  const [batchTags, setBatchTags] = useState<string[]>([])
-  const [tagMode, setTagMode] = useState<"add" | "replace">("add")
-  const [batchLoading, setBatchLoading] = useState(false)
-
-  const handleBatchTag = useCallback(async () => {
-    if (batchTags.length === 0) {
-      message.warning("请至少输入一个标签")
-      return
-    }
-    const ids = Array.from(selectedIds)
-    setBatchLoading(true)
-    try {
-      const result = await batchTagAssets({
-        asset_ids: ids,
-        tags: batchTags,
-        mode: tagMode,
-      })
-      queryClient.invalidateQueries({ queryKey: ["assets"] })
-      showResult(result, "批量打标签")
-      setTagModalOpen(false)
-      setBatchTags([])
-      setBatchTagInput("")
-      setTagMode("add")
-      if (result.failure_count === 0) {
-        message.success(`成功为 ${result.success_count} 个素材打标签`)
-      } else {
-        message.warning(
-          `打标签完成：成功 ${result.success_count} 个，失败 ${result.failure_count} 个`,
-        )
-      }
-    } catch {
-      message.error("批量打标签失败，请重试")
-    } finally {
-      setBatchLoading(false)
-    }
-  }, [batchTags, selectedIds, tagMode, queryClient, showResult])
-
-  const handleTagInputKeyDown = useCallback(
-    (e: React.KeyboardEvent) => {
-      if (e.key === "Enter" && batchTagInput.trim()) {
-        e.preventDefault()
-        const tag = batchTagInput.trim()
-        if (!batchTags.includes(tag)) {
-          setBatchTags([...batchTags, tag])
-        }
-        setBatchTagInput("")
-      }
-    },
-    [batchTagInput, batchTags],
-  )
-
-  const removeBatchTag = useCallback(
-    (tag: string) => {
-      setBatchTags(batchTags.filter((t) => t !== tag))
-    },
-    [batchTags],
-  )
-
-  return {
-    tagModalOpen,
-    setTagModalOpen,
-    batchTagInput,
-    setBatchTagInput,
-    batchTags,
-    setBatchTags,
-    tagMode,
-    setTagMode,
-    batchLoading,
-    handleBatchTag,
-    handleTagInputKeyDown,
-    removeBatchTag,
-  }
-}

apps/web/src/pages/assets/hooks/asset-operations/batchOperations.ts

@@ -1,8 +0,0 @@
-/**
- * @deprecated 请从 ./batch/ 目录导入子模块
- * 保持向后兼容，re-export 所有批量操作 Hook
- */
-export { useBatchDelete } from "./batch/useBatchDelete"
-export { useBatchTag } from "./batch/useBatchTag"
-export { useBatchClassify } from "./batch/useBatchClassify"
-export { useBatchMark } from "./batch/useBatchMark"

apps/web/src/pages/editing-planner/components/pip-config/LayerConfig.tsx

@@ -1,5 +0,0 @@
-/**
- * LayerConfig 入口（向后兼容）
- * 实际实现位于 ./layer-config/ 目录
- */
-export { default } from "./layer-config"

apps/web/src/pages/editing-planner/components/pip-config/LayerList.tsx

@@ -1,65 +0,0 @@
-/**
- * 混剪图层列表
- */
-import React from "react"
-import type { PipLayer } from "@/pages/editing-planner/types"
-import { LAYER_COLORS } from "@/pages/editing-planner/constants/pipConfig"
-
-interface LayerListProps {
-  layers: PipLayer[]
-  selectedId: string
-  onSelect: (id: string) => void
-  onAdd: () => void
-  onDelete: (id: string) => void
-}
-
-const LayerList: React.FC<LayerListProps> = ({ layers, selectedId, onSelect, onAdd, onDelete }) => {
-  return (
-    <div className="pip-layer-list">
-      <div className="pip-toolbar" style={{ marginBottom: 8 }}>
-        <button className="pip-add-btn" onClick={onAdd}>
-          + 添加图层
-        </button>
-      </div>
-      {layers.length === 0 ? (
-        <div className="pip-layer-empty">暂无图层，点击上方添加</div>
-      ) : (
-        layers.map((layer, idx) => (
-          <div
-            key={layer.id}
-            className={`pip-layer-item${selectedId === layer.id ? " active" : ""}`}
-            onClick={() => onSelect(layer.id)}
-          >
-            {layer.thumbnail_url || layer.material_url ? (
-              <img
-                className="pip-layer-thumb"
-                src={layer.thumbnail_url || layer.material_url}
-                alt={layer.name}
-              />
-            ) : (
-              <div
-                className="pip-layer-thumb"
-                style={{
-                  background: LAYER_COLORS[idx % LAYER_COLORS.length],
-                }}
-              />
-            )}
-            <span className="pip-layer-name">{layer.name}</span>
-            <button
-              className="pip-layer-delete"
-              onClick={(e) => {
-                e.stopPropagation()
-                onDelete(layer.id)
-              }}
-              title="删除图层"
-            >
-              ✕
-            </button>
-          </div>
-        ))
-      )}
-    </div>
-  )
-}
-
-export default LayerList

apps/web/src/pages/editing-planner/components/pip-config/layer-config/LayerPositionSize.tsx

@@ -1,172 +0,0 @@
-import React from "react"
-import type { PipLayer, PipGridPosition } from "@/pages/editing-planner/types"
-import { GRID_POSITIONS } from "@/pages/editing-planner/constants/pipConfig"
-
-interface LayerPositionSizeProps {
-  layer: PipLayer
-  onUpdate: (id: string, partial: Partial<PipLayer>) => void
-  onGridClick: (pos: PipGridPosition) => void
-  onWidthChange: (val: number) => void
-  onHeightChange: (val: number) => void
-}
-
-/**
- * 图层位置与尺寸配置面板
- */
-export const LayerPositionSize: React.FC<LayerPositionSizeProps> = ({
-  layer,
-  onUpdate,
-  onGridClick,
-  onWidthChange,
-  onHeightChange,
-}) => (
-  <>
-    {/* 素材类型 */}
-    <div className="pip-field">
-      <label className="pip-field-label">素材类型</label>
-      <div className="pip-type-btns">
-        <button
-          className={`pip-type-btn${layer.material_type === "image" ? " active" : ""}`}
-          onClick={() => onUpdate(layer.id, { material_type: "image" })}
-        >
-          🖼️ 图片
-        </button>
-        <button
-          className={`pip-type-btn${layer.material_type === "video" ? " active" : ""}`}
-          onClick={() => onUpdate(layer.id, { material_type: "video" })}
-        >
-          🎬 视频
-        </button>
-      </div>
-    </div>
-
-    {/* 素材 URL */}
-    <div className="pip-field">
-      <label className="pip-field-label">
-        {layer.material_type === "image" ? "图片" : "视频"} URL
-      </label>
-      <input
-        className="pip-input"
-        type="text"
-        placeholder={
-          layer.material_type === "image"
-            ? "https://example.com/image.png"
-            : "https://example.com/video.mp4"
-        }
-        value={layer.material_url}
-        onChange={(e) => onUpdate(layer.id, { material_url: e.target.value })}
-      />
-    </div>
-
-    {/* 位置：九宫格 + 坐标 */}
-    <div className="pip-field">
-      <label className="pip-field-label">位置</label>
-      <div style={{ display: "flex", gap: 16, alignItems: "flex-start" }}>
-        <div className="pip-grid">
-          {GRID_POSITIONS.map((pos) => (
-            <button
-              key={pos}
-              className={`pip-grid-btn${layer.grid_position === pos ? " active" : ""}`}
-              onClick={() => onGridClick(pos)}
-            >
-              <span className="pip-grid-dot" />
-            </button>
-          ))}
-        </div>
-        <div className="pip-field-row" style={{ flex: 1 }}>
-          <div>
-            <label className="pip-field-label">X (%)</label>
-            <input
-              className="pip-number"
-              type="number"
-              min={0}
-              max={100}
-              value={layer.x}
-              onChange={(e) => onUpdate(layer.id, { x: Number(e.target.value) })}
-            />
-          </div>
-          <div>
-            <label className="pip-field-label">Y (%)</label>
-            <input
-              className="pip-number"
-              type="number"
-              min={0}
-              max={100}
-              value={layer.y}
-              onChange={(e) => onUpdate(layer.id, { y: Number(e.target.value) })}
-            />
-          </div>
-        </div>
-      </div>
-    </div>
-
-    {/* 尺寸 */}
-    <div className="pip-field">
-      <label className="pip-field-label">尺寸</label>
-      <div className="pip-slider-row">
-        <span style={{ fontSize: 12, color: "#999", width: 20 }}>宽</span>
-        <input
-          className="pip-slider"
-          type="range"
-          min={10}
-          max={80}
-          value={layer.width}
-          onChange={(e) => onWidthChange(Number(e.target.value))}
-        />
-        <span className="pip-slider-value">{layer.width}%</span>
-      </div>
-      <div className="pip-slider-row" style={{ marginTop: 6 }}>
-        <span style={{ fontSize: 12, color: "#999", width: 20 }}>高</span>
-        <input
-          className="pip-slider"
-          type="range"
-          min={10}
-          max={80}
-          value={layer.height}
-          onChange={(e) => onHeightChange(Number(e.target.value))}
-        />
-        <span className="pip-slider-value">{layer.height}%</span>
-      </div>
-      <div
-        className="pip-lock-row"
-        style={{ marginTop: 6 }}
-        onClick={() => onUpdate(layer.id, { aspect_lock: !layer.aspect_lock })}
-      >
-        <span className="pip-lock-icon">{layer.aspect_lock ? "🔒" : "🔓"}</span>
-        <span>{layer.aspect_lock ? "已锁定比例" : "锁定宽高比"}</span>
-      </div>
-    </div>
-
-    {/* 圆角 */}
-    <div className="pip-field">
-      <label className="pip-field-label">圆角</label>
-      <div className="pip-slider-row">
-        <input
-          className="pip-slider"
-          type="range"
-          min={0}
-          max={50}
-          value={layer.border_radius}
-          onChange={(e) => onUpdate(layer.id, { border_radius: Number(e.target.value) })}
-        />
-        <span className="pip-slider-value">{layer.border_radius}%</span>
-      </div>
-    </div>
-
-    {/* 透明度 */}
-    <div className="pip-field">
-      <label className="pip-field-label">透明度</label>
-      <div className="pip-slider-row">
-        <input
-          className="pip-slider"
-          type="range"
-          min={0}
-          max={100}
-          value={layer.opacity}
-          onChange={(e) => onUpdate(layer.id, { opacity: Number(e.target.value) })}
-        />
-        <span className="pip-slider-value">{layer.opacity}%</span>
-      </div>
-    </div>
-  </>
-)

apps/web/src/pages/editing-planner/components/pip-config/layer-config/LayerTimingAnimation.tsx

@@ -1,87 +0,0 @@
-import React from "react"
-import type { PipLayer, PipAnimType, PipSlideDirection } from "@/pages/editing-planner/types"
-import { ANIM_OPTIONS, SLIDE_DIR_OPTIONS } from "@/pages/editing-planner/constants/pipConfig"
-
-interface LayerTimingAnimationProps {
-  layer: PipLayer
-  totalDuration: number
-  onUpdate: (id: string, partial: Partial<PipLayer>) => void
-}
-
-/**
- * 图层时间与动画配置面板
- */
-export const LayerTimingAnimation: React.FC<LayerTimingAnimationProps> = ({
-  layer,
-  totalDuration,
-  onUpdate,
-}) => (
-  <>
-    {/* 时间 */}
-    <div className="pip-field">
-      <label className="pip-field-label">时间</label>
-      <div className="pip-field-row">
-        <div>
-          <label className="pip-field-label">开始 (s)</label>
-          <input
-            className="pip-number"
-            type="number"
-            min={0}
-            max={totalDuration || 999}
-            step={0.1}
-            value={layer.start_time}
-            onChange={(e) => onUpdate(layer.id, { start_time: Number(e.target.value) })}
-          />
-        </div>
-        <div>
-          <label className="pip-field-label">持续 (s)</label>
-          <input
-            className="pip-number"
-            type="number"
-            min={0.1}
-            max={totalDuration || 999}
-            step={0.1}
-            value={layer.duration}
-            onChange={(e) => onUpdate(layer.id, { duration: Number(e.target.value) })}
-          />
-        </div>
-      </div>
-    </div>
-
-    {/* 入场动画 */}
-    <div className="pip-field">
-      <label className="pip-field-label">入场动画</label>
-      <select
-        className="pip-select"
-        value={layer.animation}
-        onChange={(e) => onUpdate(layer.id, { animation: e.target.value as PipAnimType })}
-      >
-        {ANIM_OPTIONS.map((opt) => (
-          <option key={opt.value} value={opt.value}>
-            {opt.label}
-          </option>
-        ))}
-      </select>
-    </div>
-
-    {/* 滑入方向（仅 slide_in 时显示） */}
-    {layer.animation === "slide_in" && (
-      <div className="pip-field">
-        <label className="pip-field-label">滑入方向</label>
-        <select
-          className="pip-select"
-          value={layer.slide_direction}
-          onChange={(e) =>
-            onUpdate(layer.id, { slide_direction: e.target.value as PipSlideDirection })
-          }
-        >
-          {SLIDE_DIR_OPTIONS.map((opt) => (
-            <option key={opt.value} value={opt.value}>
-              {opt.label}
-            </option>
-          ))}
-        </select>
-      </div>
-    )}
-  </>
-)

apps/web/src/pages/editing-planner/components/pip-config/layer-config/PipPreview.tsx

@@ -1,33 +0,0 @@
-import React from "react"
-import type { PipLayer } from "@/pages/editing-planner/types"
-import { LAYER_COLORS } from "@/pages/editing-planner/constants/pipConfig"
-
-interface PipPreviewProps {
-  layers: PipLayer[]
-  selectedId: string
-}
-
-/**
- * PIP 图层迷你预览组件
- */
-export const PipPreview: React.FC<PipPreviewProps> = ({ layers, selectedId }) => (
-  <div className="pip-preview-box">
-    {layers.map((l, idx) => (
-      <div
-        key={l.id}
-        className={`pip-preview-layer${selectedId === l.id ? " selected" : ""}`}
-        style={{
-          left: `${l.x}%`,
-          top: `${l.y}%`,
-          width: `${l.width}%`,
-          height: `${l.height}%`,
-          background: LAYER_COLORS[idx % LAYER_COLORS.length],
-          opacity: l.opacity / 100,
-          borderRadius: `${l.border_radius}%`,
-        }}
-      >
-        <span className="pip-preview-label">{l.name}</span>
-      </div>
-    ))}
-  </div>
-)

apps/web/src/pages/editing-planner/components/pip-config/layer-config/index.tsx

@@ -1,57 +0,0 @@
-/**
- * 混剪单图层配置区
- */
-import React from "react"
-import type { PipLayer, PipGridPosition } from "@/pages/editing-planner/types"
-import { PipPreview } from "./PipPreview"
-import { LayerPositionSize } from "./LayerPositionSize"
-import { LayerTimingAnimation } from "./LayerTimingAnimation"
-
-interface LayerConfigProps {
-  layer: PipLayer | null
-  layers: PipLayer[]
-  totalDuration: number
-  onUpdate: (id: string, partial: Partial<PipLayer>) => void
-  onGridClick: (pos: PipGridPosition) => void
-  onWidthChange: (val: number) => void
-  onHeightChange: (val: number) => void
-}
-
-const LayerConfig: React.FC<LayerConfigProps> = ({
-  layer,
-  layers,
-  totalDuration,
-  onUpdate,
-  onGridClick,
-  onWidthChange,
-  onHeightChange,
-}) => {
-  if (!layer) {
-    return (
-      <div className="pip-config-area">
-        <div className="pip-config-empty">选择或添加图层以配置</div>
-      </div>
-    )
-  }
-
-  return (
-    <div className="pip-config-area">
-      {/* 迷你预览 */}
-      <PipPreview layers={layers} selectedId={layer.id} />
-
-      {/* 位置与尺寸 */}
-      <LayerPositionSize
-        layer={layer}
-        onUpdate={onUpdate}
-        onGridClick={onGridClick}
-        onWidthChange={onWidthChange}
-        onHeightChange={onHeightChange}
-      />
-
-      {/* 时间与动画 */}
-      <LayerTimingAnimation layer={layer} totalDuration={totalDuration} onUpdate={onUpdate} />
-    </div>
-  )
-}
-
-export default LayerConfig

apps/web/src/pages/editing-planner/components/sticker/StickerLibrary.tsx

@@ -1,142 +0,0 @@
-/**
- * 贴纸素材库（emoji / 图片 / 文字花字）
- */
-import React, { useState } from "react"
-import type { StickerType, TextStickerPreset } from "@/pages/editing-planner/types"
-import {
-  EMOJI_LIST,
-  STICKER_TYPE_TABS,
-  TEXT_PRESET_STYLES,
-  TEXT_STICKER_PRESET_LABELS,
-} from "@/pages/editing-planner/constants/sticker"
-
-interface StickerLibraryProps {
-  activeTab: StickerType
-  onTabChange: (tab: StickerType) => void
-  onAddSticker: (type: StickerType, content: string) => void
-}
-
-const StickerLibrary: React.FC<StickerLibraryProps> = ({
-  activeTab,
-  onTabChange,
-  onAddSticker,
-}) => {
-  const [textInput, setTextInput] = useState("")
-  const imageInputRef = React.useRef<HTMLInputElement>(null)
-
-  const handleAddImage = () => {
-    const val = imageInputRef.current?.value.trim()
-    if (val) {
-      onAddSticker("image", val)
-      if (imageInputRef.current) imageInputRef.current.value = ""
-    }
-  }
-
-  const handleAddText = () => {
-    if (textInput.trim()) {
-      onAddSticker("text", textInput.trim())
-      setTextInput("")
-    }
-  }
-
-  return (
-    <>
-      {/* 类型 Tab */}
-      <div className="sticker-tabs">
-        {STICKER_TYPE_TABS.map((t) => (
-          <button
-            key={t.value}
-            className={`sticker-tab${activeTab === t.value ? " active" : ""}`}
-            onClick={() => onTabChange(t.value)}
-          >
-            {t.label}
-          </button>
-        ))}
-      </div>
-
-      {/* Tab 内容区 */}
-      <div className="sticker-tab-content">
-        {/* Emoji 素材库 */}
-        {activeTab === "emoji" && (
-          <div className="sticker-emoji-grid">
-            {EMOJI_LIST.map((emoji) => (
-              <button
-                key={emoji}
-                className="sticker-emoji-btn"
-                onClick={() => onAddSticker("emoji", emoji)}
-              >
-                {emoji}
-              </button>
-            ))}
-          </div>
-        )}
-
-        {/* 图片贴纸 */}
-        {activeTab === "image" && (
-          <div className="sticker-image-input">
-            <input
-              ref={imageInputRef}
-              type="text"
-              className="sticker-url-input"
-              placeholder="输入图片 URL 添加贴纸..."
-              onKeyDown={(e) => {
-                if (e.key === "Enter" && e.currentTarget.value.trim()) {
-                  handleAddImage()
-                }
-              }}
-            />
-            <button className="sticker-url-add-btn" onClick={handleAddImage}>
-              添加
-            </button>
-          </div>
-        )}
-
-        {/* 文字花字 */}
-        {activeTab === "text" && (
-          <div className="sticker-text-section">
-            <div className="sticker-text-input-row">
-              <input
-                type="text"
-                className="sticker-text-input"
-                placeholder="输入文字内容..."
-                value={textInput}
-                onChange={(e) => setTextInput(e.target.value)}
-              />
-              <button
-                className="sticker-text-add-btn"
-                disabled={!textInput.trim()}
-                onClick={handleAddText}
-              >
-                添加
-              </button>
-            </div>
-            <div className="sticker-text-presets">
-              <div className="sticker-preset-title">花字预设预览</div>
-              <div className="sticker-preset-grid">
-                {(Object.keys(TEXT_STICKER_PRESET_LABELS) as TextStickerPreset[]).map((p) => (
-                  <div
-                    key={p}
-                    className="sticker-preset-preview"
-                    style={{
-                      background:
-                        p === "bubble"
-                          ? "rgba(0,0,0,0.5)"
-                          : p === "gradient"
-                            ? "linear-gradient(90deg,#f093fb,#f5576c)"
-                            : "#1a1a2e",
-                    }}
-                  >
-                    <span style={TEXT_PRESET_STYLES[p]}>示例</span>
-                    <div className="sticker-preset-name">{TEXT_STICKER_PRESET_LABELS[p]}</div>
-                  </div>
-                ))}
-              </div>
-            </div>
-          </div>
-        )}
-      </div>
-    </>
-  )
-}
-
-export default StickerLibrary

apps/web/src/pages/editing-planner/components/sticker/StickerList.tsx

@@ -1,54 +0,0 @@
-/**
- * 已添加贴纸列表
- */
-import React from "react"
-import type { StickerItem } from "@/pages/editing-planner/types"
-
-interface StickerListProps {
-  items: StickerItem[]
-  selectedId: string | null
-  onSelect: (id: string) => void
-  onDelete: (id: string) => void
-}
-
-const StickerList: React.FC<StickerListProps> = ({ items, selectedId, onSelect, onDelete }) => {
-  if (items.length === 0) return null
-
-  const getDisplayContent = (item: StickerItem) => {
-    if (item.type === "emoji") return { icon: item.content, name: "表情贴纸" }
-    if (item.type === "text") return { icon: "T", name: item.content.slice(0, 10) }
-    return { icon: "🖼", name: "图片贴纸" }
-  }
-
-  return (
-    <div className="sticker-list-section">
-      <div className="sticker-section-title">已添加贴纸 ({items.length})</div>
-      <div className="sticker-list">
-        {items.map((item) => {
-          const { icon, name } = getDisplayContent(item)
-          return (
-            <div
-              key={item.id}
-              className={`sticker-list-item${selectedId === item.id ? " active" : ""}`}
-              onClick={() => onSelect(item.id)}
-            >
-              <span className="sticker-list-icon">{icon}</span>
-              <span className="sticker-list-name">{name}</span>
-              <button
-                className="sticker-list-delete"
-                onClick={(e) => {
-                  e.stopPropagation()
-                  onDelete(item.id)
-                }}
-              >
-                ✕
-              </button>
-            </div>
-          )
-        })}
-      </div>
-    </div>
-  )
-}
-
-export default StickerList

apps/web/src/pages/editing-planner/components/sticker/StickerPreview.tsx

@@ -1,39 +0,0 @@
-import React from "react"
-import type { StickerItem } from "@/pages/editing-planner/types"
-import { TEXT_PRESET_STYLES } from "@/pages/editing-planner/constants/sticker"
-
-interface StickerPreviewProps {
-  sticker: StickerItem
-}
-
-export const StickerPreview: React.FC<StickerPreviewProps> = ({ sticker }) => (
-  <div className="sticker-preview-box">
-    <div
-      className="sticker-preview-item"
-      style={{
-        left: `${sticker.x}%`,
-        top: `${sticker.y}%`,
-        width: `${sticker.width}%`,
-        height: `${sticker.height}%`,
-        transform: `translate(-50%, -50%) rotate(${sticker.rotation}deg)`,
-        opacity: sticker.opacity / 100,
-        fontSize: sticker.type === "text" ? `${sticker.font_size}px` : undefined,
-        ...TEXT_PRESET_STYLES[sticker.text_preset],
-      }}
-    >
-      {sticker.type === "emoji" && sticker.content}
-      {sticker.type === "text" && sticker.content}
-      {sticker.type === "image" && (
-        <img
-          src={sticker.content}
-          alt="sticker"
-          style={{
-            width: "100%",
-            height: "100%",
-            objectFit: "contain",
-          }}
-        />
-      )}
-    </div>
-  </div>
-)

apps/web/src/pages/editing-planner/components/sticker/StickerPropsEditor.tsx

@@ -1,130 +0,0 @@
-/**
- * 选中贴纸的属性编辑器
- */
-import React from "react"
-import type { StickerItem } from "@/pages/editing-planner/types"
-import { StickerPreview } from "./StickerPreview"
-import { TextStickerPropsEditor } from "./TextStickerPropsEditor"
-
-interface StickerPropsEditorProps {
-  sticker: StickerItem
-  totalDuration: number
-  onUpdate: (id: string, partial: Partial<StickerItem>) => void
-}
-
-const StickerPropsEditor: React.FC<StickerPropsEditorProps> = ({
-  sticker,
-  totalDuration,
-  onUpdate,
-}) => {
-  return (
-    <div className="sticker-props-section">
-      <div className="sticker-section-title">属性调整</div>
-
-      {/* 位置 */}
-      <div className="sticker-prop-row">
-        <span className="sticker-prop-label">位置 X</span>
-        <input
-          type="range"
-          className="sticker-prop-slider"
-          min={0}
-          max={100}
-          value={sticker.x}
-          onChange={(e) => onUpdate(sticker.id, { x: Number(e.target.value) })}
-        />
-        <span className="sticker-prop-value">{sticker.x}%</span>
-      </div>
-      <div className="sticker-prop-row">
-        <span className="sticker-prop-label">位置 Y</span>
-        <input
-          type="range"
-          className="sticker-prop-slider"
-          min={0}
-          max={100}
-          value={sticker.y}
-          onChange={(e) => onUpdate(sticker.id, { y: Number(e.target.value) })}
-        />
-        <span className="sticker-prop-value">{sticker.y}%</span>
-      </div>
-
-      {/* 尺寸 */}
-      <div className="sticker-prop-row">
-        <span className="sticker-prop-label">大小</span>
-        <input
-          type="range"
-          className="sticker-prop-slider"
-          min={5}
-          max={50}
-          value={sticker.width}
-          onChange={(e) =>
-            onUpdate(sticker.id, {
-              width: Number(e.target.value),
-              height: Number(e.target.value),
-            })
-          }
-        />
-        <span className="sticker-prop-value">{sticker.width}%</span>
-      </div>
-
-      {/* 旋转 */}
-      <div className="sticker-prop-row">
-        <span className="sticker-prop-label">旋转</span>
-        <input
-          type="range"
-          className="sticker-prop-slider"
-          min={-180}
-          max={180}
-          value={sticker.rotation}
-          onChange={(e) => onUpdate(sticker.id, { rotation: Number(e.target.value) })}
-        />
-        <span className="sticker-prop-value">{sticker.rotation}°</span>
-      </div>
-
-      {/* 透明度 */}
-      <div className="sticker-prop-row">
-        <span className="sticker-prop-label">透明度</span>
-        <input
-          type="range"
-          className="sticker-prop-slider"
-          min={0}
-          max={100}
-          value={sticker.opacity}
-          onChange={(e) => onUpdate(sticker.id, { opacity: Number(e.target.value) })}
-        />
-        <span className="sticker-prop-value">{sticker.opacity}%</span>
-      </div>
-
-      {/* 时间 */}
-      <div className="sticker-prop-row">
-        <span className="sticker-prop-label">开始</span>
-        <input
-          type="number"
-          className="sticker-prop-number"
-          min={0}
-          max={totalDuration}
-          step={0.1}
-          value={sticker.start_time}
-          onChange={(e) => onUpdate(sticker.id, { start_time: Number(e.target.value) })}
-        />
-        <span className="sticker-prop-label">时长</span>
-        <input
-          type="number"
-          className="sticker-prop-number"
-          min={0}
-          max={totalDuration}
-          step={0.1}
-          value={sticker.duration}
-          onChange={(e) => onUpdate(sticker.id, { duration: Number(e.target.value) })}
-        />
-      </div>
-
-      {/* 文字贴纸特有属性 */}
-      <TextStickerPropsEditor sticker={sticker} onUpdate={onUpdate} />
-
-      {/* 预览 */}
-      <StickerPreview sticker={sticker} />
-    </div>
-  )
-}
-
-export default StickerPropsEditor

apps/web/src/pages/editing-planner/components/sticker/TextStickerPropsEditor.tsx

@@ -1,57 +0,0 @@
-import React from "react"
-import type { StickerItem, TextStickerPreset } from "@/pages/editing-planner/types"
-import { TEXT_STICKER_PRESET_LABELS } from "@/pages/editing-planner/constants/sticker"
-
-interface TextStickerPropsEditorProps {
-  sticker: StickerItem
-  onUpdate: (id: string, partial: Partial<StickerItem>) => void
-}
-
-export const TextStickerPropsEditor: React.FC<TextStickerPropsEditorProps> = ({
-  sticker,
-  onUpdate,
-}) => {
-  if (sticker.type !== "text") return null
-
-  return (
-    <>
-      <div className="sticker-prop-row">
-        <span className="sticker-prop-label">花字</span>
-        <select
-          className="sticker-prop-select"
-          value={sticker.text_preset}
-          onChange={(e) =>
-            onUpdate(sticker.id, { text_preset: e.target.value as TextStickerPreset })
-          }
-        >
-          {(Object.keys(TEXT_STICKER_PRESET_LABELS) as TextStickerPreset[]).map((p) => (
-            <option key={p} value={p}>
-              {TEXT_STICKER_PRESET_LABELS[p]}
-            </option>
-          ))}
-        </select>
-      </div>
-      <div className="sticker-prop-row">
-        <span className="sticker-prop-label">字号</span>
-        <input
-          type="range"
-          className="sticker-prop-slider"
-          min={12}
-          max={72}
-          value={sticker.font_size}
-          onChange={(e) => onUpdate(sticker.id, { font_size: Number(e.target.value) })}
-        />
-        <span className="sticker-prop-value">{sticker.font_size}px</span>
-      </div>
-      <div className="sticker-prop-row">
-        <span className="sticker-prop-label">颜色</span>
-        <input
-          type="color"
-          className="sticker-prop-color"
-          value={sticker.text_color}
-          onChange={(e) => onUpdate(sticker.id, { text_color: e.target.value })}
-        />
-      </div>
-    </>
-  )
-}

apps/web/src/pages/editing-planner/components/tts/TtsSlider.tsx

@@ -1,62 +0,0 @@
-/**
- * TTS 滑块组件（语速/语调/音量）
- */
-import React from "react"
-import { Slider } from "antd"
-
-interface TtsSliderProps {
-  label: string
-  value: number
-  min: number
-  max: number
-  step: number
-  unit?: string
-  onChange: (val: number) => void
-  marks?: string[]
-  tooltipFormatter?: (v: number) => string
-}
-
-const TtsSlider: React.FC<TtsSliderProps> = ({
-  label,
-  value,
-  min,
-  max,
-  step,
-  unit = "",
-  onChange,
-  marks,
-  tooltipFormatter,
-}) => {
-  const displayValue =
-    unit === "x"
-      ? `${value.toFixed(2)}x`
-      : label === "语调"
-        ? `${value > 0 ? "+" : ""}${value} 半音`
-        : `${value}${unit}`
-
-  return (
-    <div className="tts-slider-section">
-      <div className="tts-slider-header">
-        <span className="tts-slider-label">{label}</span>
-        <span className="tts-slider-value">{displayValue}</span>
-      </div>
-      <Slider
-        min={min}
-        max={max}
-        step={step}
-        value={value}
-        onChange={(v) => onChange(v as number)}
-        tooltip={tooltipFormatter ? { formatter: (v) => tooltipFormatter(v as number) } : undefined}
-      />
-      {marks && (
-        <div className="tts-slider-marks">
-          {marks.map((m, i) => (
-            <span key={i}>{m}</span>
-          ))}
-        </div>
-      )}
-    </div>
-  )
-}
-
-export default TtsSlider

apps/web/src/pages/editing-planner/components/tts/VoiceSelector.tsx

@@ -1,50 +0,0 @@
-/**
- * TTS 音色选择组件
- */
-import React from "react"
-import type { TTSVoice } from "@/api/tts"
-import { VOICE_CATEGORY_MAP } from "../../constants/tts"
-
-interface VoiceSelectorProps {
-  voices: TTSVoice[]
-  voicesLoading: boolean
-  selectedVoiceId: string
-  onVoiceSelect: (voiceId: string) => void
-}
-
-const VoiceSelector: React.FC<VoiceSelectorProps> = ({
-  voices,
-  voicesLoading,
-  selectedVoiceId,
-  onVoiceSelect,
-}) => {
-  const voiceCategories = Object.entries(VOICE_CATEGORY_MAP)
-
-  return (
-    <div className="tts-voice-section">
-      <div className="tts-voice-label">
-        选择音色
-        {voicesLoading && <span className="tts-voice-loading">加载中...</span>}
-      </div>
-      <div className="tts-voice-grid">
-        {voiceCategories.map(([cat, info]) => {
-          const voice = voices.find((v) => v.category === cat)
-          const isSelected = voice && selectedVoiceId === voice.id
-          return (
-            <button
-              key={cat}
-              className={`tts-voice-card${isSelected ? " active" : ""}`}
-              onClick={() => voice && onVoiceSelect(voice.id)}
-              disabled={!voice || voicesLoading}
-            >
-              <span className="tts-voice-card-icon">{info.icon}</span>
-              <span className="tts-voice-card-name">{voice?.name || info.label}</span>
-            </button>
-          )
-        })}
-      </div>
-    </div>
-  )
-}
-
-export default VoiceSelector

apps/web/src/pages/generate/GeneratePage.tsx

@@ -548,10 +548,21 @@ const GeneratePage: React.FC = () => {
             <div
               style={{
                 display: "flex",
-                justifyContent: "center",
+                flexDirection: "column",
+                alignItems: "center",
                 marginTop: 16,
               }}
             >
+              <h2
+                style={{
+                  textAlign: "center",
+                  marginBottom: 12,
+                  fontSize: "1.5rem",
+                  fontWeight: 600,
+                }}
+              >
+                🎬 确认生成
+              </h2>
               <div
                 style={{
                   display: "flex",

apps/web/src/pages/products/ProductLibrary.tsx

@@ -1,17 +1,22 @@
 /**
  * 成片库页面 — V21 设计系统
- * 卡片网格布局，支持视频内联播放/下载/分享、批量操作、筛选
+ * 卡片网格布局，支持视频内联播放/下载/分享、批量操作、筛选、无限滚动分页
  *
  * 主组件仅保留 Hook 组装与整体布局
- * 列表查询 → hooks/useProductList
+ * 列表查询 → hooks/useProductList（useInfiniteQuery 分页）
  * 操作逻辑 → hooks/useProductActions
  * 筛选栏 → components/ProductFilterBar
  * 批量操作栏 → components/ProductBatchBar
  * 空状态 → components/ProductEmptyState
  * 产品卡片 → components/ProductCard（内联视频播放）
  */
-import React from "react"
-import { VideoCameraOutlined, DownloadOutlined, ReloadOutlined } from "@ant-design/icons"
+import React, { useEffect, useRef } from "react"
+import {
+  VideoCameraOutlined,
+  DownloadOutlined,
+  ReloadOutlined,
+  LoadingOutlined,
+} from "@ant-design/icons"
 import { Button } from "@/components/ui"
 import { ProductCard } from "./components/ProductCard"
 import { ProductFilterBar } from "./components/ProductFilterBar"
@@ -24,11 +29,13 @@ import "./products.css"
 
 const ProductLibrary: React.FC = () => {
   const {
-    products,
     filteredProducts,
     isLoading,
+    isFetchingNextPage,
     isError,
     error,
+    hasNextPage,
+    fetchNextPage,
     refetch,
     searchText,
     setSearchText,
@@ -64,19 +71,40 @@ const ProductLibrary: React.FC = () => {
   } = useProductActions({
     selectedIds,
     clearSelection,
-    products,
+    products: filteredProducts,
     setPlayingProduct: () => {}, // 不再使用弹窗播放
   })
 
   const { recomputeDedup, isRecomputing } = useRecomputeDedup()
 
-  // ── Loading 状态 ──
-  if (isLoading) {
+  /* ── 无限滚动：IntersectionObserver 监听底部哨兵元素 ── */
+  const sentinelRef = useRef<HTMLDivElement>(null)
+
+  useEffect(() => {
+    const el = sentinelRef.current
+    if (!el) return
+    // 已有数据但正在加载中/没有更多页时不触发
+    if (isFetchingNextPage || !hasNextPage) return
+
+    const observer = new IntersectionObserver(
+      (entries) => {
+        if (entries[0]?.isIntersecting) {
+          void fetchNextPage()
+        }
+      },
+      { rootMargin: "200px" },
+    )
+    observer.observe(el)
+    return () => observer.disconnect()
+  }, [fetchNextPage, hasNextPage, isFetchingNextPage])
+
+  // ── Loading 状态（仅首次加载）──
+  if (isLoading && filteredProducts.length === 0) {
     return <ProductEmptyState type="loading" />
   }
 
   // ── Error 状态 ──
-  if (isError) {
+  if (isError && filteredProducts.length === 0) {
     console.error("[ProductLibrary] 加载失败:", error)
     const errorMsg = error?.message || "加载失败"
     const is404 = errorMsg.includes("404") || errorMsg.includes("Not Found")
@@ -143,22 +171,46 @@ const ProductLibrary: React.FC = () => {
 
       {/* 卡片网格 */}
       {filteredProducts.length > 0 ? (
-        <div className="xx-products-grid">
-          {filteredProducts.map((product) => (
-            <ProductCard
-              key={product.id}
-              product={product}
-              isSelected={selectedIds.has(product.id)}
-              batchMode={batchMode}
-              onToggleSelect={handleToggleSelect}
-              onDownload={handleDownload}
-              onShare={handleShare}
-              onDelete={handleDelete}
-              onPublish={handlePublish}
-              onReviewStatusChange={handleReviewStatusChange}
-            />
-          ))}
-        </div>
+        <>
+          <div className="xx-products-grid">
+            {filteredProducts.map((product) => (
+              <ProductCard
+                key={product.id}
+                product={product}
+                isSelected={selectedIds.has(product.id)}
+                batchMode={batchMode}
+                onToggleSelect={handleToggleSelect}
+                onDownload={handleDownload}
+                onShare={handleShare}
+                onDelete={handleDelete}
+                onPublish={handlePublish}
+                onReviewStatusChange={handleReviewStatusChange}
+              />
+            ))}
+          </div>
+
+          {/* 底部哨兵 + 状态提示 */}
+          <div
+            ref={sentinelRef}
+            style={{
+              gridColumn: "1 / -1",
+              textAlign: "center",
+              padding: "24px 0",
+              fontSize: 13,
+              color: "#8c8ca1",
+            }}
+          >
+            {isFetchingNextPage ? (
+              <>
+                <LoadingOutlined /> 加载中…
+              </>
+            ) : hasNextPage ? (
+              <span style={{ opacity: 0 }}>加载更多</span>
+            ) : (
+              <span>—— 已加载全部 ——</span>
+            )}
+          </div>
+        </>
       ) : (
         <ProductEmptyState type="empty" />
       )}

apps/web/src/pages/products/hooks/useProductList/index.ts

@@ -1,28 +1,53 @@
 import { useMemo } from "react"
-import { useQuery } from "@tanstack/react-query"
+import { useInfiniteQuery } from "@tanstack/react-query"
 import { getProducts, type ProductItem as ApiProductItem } from "@/api/products"
 import { mapApiProduct } from "../../utils"
+import type { ProductItem } from "../../types"
 import { useProductFiltering } from "./useProductFiltering"
 import { useBatchSelection } from "./useBatchSelection"
 
 export type { Filters } from "./useProductFiltering"
 
+const PAGE_SIZE = 20
+
 export const useProductList = () => {
-  /* ── 获取成品列表 ── */
+  /* ── 无限滚动获取成品列表（每页 20 条） ── */
   const {
-    data: apiProducts = [],
+    data,
     isLoading,
+    isFetchingNextPage,
     isError,
     error,
+    hasNextPage,
+    fetchNextPage,
     refetch,
-  } = useQuery<ApiProductItem[], Error>({
+  } = useInfiniteQuery<
+    {
+      items: ApiProductItem[]
+      total: number
+      page: number
+      page_size: number
+    },
+    Error
+  >({
     queryKey: ["products"],
-    queryFn: () => getProducts(),
+    queryFn: async ({ pageParam = 1 }) =>
+      getProducts({ page: pageParam as number, page_size: PAGE_SIZE }),
+    initialPageParam: 1,
+    getNextPageParam: (lastPage) => {
+      const loadedCount = lastPage.page * lastPage.page_size
+      return loadedCount < lastPage.total ? lastPage.page + 1 : undefined
+    },
     staleTime: 30_000,
   })
 
-  // 映射为前端类型，按创建时间倒序排列，防御非数组返回
-  const products = useMemo(
+  // 将所有页拼接为一维数组，再做前端映射+排序
+  const apiProducts = useMemo<ApiProductItem[]>(() => {
+    if (!data?.pages) return []
+    return data.pages.flatMap((p) => p.items)
+  }, [data])
+
+  const products = useMemo<ProductItem[]>(
     () =>
       (Array.isArray(apiProducts) ? apiProducts : []).map(mapApiProduct).sort((a, b) => {
         if (!a.date || a.date === "—") return 1
@@ -65,8 +90,11 @@ export const useProductList = () => {
     products,
     filteredProducts,
     isLoading,
+    isFetchingNextPage,
     isError,
     error,
+    hasNextPage,
+    fetchNextPage,
     refetch,
     // 筛选
     searchText,

apps/web/src/test/pages/assets/smoke.test.tsx

@@ -36,10 +36,10 @@ import "@/pages/assets/hooks/useLibraryManagement"
 import "@/pages/assets/hooks/useAssetUpload"
 import "@/pages/assets/hooks/useAssetSelection"
 import "@/pages/assets/hooks/useAssetOperations"
-import "@/pages/assets/hooks/asset-operations/batch/useBatchDelete"
-import "@/pages/assets/hooks/asset-operations/batch/useBatchTag"
-import "@/pages/assets/hooks/asset-operations/batch/useBatchClassify"
-import "@/pages/assets/hooks/asset-operations/batch/useBatchMark"
+import "@/pages/assets/hooks/asset-operations/batch-operations/useBatchDelete"
+import "@/pages/assets/hooks/asset-operations/batch-operations/useBatchTag"
+import "@/pages/assets/hooks/asset-operations/batch-operations/useBatchClassify"
+import "@/pages/assets/hooks/asset-operations/batch-operations/useBatchMark"
 
 describe("AssetLibrary module smoke test", () => {
   it("should load all asset modules", () => {

apps/web/src/test/pages/editing-planner/smoke.test.tsx

@@ -53,13 +53,6 @@ import "@/pages/editing-planner/components/clip-properties/SubtitleSettingsSecti
 import "@/pages/editing-planner/components/clip-properties/BgmSettingsSection"
 import "@/pages/editing-planner/components/clip-properties/ClipDetailSection"
 import "@/pages/editing-planner/components/clip-properties/StatsSection"
-import "@/pages/editing-planner/components/pip-config/LayerList"
-import "@/pages/editing-planner/components/pip-config/LayerConfig"
-import "@/pages/editing-planner/components/sticker/StickerLibrary"
-import "@/pages/editing-planner/components/sticker/StickerList"
-import "@/pages/editing-planner/components/sticker/StickerPropsEditor"
-import "@/pages/editing-planner/components/sticker/StickerPreview"
-import "@/pages/editing-planner/components/sticker/TextStickerPropsEditor"
 import "@/pages/editing-planner/components/filter/FilterPresetGrid"
 import "@/pages/editing-planner/components/filter/FilterManualAdjust"
 import "@/pages/editing-planner/components/intro-outro/IntroOutroBlock"
@@ -67,8 +60,6 @@ import "@/pages/editing-planner/components/subtitle-style/SubtitlePreview"
 import "@/pages/editing-planner/components/subtitle-style/SubtitleModeSwitch"
 import "@/pages/editing-planner/components/subtitle-style/SubtitlePositionSelector"
 import "@/pages/editing-planner/components/subtitle-style/SubtitleEffectButtons"
-import "@/pages/editing-planner/components/tts/VoiceSelector"
-import "@/pages/editing-planner/components/tts/TtsSlider"
 import "@/pages/editing-planner/components/watermark/WatermarkTypeTabs"
 import "@/pages/editing-planner/components/watermark/ImageWatermarkSection"
 import "@/pages/editing-planner/components/watermark/TextWatermarkSection"

apps/worker/worker_app/celery_app.py

@@ -34,9 +34,15 @@ celery_app.conf.imports = (
     "worker_app.tasks.tts_synthesis",
     "worker_app.tasks.batch_download",
     "worker_app.tasks.duplication_check",
+    # #1798 AI 数字人渲染：必须在 Worker 实例上注册同名任务，否则消息无人消费（渲染卡 0%）
+    "worker_app.tasks.ai_avatar_render",
     "worker_app.tasks._startup",
     "apps.worker.video_processing.dedup",
     "worker_app.tasks.cleanup",
+    # 注意：必须用 app.* 路径，不能用 apps.api.app.* 路径！
+    # PYTHONPATH=/app/apps/api 下，app.tasks.lipsync_tts 可直接导入且不触发 apps/api/__init__.py
+    # （apps/api/__init__.py 会 from .main import app，级联加载整个 FastAPI 栈，Worker 中不需要且会导致注册失败）
+    "app.tasks.lipsync_tts",
 )
 
 # Celery Beat 定时任务调度

apps/worker/worker_app/tasks/ai_avatar_render.py

@@ -0,0 +1,48 @@
+"""AI 数字人渲染任务 — Worker 侧 Celery 任务注册.
+
+#1798 渲染进度卡在 0% 的根因：渲染任务定义在 API 侧（`app.tasks.ai_avatar_render`），
+装饰在 API 自己的 Celery 实例（`xiaoxia-saas-api`）上；而 Worker 用的是
+`worker_app.celery_app` 实例，`conf.imports` 从未导入该任务，Worker 的任务
+注册表里没有 `ai_avatar_render.execute`，消息被路由到默认 `celery` 队列后
+无人消费，任务永远停在 0%。
+
+修复：在 Worker 侧用 `worker_app.celery_app` 注册同名任务，直接调用与 API
+服务一致的 `AiAvatarRenderService.execute_render` 核心管线（业务逻辑在
+`apps.api.app.services`，worker 镜像已复制 `apps/api/app`）。任务名保持
+`ai_avatar_render.execute`，与 API 生产端 `.delay()` 的消息路由一致；未在
+task_routes 显式配置，走默认 `celery` 队列，由 transcode worker 消费。
+"""
+
+from __future__ import annotations
+
+import logging
+
+from worker_app.celery_app import celery_app
+from worker_app.db import SessionLocal
+
+logger = logging.getLogger(__name__)
+
+
+@celery_app.task(bind=True, name="ai_avatar_render.execute", max_retries=2)
+def execute_ai_avatar_render(self, job_id: str) -> dict:
+    """执行 AI 数字人渲染管线（Worker 侧入口）.
+
+    进度由 service 直接写入 DB（AiAvatarRenderJob.progress:
+    0→5→20→40→80→90→95→100），API 通过轮询 progress 字段展示。
+    """
+    logger.info("开始执行渲染任务: %s", job_id)
+    self.update_state(state="PROCESSING", meta={"progress": 0, "job_id": job_id})
+
+    session = SessionLocal()
+    try:
+        from app.services.ai_avatar_render_service import AiAvatarRenderService
+
+        service = AiAvatarRenderService(session)
+        service.execute_render(job_id)
+        return {"status": "completed", "job_id": job_id}
+    except Exception as exc:
+        logger.exception("渲染任务执行异常 [%s]: %s", job_id, exc)
+        self.update_state(state="FAILED", meta={"progress": 0, "error": str(exc)})
+        raise
+    finally:
+        session.close()

docs/ai-avatar-api-contract-1822.md

@@ -0,0 +1,169 @@
+# AI 数字人前后端接口契约（#1797 / #1822）
+
+> 分支：`fix/ai-avatar-v3-1797`
+> 范围：TTS→对口型链路打通、语速/情绪透传、封面智能选帧、标题字段对齐
+> 本文档为前后端联调的唯一字段口径。
+
+---
+
+## 1. 对口型创建接口 `POST /api/v1/lipsync/jobs`
+
+支持两种输入模式，**二选一**：
+
+### 模式 A（推荐）：TTS 直生 —— 传音色 + 文案，后端内部合成音频
+
+前端无需先调 TTS。后端收到请求后：先调 CosyVoice 合成音频 → 转存 OSS → 再提交 MediaKit 对口型。
+
+```jsonc
+{
+  "video_url": "https://oss.../person.mp4",   // 必填，人物视频（MP4）
+  "voice_id": "cosyvoice-v3-flash-99-xxxx",    // 必填，音色 ID（预置音色 或 克隆 profile UUID）
+  "script_text": "省是浙江省，市是永康市……",     // 必填，要合成的文案
+  "speed": 1.0,                                 // 可选，语速 0.5~2.0，默认 1.0
+  "emotion": "excited",                         // 可选，情绪，见 §3
+  "enable_video_loop": false,                   // 可选，音频长于视频时是否循环画面
+  "project_id": ""                              // 可选
+}
+```
+
+### 模式 B：直接音频 —— 前端已准备好音频
+
+```jsonc
+{
+  "video_url": "https://oss.../person.mp4",     // 必填
+  "audio_url": "https://oss.../voice.mp3",      // 必填，mp3/aac/wav/m4a/flac
+  "enable_video_loop": false
+}
+```
+
+### 校验与错误码
+
+| 场景 | HTTP | detail.code |
+|------|------|-------------|
+| 既无 audio_url 又无 voice_id+script_text | 422 | （schema 校验） |
+| video_url 非 MP4 / audio_url 格式不支持 | 422 | （schema 校验） |
+| 克隆音色不属于当前用户 | 403 | `VoiceForbidden` |
+| 克隆音色尚未合成完成 | 400 | `VoiceNotReady` |
+| TTS 合成失败（如 CosyVoice 欠费） | 502 | `TTSSynthesisFailed` |
+| MediaKit 提交失败 | 502 | `*`（透传 MediaKit code） |
+
+### 轮询
+
+- `GET /api/v1/lipsync/jobs/{id}`：非终态任务先返回 DB 缓存，**后台异步刷新 MediaKit**（不会阻塞轮询）。
+- `status` 流转：`pending` → `submitted` → `running`/`processing`（MediaKit 中间态同步）→ `completed` / `failed`。
+- `completed` 时 `output_video_url` 为**已转存自家 OSS 的非临时 URL**（不会过期）。
+- 前端每 3s 轮询，命中 `completed`/`failed` 即停。
+
+---
+
+## 2. TTS 合成接口语速/情绪透传
+
+- `POST /api/v1/tts/synthesize`（异步任务）与 `POST /api/v1/tts/preview`（即时试听）均新增：
+  - `speed`：float，0.5~2.0，默认 1.0 → 透传 CosyVoice payload 的 `rate`
+  - `emotion`：string，见 §3 映射 → 透传 `emotion`
+- 透传链路：`route → CreateTTSJobUseCase(metadata) → TTSJobWorkflow.start_synthesis / 分段合成 → CosyVoiceService.submit_synthesize_task(rate/emotion)`。
+- 分段合成（长文案）与失败重合成路径同样透传 speed/emotion。
+
+---
+
+## 3. 情绪枚举（前后端统一）
+
+前端把中文选项映射成英文后传后端；后端同时接受中文/英文，非法值忽略（走默认自然）。
+
+| 前端选项 | 传参值 | CosyVoice 枚举 |
+|---------|--------|---------------|
+| 自然 | `natural` | natural |
+| 兴奋 | `excited` | excited |
+| 沉稳 | `calm` | calm |
+| 亲切 | `friendly` | friendly |
+
+后端 `normalize_emotion()` 也接受中文（自然/兴奋/沉稳/亲切）做兜底映射。
+
+---
+
+## 4. 智能封面接口 `POST /api/v1/ai-avatar/render/smart-cover`
+
+独立接口，**不依赖渲染任务**，前端「智能获取封面」按钮直接调用。
+
+**请求**
+```jsonc
+{
+  "video_url": "https://oss.../avatar_output.mp4",  // 必填，数字人视频
+  "max_frames": 5                                    // 可选，抽帧数量 1~10，默认 5
+}
+```
+
+**响应**
+```jsonc
+{
+  "cover_url": "https://oss.../ai-avatar/covers/xxx/cover_yy.jpg", // OSS 非临时 URL
+  "status": "completed",       // completed / fallback_failed
+  "message": ""                // 失败原因
+}
+```
+
+**实现**：复用智能剪辑同款能力 —— MediaKit `extract_frames(SpecifiedFrames)` 抽 5 帧 → `cover_frame_scorer.score_frames`（清晰度+亮度+色彩）评分选最佳 → 转存 OSS。
+**不再使用 FFmpeg 简单首帧**。渲染管线最终封面也优先走该智能选帧，MediaKit 不可用时才回退 FFmpeg。
+
+---
+
+## 5. 渲染接口 `POST /api/v1/ai-avatar/render`
+
+```jsonc
+{
+  "lipsync_job_id": "7c29a3b2-...",     // 必填，已 completed 的对口型任务
+  "script_id": "",                       // 可选！见下方说明
+  "b_roll_segments": [],                 // 可选，B-roll 片段
+  "title_config": { ... },               // 可选，单个标题配置 dict（见 §6）
+  "cover_config": { ... },               // 可选，封面配置（建议改用 smart-cover）
+  "project_id": ""
+}
+```
+
+**`script_id` 是否必填：可选。**
+- 从文案库选了文案时传对应文案 ID（后端做归属校验）。
+- **手动输入文案、走 TTS 直生模式时不传（留空）即可**——渲染管线不依赖文案内容，`script_id` 仅用于归属校验。留空不会卡手动文案用户。
+
+---
+
+## 6. 标题配置 `title_config` 字段清单（以 build_title_drawtext_filter 为准）
+
+渲染请求收的是**单个 `title_config` dict**（不是 `titles[]` 数组），字段与 `packages/domain/video_filter_builder.py` 的 `build_title_drawtext_filter()` 完全对齐：
+
+| 字段 | 别名 | 类型 | 必填 | 默认 | 说明 |
+|------|------|------|------|------|------|
+| `text` | `content` | string | ✅ | — | 标题文字；为空或 `enabled=false` 时不渲染标题 |
+| `enabled` | — | bool | ❌ | `true` | 是否启用标题；false 跳过 |
+| `font` | `font_preset` | string | ❌ | 思源黑体 | 字体名（后端按名字解析字体文件） |
+| `font_size` | `size` | int | ❌ | 36 | 字号（像素） |
+| `font_color` | `color` | string | ❌ | `#ffffff` | 文字颜色，`#RRGGBB`；后端自动去掉 `#`，也可传 `RRGGBB` 或颜色名 |
+| `position` | — | string | ❌ | `top` | 预设位置：`top`(y=50) / `center`(垂直居中) / `bottom`(底部上移50px) / `custom` |
+| `pos_x` | — | int/float | ❌ | — | 自定义 X 坐标（像素），仅 `position=custom` 生效 |
+| `pos_y` | — | int/float | ❌ | — | 自定义 Y 坐标（像素），仅 `position=custom` 生效 |
+| `bold` | — | bool | ❌ | `true` | 粗体（Bold 字体变体，回退 borderw 模拟） |
+| `stroke` | — | bool/object | ❌ | — | 描边。`true`=黑描边宽2；object 见下 |
+| `stroke.enabled` | — | bool | ❌ | true | 是否描边 |
+| `stroke.width` | — | int | ❌ | 2 | 描边宽度 |
+| `stroke.color` | — | string | ❌ | `#000000` | 描边颜色 |
+| `shadow` | — | bool/object | ❌ | — | 阴影。`true`=黑色阴影偏移2px；object 见下 |
+| `shadow.enabled` | — | bool | ❌ | true | 是否阴影 |
+| `shadow.color` | — | string | ❌ | `#000000` | 阴影颜色 |
+| `shadow.offset_x` | — | int | ❌ | 2 | 阴影 X 偏移 |
+| `shadow.offset_y` | — | int | ❌ | 2 | 阴影 Y 偏移 |
+
+**前端注意事项**
+- 标题是**整条成片一个标题**（单个 dict），不是按时间段的标题数组；没有 `start`/`end`/`frame`/`fontSize` 这些字段。
+- 位置用 `position` 四档枚举；自由摆放用 `position="custom"` + `pos_x`/`pos_y`（像素坐标，非比例）。
+- 颜色统一传 `#RRGGBB` 即可，后端会处理 `#`；三档预设位置下标题始终水平居中。
+- `stroke`/`shadow` 传 `true` 用默认样式，或传 object 精细控制颜色/宽度/偏移。
+
+---
+
+## 7. 前端对接清单
+
+1. 对口型：改用**模式 A**（voice_id + script_text + speed + emotion），不要再先调 TTS 拿 audio_url。
+2. 音色 ID：`voice_id` 可直接传克隆音色的 profile UUID，后端会解析为 CosyVoice voice_id（与 /tts 一致）。
+3. 情绪下拉：自然/兴奋/沉稳/亲切 → natural/excited/calm/friendly。
+4. 封面：点「智能获取封面」→ POST `/ai-avatar/render/smart-cover`，用返回的 `cover_url`。
+5. 渲染：手动文案直生场景 `script_id` 留空；标题传**单个** `title_config` dict（字段见 §6）。
+6. 轮询：识别 `running` 等中间态，不要只认 `submitted`。

infra/docker/compose.yml

@@ -128,9 +128,9 @@ services:
       - xiaoxia-net
 
     # 健康检查配置
-    # 注：celery inspect ping 依赖 broker 连接，在容器内不可靠，改用进程检查
+    # 注：容器内无 pgrep/ps，扫描 /proc 所有进程的 cmdline 查找 celery 进程
     healthcheck:
-      test: ["CMD-SHELL", "for pid in /proc/[0-9]*/cmdline; do if grep -ql celery \"$pid\" 2>/dev/null; then exit 0; fi; done; exit 1"]
+      test: ["CMD-SHELL", "grep -lq celery /proc/[0-9]*/cmdline 2>/dev/null || exit 1"]
       interval: 30s
       timeout: 10s
       retries: 3
@@ -163,7 +163,7 @@ services:
     #   context: ../..
     #   dockerfile: ${WEB_DOCKERFILE:-infra/docker/web.Dockerfile}
     #   args:
-    #     NGINX_CONF: ${WEB_NGINX_CONF:-infra/docker/nginx.conf}
+    #     (NGINX_CONF no longer needed - all configs baked into image)
     
     container_name: xiaoxia-web-${ENV:-staging}
     restart: unless-stopped
@@ -178,12 +178,12 @@ services:
       - xiaoxia-net
     
     # =========================================
-    # Nginx 配置运行时覆盖
+    # Nginx 配置运行时覆盖（双保险：entrypoint 也按 APP_ENV 选择配置）
     # 确保容器使用正确环境的 nginx 配置，即使镜像构建时使用了默认配置
     # 注意: 只覆盖 /etc/nginx/conf.d/default.conf，不挂载 /usr/share/nginx/html
     # =========================================
     environment:
-      - NGINX_ENV=${ENV:-staging}
+      - APP_ENV=${ENV:-staging}
     volumes:
       - ./nginx-${ENV:-staging}.conf:/etc/nginx/conf.d/default.conf:ro
     

infra/docker/deploy-production-registry.sh

@@ -14,11 +14,11 @@ REGISTRY_TOKEN="${REGISTRY_TOKEN:-}"
 ENV_FILE="${ENV_FILE:-/var/lib/xiaoxia-saas-production/.env}"
 GENERATED_DIR="${GENERATED_DIR:-/var/lib/xiaoxia-saas-production/generated}"
 LEGACY_ASSETS_DIR="${LEGACY_ASSETS_DIR:-/var/lib/xiaoxia-saas-production/legacy-assets}"
+REPO_DIR="${REPO_DIR:-/var/lib/xiaoxia-saas-production/repo}"
 
 if [ -z "$IMAGE_TAG" ]; then
   echo "ERROR: IMAGE_TAG is required"
   exit 1
-fi
 
 test -f "$ENV_FILE"
 mkdir -p "$GENERATED_DIR"
@@ -31,7 +31,6 @@ if [ -n "$REGISTRY_TOKEN" ]; then
   printf %s "$REGISTRY_TOKEN" | docker login "$REGISTRY_HOST" -u "$REGISTRY_USER" --password-stdin 2>/dev/null || {
     echo "WARN: docker login failed, will try to pull anyway"
   }
-fi
 
 # ---- Pull 三镜像 ----
 REGISTRY_API="${REGISTRY}/xiaoxia-saas-api:${IMAGE_TAG}"
@@ -66,17 +65,14 @@ if docker inspect xiaoxia-web-production >/dev/null 2>&1; then
   if [ -d "$_tmpdir" ] && [ "$(ls -A "$_tmpdir" 2>/dev/null)" ]; then
     cp -an "$_tmpdir"/. "$LEGACY_ASSETS_DIR"/ 2>/dev/null || true
     echo "Legacy assets backed up: $(ls "$_tmpdir" | wc -l) files"
-  fi
   rm -rf "$_tmpdir"
 else
   echo "No existing web container, skipping legacy assets backup"
-fi
 
 # 清理超过 7 天的旧 assets 文件（避免无限增长）
 if [ -d "$LEGACY_ASSETS_DIR" ]; then
   find "$LEGACY_ASSETS_DIR" -type f -mtime +7 -delete 2>/dev/null || true
   echo "Legacy assets cleanup done (retain 7 days)"
-fi
 
 # ---- 确保基础设施容器在运行 ----
 echo "Checking infrastructure containers..."
@@ -84,12 +80,10 @@ for c in xiaoxia-postgres-production xiaoxia-redis-production; do
   if ! docker inspect "$c" >/dev/null 2>&1; then
     echo "ERROR: Required container not found: $c"
     exit 1
-  fi
   state=$(docker inspect -f '{{.State.Status}}' "$c")
   if [ "$state" != "running" ]; then
     echo "ERROR: Container not running: $c ($state)"
     exit 1
-  fi
 done
 
 # ---- 确保生产网络存在 ----
@@ -108,7 +102,6 @@ echo "Migrations completed."
 echo "Stopping old containers..."
 docker rm -f xiaoxia-api-production 2>/dev/null || true
 docker rm -f xiaoxia-worker-production 2>/dev/null || true
-docker rm -f xiaoxia-web-production 2>/dev/null || true
 
 # ---- 日志配置（所有容器共用） ----
 LOG_OPTS="--log-driver json-file --log-opt max-size=50m --log-opt max-file=3"
@@ -155,7 +148,7 @@ docker run -d \
   --restart unless-stopped \
   --cpus 2 \
   --memory 2g \
-  --health-cmd "sh -c \"for pid in /proc/[0-9]*/cmdline; do if grep -ql celery \"$pid\" 2>/dev/null; then exit 0; fi; done; exit 1\"" \
+  --health-cmd "sh -c \"grep -lq celery /proc/[0-9]*/cmdline 2>/dev/null || exit 1\"" \
   --health-interval 30s \
   --health-timeout 10s \
   --health-retries 3 \
@@ -166,15 +159,16 @@ docker run -d \
 # ---- 启动 Web ----
 # Legacy assets 挂载到 /usr/share/nginx/html/assets-legacy/assets/
 # nginx 配置中 assets location 有 fallback 逻辑
-LEGACY_VOLUME=""
+WEB_VOLUMES=""
 if [ -d "$LEGACY_ASSETS_DIR" ] && [ "$(ls -A "$LEGACY_ASSETS_DIR" 2>/dev/null)" ]; then
-  LEGACY_VOLUME="-v ${LEGACY_ASSETS_DIR}:/usr/share/nginx/html/assets-legacy/assets:ro"
+  WEB_VOLUMES="-v ${LEGACY_ASSETS_DIR}:/usr/share/nginx/html/assets-legacy/assets:ro"
   echo "Web container: legacy assets mounted (fallback)"
 else
   echo "Web container: no legacy assets to mount"
 fi
 
 echo "Starting Web container..."
+docker rm -f xiaoxia-web-production 2>/dev/null || true
 docker run -d \
   --name xiaoxia-web-production \
   --network xiaoxia-net-production \
@@ -182,7 +176,8 @@ docker run -d \
   --restart unless-stopped \
   --cpus 0.5 \
   --memory 512m \
-  $LEGACY_VOLUME \
+  -e APP_ENV=production \
+  $WEB_VOLUMES \
   --health-cmd "wget --spider -q http://127.0.0.1:80" \
   --health-interval 30s \
   --health-timeout 5s \
@@ -197,7 +192,6 @@ while [ "$i" -lt 40 ]; do
   if curl -sf --max-time 5 http://127.0.0.1:8001/health >/dev/null 2>&1; then
     echo "API is healthy!"
     break
-  fi
   i=$((i + 1))
   echo "  Waiting... ($i/40)"
   sleep 3
@@ -207,7 +201,6 @@ if [ "$i" -ge 40 ]; then
   echo "ERROR: API did not become healthy within 120s"
   docker logs --tail 50 xiaoxia-api-production
   exit 1
-fi
 
 # ---- 等待 Web 健康 ----
 echo "Waiting for Web to become healthy..."
@@ -216,7 +209,6 @@ while [ "$i" -lt 15 ]; do
   if curl -sf --max-time 5 http://127.0.0.1:3002/ >/dev/null 2>&1; then
     echo "Web is healthy!"
     break
-  fi
   i=$((i + 1))
   echo "  Waiting... ($i/15)"
   sleep 2
@@ -226,7 +218,6 @@ if [ "$i" -ge 15 ]; then
   echo "ERROR: Web did not become healthy within 30s"
   docker logs --tail 30 xiaoxia-web-production
   exit 1
-fi
 
 # ---- 清理旧镜像 ----
 echo "Cleaning up old images..."

infra/docker/deploy-staging-registry.sh

@@ -116,7 +116,7 @@ docker run -d \
   -v "$GENERATED_DIR:/app/generated" \
   --restart unless-stopped \
   --label com.centurylinklabs.watchtower.enable=true \
-  --health-cmd "sh -c \"for pid in /proc/[0-9]*/cmdline; do if grep -ql celery \"$pid\" 2>/dev/null; then exit 0; fi; done; exit 1\"" \
+  --health-cmd "sh -c \"grep -lq celery /proc/[0-9]*/cmdline 2>/dev/null || exit 1\"" \
   --health-interval 30s \
   --health-timeout 10s \
   --health-retries 3 \
@@ -124,23 +124,15 @@ docker run -d \
   "$LOCAL_WORKER"
 
 # ---- 启动 Web ----
-# Web 镜像默认打包 production nginx.conf，staging 需要挂载 staging 配置
-NGINX_CONF="${NGINX_CONF:-${COMPOSE_DIR}/nginx-staging.conf}"
-if [ ! -f "$NGINX_CONF" ]; then
-  echo "WARN: nginx config not found at $NGINX_CONF, using image default"
-  NGINX_VOLUME=""
-else
-  NGINX_VOLUME="-v ${NGINX_CONF}:/etc/nginx/conf.d/default.conf:ro"
-fi
-
 echo "Starting Web container..."
+docker rm -f xiaoxia-web-staging 2>/dev/null || true
 docker run -d \
   --name xiaoxia-web-staging \
   --network xiaoxia-net-staging \
   -p 127.0.0.1:3001:80 \
   --restart unless-stopped \
   --label com.centurylinklabs.watchtower.enable=true \
-  $NGINX_VOLUME \
+  -e APP_ENV=staging \
   --health-cmd "wget --spider -q http://127.0.0.1:80" \
   --health-interval 30s \
   --health-timeout 5s \

infra/docker/deploy-staging.sh

@@ -53,7 +53,6 @@ export API_IMAGE="${API_IMAGE:-${REGISTRY}/xiaoxia-saas-api:dev}"
 export WORKER_IMAGE="${WORKER_IMAGE:-${REGISTRY}/xiaoxia-saas-worker:dev}"
 
 # Use staging-specific nginx config (proxy_pass → xiaoxia-api-staging:8000)
-export WEB_NGINX_CONF=infra/docker/nginx-staging.conf
 
 if [ "${REBUILD_BACKEND:-0}" = "1" ] || [ "${BUILD_WEB:-0}" = "1" ]; then
   if [ "${ALLOW_STAGING_BUILDS:-false}" != "true" ]; then

infra/docker/nginx-entrypoint.sh

@@ -0,0 +1,49 @@
+#!/bin/sh
+# Select nginx config based on APP_ENV (staging/production).
+#
+# 两种运行模式：
+# 1. CI/CD 部署（staging/production）：部署脚本通过 `-v 宿主机文件:/etc/nginx/conf.d/default.conf:ro`
+#    把宿主机生成的带 resolver/docker upstream 的配置 bind mount 进来，entrypoint 不应改动。
+#    bind mount 的文件是 readonly 的，rm 会报 EBUSY ("Resource busy")，直接 exec nginx 即可。
+# 2. 本地 docker-compose / 直接 `docker run`（无外部挂载）：镜像烤入了 nginx-staging.conf 与
+#    nginx-production.conf 到 /etc/nginx/，entrypoint 根据 APP_ENV 把 default.conf 换成正确的 symlink。
+#
+# 策略：
+# - 如果 /etc/nginx/conf.d/default.conf 已经是指向目标 conf 的 symlink，什么都不做；
+# - 否则尝试 rm -f 再 ln -s；rm 失败说明是外部 bind mount（已有正确配置），不阻塞启动；
+# - 兜底：只要 conf.d 目录里有 .conf 文件（含 bind mount 来的），就直接启动 nginx。
+set -e
+
+NGINX_CONF_DIR="/etc/nginx/conf.d"
+TARGET_CONF=""
+
+case "${APP_ENV:-production}" in
+  staging)
+    TARGET_CONF="/etc/nginx/nginx-staging.conf"
+    ;;
+  *)
+    TARGET_CONF="/etc/nginx/nginx-production.conf"
+    ;;
+esac
+
+DEFAULT_CONF="$NGINX_CONF_DIR/default.conf"
+
+# 1. 已经是正确的 symlink：直接启动
+if [ -L "$DEFAULT_CONF" ] && [ "$(readlink "$DEFAULT_CONF" 2>/dev/null)" = "$TARGET_CONF" ]; then
+  exec nginx -g "daemon off;"
+fi
+
+# 2. 尝试替换为目标 symlink（无 bind mount 的场景）
+#    若 rm 失败（bind mount readonly，EBUSY/EPERM），则认为外部已注入配置，不阻塞。
+rm -f "$DEFAULT_CONF" 2>/dev/null || true
+if [ -f "$TARGET_CONF" ] && [ ! -e "$DEFAULT_CONF" ]; then
+  ln -s "$TARGET_CONF" "$DEFAULT_CONF" 2>/dev/null || true
+fi
+
+# 3. 兜底：至少要有一个 .conf 文件，否则 nginx 起不来
+if ! ls "$NGINX_CONF_DIR"/*.conf >/dev/null 2>&1; then
+  echo "ERROR: no nginx config found in $NGINX_CONF_DIR (tried $TARGET_CONF and external bind mount)" >&2
+  exit 1
+fi
+
+exec nginx -g "daemon off;"

infra/docker/web-artifact.Dockerfile

@@ -1,7 +1,11 @@
 FROM git.xiaoxiajianji.com/xiaoxia/base/nginx:alpine AS runner
-ARG NGINX_CONF=infra/docker/nginx.conf
 WORKDIR /usr/share/nginx/html
 COPY apps/web/dist ./
-COPY ${NGINX_CONF} /etc/nginx/conf.d/default.conf
+# 将所有 nginx 配置烤入镜像，entrypoint 按 APP_ENV 选择
+COPY infra/docker/nginx.conf /etc/nginx/nginx-production.conf
+COPY infra/docker/nginx-staging.conf /etc/nginx/nginx-staging.conf
+COPY infra/docker/nginx-production.conf /etc/nginx/nginx-production.conf
+COPY infra/docker/nginx-entrypoint.sh /docker-entrypoint.sh
+RUN chmod +x /docker-entrypoint.sh
 EXPOSE 80
-CMD ["nginx", "-g", "daemon off;"]
+ENTRYPOINT ["/docker-entrypoint.sh"]

infra/docker/web.Dockerfile

@@ -28,9 +28,13 @@ RUN --mount=type=cache,target=/app/apps/web/.tscache,sharing=locked \
 
 # Production stage with nginx
 FROM git.xiaoxiajianji.com/xiaoxia/base/nginx:alpine AS runner
-ARG NGINX_CONF=infra/docker/nginx.conf
 WORKDIR /usr/share/nginx/html
 COPY --from=builder /app/apps/web/dist ./
-COPY ${NGINX_CONF} /etc/nginx/conf.d/default.conf
+# 将所有 nginx 配置烤入镜像，entrypoint 按 APP_ENV 选择
+COPY infra/docker/nginx.conf /etc/nginx/nginx-production.conf
+COPY infra/docker/nginx-staging.conf /etc/nginx/nginx-staging.conf
+COPY infra/docker/nginx-production.conf /etc/nginx/nginx-production.conf
+COPY infra/docker/nginx-entrypoint.sh /docker-entrypoint.sh
+RUN chmod +x /docker-entrypoint.sh
 EXPOSE 80
-CMD ["nginx", "-g", "daemon off;"]
+ENTRYPOINT ["/docker-entrypoint.sh"]

infra/docker/worker.Dockerfile

@@ -20,7 +20,7 @@ WORKDIR /app
 
 # 设置 Python 环境变量
 ENV PATH="/opt/venv/bin:$PATH"
-ENV PYTHONPATH=/app:/app/packages
+ENV PYTHONPATH=/app:/app/apps/api:/app/packages
 ENV PYTHONUNBUFFERED=1
 ENV APP_VERSION=$APP_VERSION
 
@@ -28,8 +28,10 @@ ENV APP_VERSION=$APP_VERSION
 COPY alembic.ini /app/alembic.ini
 COPY migrations/ /app/migrations/
 COPY packages/ /app/packages/
-COPY apps/api/app/config.py /app/apps/api/app/config.py
-COPY apps/api/app/core/ /app/apps/api/app/core/
+# PR #1844 起，worker 还需要加载 apps.api.app.tasks.lipsync_tts，
+# 该 task 依赖 app.services.* 与 app.core.celery_app（PYTHONPATH=/app/apps/api 下解析）。
+# 为避免后续新增 task 再次漏 COPY，直接把整个 apps/api/app/ 复制进 worker 镜像。
+COPY apps/api/app/ /app/apps/api/app/
 
 # Worker 启动脚本
 COPY infra/docker/entrypoint-worker.sh /usr/local/bin/entrypoint-worker.sh
@@ -38,6 +40,10 @@ RUN chmod +x /usr/local/bin/entrypoint-worker.sh
 # 业务代码（变化最频繁，放最后）
 COPY apps/worker/ /app/apps/worker/
 
+# 健康检查：扫描所有进程的 cmdline 查找 celery 进程
+HEALTHCHECK --interval=30s --timeout=10s --start-period=40s --retries=3 \
+    CMD grep -lq celery /proc/[0-9]*/cmdline 2>/dev/null || exit 1
+
 USER celery
 WORKDIR /app/apps/worker
 CMD ["/usr/local/bin/entrypoint-worker.sh"]

packages/__init__.py

@@ -0,0 +1 @@
+# xiaoxia-saas shared packages namespace

packages/adapters/__init__.py

@@ -0,0 +1 @@
+# adapter implementations namespace

packages/adapters/sqlalchemy_impl/models.py

@@ -684,9 +684,15 @@ class LipsyncJobModel(Base):
 
     # 输入参数
     video_url = Column(Text, nullable=False)
-    audio_url = Column(Text, nullable=False)
+    audio_url = Column(Text, nullable=True)  # 直生模式（voice_id+script_text）下 TTS 合成后回填
     enable_video_loop = Column(Boolean, nullable=False, default=False)
 
+    # TTS 直生字段：传音色 + 文案，由后端先合成音频再对口型
+    voice_id = Column(String(200), nullable=False, default="")
+    script_text = Column(Text, nullable=False, default="")
+    speed = Column(Float, nullable=False, default=1.0)
+    emotion = Column(String(20), nullable=False, default="")
+
     # MediaKit 任务状态
     mediakit_task_id = Column(String(200), nullable=False, default="", index=True)
     status = Column(
@@ -697,6 +703,9 @@ class LipsyncJobModel(Base):
     error_message = Column(Text, nullable=False, default="")
     error_code = Column(String(100), nullable=False, default="")
 
+    # 精确句子时间戳（TTS 合成后由 silencedetect 计算，用于 B-roll 精确定位）
+    sentence_timings = Column(JSON, nullable=True)  # list[{index,text,start_time,end_time}]
+
     # 时间戳
     submitted_at = Column(DateTime, nullable=True)
     completed_at = Column(DateTime, nullable=True)
@@ -715,7 +724,8 @@ class AiAvatarRenderJob(Base):
 
     # 输入参数
     lipsync_job_id = Column(String(36), nullable=False)
-    script_id = Column(String(36), nullable=False)
+    # 文案 ID 可选：手动输入文案（TTS 直生）场景不关联文案库条目
+    script_id = Column(String(36), nullable=False, default="")
     b_roll_segments = Column(JSON, nullable=False, default=list)
     # b_roll_segments 格式: [{"script_segment_index": 0, "asset_url": "...", "mode": "fullscreen|pip", "start_time": 5.0, "end_time": 10.0}, ...]
     title_config = Column(JSON, nullable=False, default=dict)

packages/application/cosyvoice_service.py

@@ -25,6 +25,35 @@ from packages.shared.config import get_shared_settings
 logger = logging.getLogger(__name__)
 
 
+# CosyVoice 支持的情绪：中文标签 → API 英文值
+EMOTION_MAP = {
+    "自然": "natural",
+    "兴奋": "excited",
+    "沉稳": "calm",
+    "亲切": "friendly",
+    "natural": "natural",
+    "excited": "excited",
+    "calm": "calm",
+    "friendly": "friendly",
+}
+VALID_EMOTIONS = {"natural", "excited", "calm", "friendly"}
+
+
+def normalize_emotion(emotion: str) -> str:
+    """将前端情绪值归一化为 CosyVoice 英文枚举。
+
+    支持中文（自然/兴奋/沉稳/亲切）和英文；非法值返回空串（不传，走默认）。
+    """
+    if not emotion:
+        return ""
+    key = emotion.strip().lower()
+    mapped = EMOTION_MAP.get(emotion.strip()) or EMOTION_MAP.get(key)
+    if mapped and mapped in VALID_EMOTIONS:
+        return mapped
+    logger.warning("未知的 emotion 值，忽略: %r", emotion)
+    return ""
+
+
 class CosyVoiceError(Exception):
     """CosyVoice API 调用异常。"""
 
@@ -431,6 +460,7 @@ class CosyVoiceService:
         format: str = "",
         speed: float = 1.0,
         volume: int = 50,
+        emotion: str = "",
     ) -> dict:
         """提交语音合成任务（同步非流式，直接返回结果）.
 
@@ -444,6 +474,7 @@ class CosyVoiceService:
             format: 输出格式（mp3/wav/pcm），空表示使用配置默认值
             speed: 语速（0.5-2.0），1.0 为正常速度
             volume: 音量（0-100），默认 50
+            emotion: 情绪（natural/excited/calm/friendly），空串不传
 
         Returns:
             dict: {"audio_url": str, "request_id": str,
@@ -463,17 +494,20 @@ class CosyVoiceService:
 
         settings = get_shared_settings()
 
-        payload = {
-            "model": self._model,
-            "input": {
-                "text": text,
-                "voice": voice_id,
-                "format": format or settings.cosyvoice_format,
-                "sample_rate": sample_rate or settings.cosyvoice_sample_rate,
-                "rate": speed,
-                "volume": volume,
-            },
+        input_payload: dict[str, Any] = {
+            "text": text,
+            "voice": voice_id,
+            "format": format or settings.cosyvoice_format,
+            "sample_rate": sample_rate or settings.cosyvoice_sample_rate,
+            "rate": speed,
+            "volume": volume,
         }
+        # 情绪：归一化（中文→英文）后透传；空/非法则不传，走 CosyVoice 默认
+        norm_emotion = normalize_emotion(emotion)
+        if norm_emotion:
+            input_payload["emotion"] = norm_emotion
+
+        payload = {"model": self._model, "input": input_payload}
 
         response = self._call_api(
             method="POST",
@@ -490,6 +524,10 @@ class CosyVoiceService:
         if not audio_url:
             raise CosyVoiceError(f"CosyVoice API 未返回 audio_url: {response}")
 
+        # DashScope 返回 http://，统一升级为 https://
+        if audio_url.startswith("http://"):
+            audio_url = audio_url.replace("http://", "https://", 1)
+
         return {
             "task_id": "",  # 同步接口无 task_id，兼容旧接口
             "audio_url": audio_url,
@@ -517,6 +555,7 @@ class CosyVoiceService:
         format: str = "",
         speed: float = 1.0,
         volume: int = 50,
+        emotion: str = "",
         timeout: float = 120.0,
     ) -> SynthesizeResult:
         """语音合成（同步非流式）.
@@ -548,6 +587,7 @@ class CosyVoiceService:
             format=format,
             speed=speed,
             volume=volume,
+            emotion=emotion,
         )
 
         return SynthesizeResult(

packages/application/tts_job/workflow.py

@@ -143,11 +143,16 @@ class TTSWorkflowService:
             return self._start_segment_synthesis(job)
 
         try:
+            _meta = dict(job.metadata)
+            _speed = float(_meta.get("speed", 1.0) or 1.0)
+            _emotion = str(_meta.get("emotion", "") or "")
             submit_result = self.cosyvoice_service.submit_synthesize_task(
                 text=job.input_text,
                 voice_id=job.voice_id,
                 sample_rate=job.sample_rate,
                 format=job.format,
+                speed=_speed,
+                emotion=_emotion,
             )
 
             # 保存 task_id / request_id 到 metadata
@@ -283,6 +288,7 @@ class TTSWorkflowService:
             job_metadata = job.metadata or {}
             speed = float(job_metadata.get("speed", 1.0))
             volume = int(job_metadata.get("volume", 50))
+            emotion = str(job_metadata.get("emotion", "") or "")
 
             result = self.cosyvoice_service.submit_synthesize_task(
                 text=job.input_text,
@@ -291,6 +297,7 @@ class TTSWorkflowService:
                 format=job.format,
                 speed=speed,
                 volume=volume,
+                emotion=emotion,
             )
             audio_url = result.get("audio_url", "")
             if not audio_url:
@@ -401,6 +408,9 @@ class TTSWorkflowService:
         """
         max_workers = min(len(segments), _MAX_SEGMENT_WORKERS)
         results: list[dict | None] = [None] * len(segments)
+        _seg_meta = job.metadata or {}
+        _seg_speed = float(_seg_meta.get("speed", 1.0) or 1.0)
+        _seg_emotion = str(_seg_meta.get("emotion", "") or "")
 
         with ThreadPoolExecutor(max_workers=max_workers) as executor:
             future_to_idx = {}
@@ -411,6 +421,8 @@ class TTSWorkflowService:
                     voice_id=job.voice_id,
                     sample_rate=job.sample_rate,
                     format=job.format,
+                    speed=_seg_speed,
+                    emotion=_seg_emotion,
                 )
                 future_to_idx[future] = idx
 
@@ -500,6 +512,7 @@ class TTSWorkflowService:
         job_metadata = job.metadata or {}
         speed = float(job_metadata.get("speed", 1.0))
         volume = int(job_metadata.get("volume", 50))
+        emotion = str(job_metadata.get("emotion", "") or "")
 
         # 分段文本（用于缺失段重新合成）
         segments = split_text(job.input_text, max_chars=_SEGMENT_THRESHOLD)
@@ -534,6 +547,7 @@ class TTSWorkflowService:
                         format=job.format,
                         speed=speed,
                         volume=volume,
+                        emotion=emotion,
                     )
                     future_to_idx[future] = idx
 

packages/domain/generated_video.py

@@ -49,19 +49,17 @@ class GeneratedVideo:
         thumbnail_url: str | None = None,
         generation_params: dict[str, Any] | None = None,
     ) -> "GeneratedVideo":
-        if not project_id.strip():
-            raise ValueError("project_id cannot be empty")
-        if not generation_task_id.strip():
-            raise ValueError("generation_task_id cannot be empty")
-        if not name.strip():
+        # project_id / generation_task_id 允许为空：AI数字人等无项目场景下，前端可能不传 project_id；
+        # lipsync 路径下 generation_task_id 也可能暂时为空。空串会被下面统一兜底为 "" 入库。
+        if not name or not name.strip():
             raise ValueError("name cannot be empty")
-        if not file_url.strip():
+        if not file_url or not file_url.strip():
             raise ValueError("file_url cannot be empty")
         return cls(
             id=uuid4().hex,
-            project_id=project_id.strip(),
-            user_id=user_id.strip(),
-            generation_task_id=generation_task_id.strip(),
+            project_id=(project_id or "").strip(),
+            user_id=(user_id or "").strip(),
+            generation_task_id=(generation_task_id or "").strip(),
             name=name.strip(),
             file_url=file_url.strip(),
             file_size=file_size,

packages/domain/sentence_timings.py

@@ -0,0 +1,205 @@
+"""共享的句子时间戳计算工具 — 供 Celery TTS 任务和 /lipsync/tts-preview 同步接口复用.
+
+- `_split_script_into_sentences`: 按标点分句（中英文逗号/句号/问号/感叹号/分号/换行）
+- `_estimate_sentence_timings_by_chars`: 按字数比例估算（静音检测失败时降级）
+- `_probe_audio_duration`: ffprobe 读取音频时长
+- `compute_sentence_timings`: 基于 ffmpeg silencedetect 精确计算每句起止时间
+"""
+
+from __future__ import annotations
+
+import logging
+import os
+import re
+import subprocess
+import tempfile
+from typing import Optional
+
+logger = logging.getLogger(__name__)
+
+
+def split_script_into_sentences(script_text: str) -> list[str]:
+    """按句号/问号/感叹号/分号/逗号/换行分句（与前端 SENTENCE_SPLIT_RE 一致）.
+
+    中文短视频文案习惯用「，」断小句（如"卖花的叫花无缺，卖姜的叫姜子牙"），
+    必须把逗号也纳入分隔符，否则多句文案会被识别成一整句，导致 B-roll 时间戳错位。
+    """
+    text = (script_text or "").strip()
+    if not text:
+        return []
+    parts = re.split(r"[。！？!?？！;；,，\n\r]+", text)
+    return [p.strip() for p in parts if p.strip()]
+
+
+def estimate_sentence_timings_by_chars(sentences: list[str], total_duration: float) -> list[dict]:
+    """降级方案：按字数比例估算句子时间（与原前端逻辑一致）."""
+    if not sentences or total_duration <= 0:
+        return []
+    total_chars = sum(len(s.replace(r"\s", "")) for s in sentences)
+    if total_chars == 0:
+        return []
+
+    timings = []
+    acc = 0
+    for i, sent in enumerate(sentences):
+        chars = len(sent.replace(r"\s", ""))
+        start = (acc / total_chars) * total_duration
+        end = ((acc + chars) / total_chars) * total_duration
+        timings.append(
+            {
+                "index": i,
+                "text": sent,
+                "start_time": round(start, 2),
+                "end_time": round(end, 2),
+            }
+        )
+        acc += chars
+    return timings
+
+
+def probe_audio_duration(audio_data: bytes, timeout: int = 10) -> float:
+    """用 ffprobe 读取音频字节流的时长（秒）.
+
+    Returns:
+        时长（秒），失败返回 0.0
+    """
+    if not audio_data:
+        return 0.0
+    tmp_path: Optional[str] = None
+    try:
+        with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as tmp:
+            tmp.write(audio_data)
+            tmp_path = tmp.name
+        result = subprocess.run(
+            [
+                "ffprobe",
+                "-v",
+                "error",
+                "-show_entries",
+                "format=duration",
+                "-of",
+                "default=noprint_wrappers=1:nokey=1",
+                tmp_path,
+            ],
+            capture_output=True,
+            text=True,
+            timeout=timeout,
+        )
+        stdout = (result.stdout or "").strip()
+        if not stdout:
+            logger.warning("[sentence_timings] ffprobe 无输出: stderr=%s", (result.stderr or "")[:200])
+            return 0.0
+        return float(stdout)
+    except Exception as exc:
+        logger.warning("[sentence_timings] ffprobe 时长探测失败: %s", exc)
+        return 0.0
+    finally:
+        if tmp_path:
+            try:
+                os.unlink(tmp_path)
+            except Exception:
+                pass
+
+
+def compute_sentence_timings(audio_data: bytes, script_text: str, total_duration: float) -> list[dict]:
+    """基于 TTS 音频的静音检测，精确计算每句文案的起止时间.
+
+    使用 ffmpeg silencedetect 检测静音段，将静音点与句子边界对齐。
+    比字数比例估算准确得多。
+
+    Args:
+        audio_data: TTS 音频二进制数据（MP3）
+        script_text: 文案全文
+        total_duration: 音频总时长（秒）
+
+    Returns:
+        list[{"index": int, "text": str, "start_time": float, "end_time": float}]
+    """
+    sentences = split_script_into_sentences(script_text)
+    if not sentences:
+        return []
+
+    tmp_path: Optional[str] = None
+    try:
+        with tempfile.NamedTemporaryFile(suffix=".mp3", delete=False) as tmp:
+            tmp.write(audio_data)
+            tmp_path = tmp.name
+
+        result = subprocess.run(
+            [
+                "ffmpeg",
+                "-i",
+                tmp_path,
+                "-af",
+                "silencedetect=noise=-25dB:d=0.3",
+                "-f",
+                "null",
+                "-",
+            ],
+            capture_output=True,
+            text=True,
+            timeout=30,
+        )
+        stderr = result.stderr or ""
+
+        silence_ends = []
+        for match in re.finditer(r"silence_end:\s*([\d.]+)", stderr):
+            t = float(match.group(1))
+            if 0 < t < total_duration:
+                silence_ends.append(t)
+
+        if len(silence_ends) < len(sentences) - 1:
+            logger.warning(
+                "[sentence_timings] 静音点不足(%d < %d)，降级为字数比例估算",
+                len(silence_ends),
+                len(sentences) - 1,
+            )
+            return estimate_sentence_timings_by_chars(sentences, total_duration)
+
+        n_boundaries = len(sentences) - 1
+        boundaries = []
+        used_indices = set()
+
+        for i in range(n_boundaries):
+            expected_pos = (i + 1) / len(sentences) * total_duration
+            best_idx = None
+            best_dist = float("inf")
+            for j, t in enumerate(silence_ends):
+                if j in used_indices:
+                    continue
+                dist = abs(t - expected_pos)
+                if dist < best_dist:
+                    best_dist = dist
+                    best_idx = j
+            if best_idx is not None:
+                used_indices.add(best_idx)
+                boundaries.append(silence_ends[best_idx])
+
+        boundaries.sort()
+
+        timings = []
+        prev_end = 0.0
+        for i, sent in enumerate(sentences):
+            start = prev_end
+            end = boundaries[i] if i < len(boundaries) else total_duration
+            timings.append(
+                {
+                    "index": i,
+                    "text": sent,
+                    "start_time": round(start, 2),
+                    "end_time": round(end, 2),
+                }
+            )
+            prev_end = end
+
+        return timings
+
+    except Exception as exc:
+        logger.warning("[sentence_timings] 静音检测异常，降级为字数比例估算: %s", exc)
+        return estimate_sentence_timings_by_chars(sentences, total_duration)
+    finally:
+        if tmp_path:
+            try:
+                os.unlink(tmp_path)
+            except Exception:
+                pass

packages/domain/video_filter_builder.py

@@ -377,26 +377,30 @@ def _append_audio_concat(parts: list[str], clip_chains: list[ClipFilterChain]) -
 
 # ── 标题 drawtext 滤镜构建（#1789）─────────────────────────────────────────────
 
-# drawtext 字体搜索路径：按优先级列出常见安装位置
-# 服务器使用 Noto Sans SC（思源黑体）作为默认字体
+# drawtext 字体搜索路径：按优先级从高到低排列
+# 服务器使用 Noto Sans SC（思源黑体）作为默认字体。
+# - NotoSansSC-VF.ttf 是 worker-base.Dockerfile 中 COPY 的 VF 字体（含所有字重，无 Mono 变体），优先级最高
+# - .ttc 系列为 fonts-noto-cjk 包预装字体（Dockerfile 已删除含 Mono 变体的旧 .ttc，存在时作为 fallback）
+# - DejaVuSans 仅含拉丁字符不支持中文，已移除
 DRAWTEXT_FONT_SEARCH_PATHS: list[str] = [
+    "/usr/share/fonts/opentype/noto/NotoSansSC-VF.ttf",
     "/usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc",
+    "/usr/share/fonts/opentype/noto/NotoSansCJK-Bold.ttc",
     "/usr/share/fonts/noto-cjk/NotoSansCJK-Regular.ttc",
     "/usr/share/fonts/google-noto-cjk/NotoSansCJK-Regular.ttc",
     "/usr/share/fonts/truetype/noto/NotoSansSC-Regular.ttf",
     "/usr/share/fonts/noto/NotoSansSC-Regular.ttf",
-    "/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf",
 ]
 
-# 前端字体名 → drawtext 字体搜索关键字
+# 前端字体名 → drawtext 字体搜索关键字（匹配 DRAWTEXT_FONT_SEARCH_PATHS 中的文件名关键字）
 DRAWTEXT_FONT_MAP: dict[str, str] = {
-    "思源黑体": "NotoSansCJK",
+    "思源黑体": "NotoSansSC",
     "思源宋体": "NotoSerifCJK",
-    "苹方": "NotoSansCJK",
-    "PingFang": "NotoSansCJK",
-    "微软雅黑": "NotoSansCJK",
+    "苹方": "NotoSansSC",
+    "PingFang": "NotoSansSC",
+    "微软雅黑": "NotoSansSC",
     "楷体": "NotoSerifCJK",
-    "华康俪金黑": "NotoSansCJK",
+    "华康俪金黑": "NotoSansSC",
 }
 
 
@@ -416,6 +420,11 @@ def _escape_drawtext_text(text: str) -> str:
     return result
 
 
+# 粗体字体已由前端 Canvas 直接渲染（Canvas 使用浏览器原生粗体 glyph），
+# FFmpeg 侧不再需要查找 Bold 字体文件；drawtext 仅作为旧版前端的降级路径，
+# 通过 borderw 黑色细描边模拟粗体（见 build_title_drawtext_filter）。
+
+
 def _resolve_font_path(font_name: str) -> str:
     """解析字体名到服务器实际字体文件路径。
 
@@ -423,6 +432,9 @@ def _resolve_font_path(font_name: str) -> str:
     1. 通过 DRAWTEXT_FONT_MAP 映射前端字体名到服务器关键字
     2. 在 DRAWTEXT_FONT_SEARCH_PATHS 中查找匹配路径
     3. 未找到则返回空字符串（drawtext 使用内置默认字体）
+
+    注：粗体已由前端 Canvas 渲染时直接用浏览器 bold glyph 绘制，
+    此处仅作为旧版前端降级路径，无需切换 Bold 字体文件。
     """
     keyword = DRAWTEXT_FONT_MAP.get(font_name, font_name)
     import os
@@ -466,8 +478,10 @@ def build_title_drawtext_filter(
     if not title_config or not isinstance(title_config, dict):
         return None
 
-    # 字段名归一化：兼容 content/text、font_preset/font 两套命名
-    text = (title_config.get("text") or title_config.get("content") or "").strip()
+    # 字段名归一化：兼容 content/text/title 三套命名
+    text = (
+        title_config.get("text") or title_config.get("content") or title_config.get("title") or ""
+    ).strip()
     if not text:
         return None
 
@@ -477,13 +491,13 @@ def build_title_drawtext_filter(
 
     # ── 样式参数 ──
     font_name = title_config.get("font") or title_config.get("font_preset") or "思源黑体"
-    font_size = int(title_config.get("font_size") or title_config.get("size") or 36)
+    font_size = int(title_config.get("font_size") or title_config.get("size") or 48)
     font_color = title_config.get("font_color") or title_config.get("color") or "#ffffff"
     # 去掉 # 前缀（drawtext 用纯 hex 或颜色名）
     if font_color.startswith("#"):
         font_color = font_color[1:]
 
-    position = title_config.get("position", "top")
+    position = title_config.get("position") or "bottom"
     bold = bool(title_config.get("bold", True))
     stroke = title_config.get("stroke")
     shadow = title_config.get("shadow")
@@ -491,7 +505,7 @@ def build_title_drawtext_filter(
     # ── 构建 drawtext 参数 ──
     params: list[str] = []
 
-    # 字体文件
+    # 字体文件（drawtext 降级路径：粗体通过 borderw 黑色描边模拟）
     font_path = _resolve_font_path(font_name)
     if font_path:
         escaped_path = font_path.replace("\\", "\\\\").replace(":", "\\\\:").replace("'", "\\\\'")
@@ -504,26 +518,28 @@ def build_title_drawtext_filter(
     params.append(f"fontsize={font_size}")
     params.append(f"fontcolor={font_color}")
 
-    # 粗体：bold 在 drawtext 中通过 font 的 Bold 变体实现
-    # 若字体有 Bold 变体可用 fontfont=bold；否则通过 borderw 模拟
-    if bold:
-        # 使用 font 参数尝试加载 Bold 变体（Noto Sans SC 有 Bold 变体文件）
-        params.append("font=bold")
-
     # 描边（borderw 需要 libfreetype 支持）
+    # 之前用 borderw=3 + font_color 同色描边模拟粗体，会在小字号/竖屏视频上造成
+    # 字形偏移、边缘重影，看起来像文字被打印了两次（用户截图中的标题"曝光曝光…"）。
+    # 修复：粗体改用黑色细描边（borderw=2, 黑色），视觉上清晰加粗且不产生偏移。
+    # 用户显式开启 stroke 时按用户配置走；粗体+无stroke 默认黑色细描边。
+    border_width = 0
+    border_color = "000000"
     if stroke:
         if isinstance(stroke, bool):
             border_width = 2
-            border_color = "black"
+            border_color = "000000"
         elif isinstance(stroke, dict):
-            border_width = int(stroke.get("width", 2)) if stroke.get("enabled", True) else 0
-            border_color = (stroke.get("color") or "#000000").lstrip("#")
-        else:
-            border_width = 0
-            border_color = "black"
-        if border_width > 0:
-            params.append(f"borderw={border_width}")
-            params.append(f"bordercolor={border_color}")
+            if stroke.get("enabled", True):
+                border_width = int(stroke.get("width", 2))
+                border_color = (stroke.get("color") or "#000000").lstrip("#")
+    elif bold:
+        # 粗体模式且未配描边：黑色细描边，模拟粗体同时保证不重影
+        border_width = 2
+        border_color = "000000"
+    if border_width > 0:
+        params.append(f"borderw={border_width}")
+        params.append(f"bordercolor={border_color}")
 
     # 阴影（shadowcolor + shadowx/y）
     if shadow:
@@ -548,8 +564,13 @@ def build_title_drawtext_filter(
         and not isinstance(pos_x, bool)
         and not isinstance(pos_y, bool)
     ):
-        params.append(f"x={int(pos_x)}")
-        params.append(f"y={int(pos_y)}")
+        # pos_x/pos_y 为百分比坐标（0-100），转换为 drawtext 表达式
+        # 例如 pos_x=50 → x=(w-text_w)*0.50（水平居中偏50%）
+        #      pos_y=30 → y=(h-text_h)*0.30
+        pct_x = max(0.0, min(100.0, float(pos_x))) / 100.0
+        pct_y = max(0.0, min(100.0, float(pos_y))) / 100.0
+        params.append(f"x=(w-text_w)*{pct_x:.4f}")
+        params.append(f"y=(h-text_h)*{pct_y:.4f}")
     else:
         # 三档预设位置：top / center / bottom
         # x 始终水平居中：(w-text_w)/2
@@ -565,6 +586,43 @@ def build_title_drawtext_filter(
     return "drawtext=" + ":".join(params)
 
 
+def build_title_overlay_filter(
+    title_config: dict[str, Any],
+    output_width: int,  # noqa: ARG001 - 保留参数签名，PNG 已按视频分辨率绘制
+    output_height: int,  # noqa: ARG001
+    title_png_path: str,
+    *,
+    title_input_label: str = "[1:v]",
+    base_label: str = "[0:v]",
+    output_label: str = "vout_titled",
+) -> str | None:
+    """构建标题 PNG 图层 overlay 滤镜（WYSIWYG 路径）。
+
+    前端用 Canvas 把标题画成与视频同分辨率的透明 PNG（所见即所得），
+    后端直接 overlay=0:0 叠加即可，PNG 透明区域不遮挡视频。
+
+    Args:
+        title_config: 标题配置 dict（仅用来判断降级）
+        output_width: 输出宽度（未使用，PNG 已按该分辨率绘制）
+        output_height: 输出高度（未使用）
+        title_png_path: 已保存到本地的标题 PNG 文件路径
+        title_input_label: 标题 PNG 在 filter_complex 中的输入标签（默认 "[1:v]"）
+        base_label: 前序滤镜输出标签（如 B-roll 输出 "[vout]"）
+        output_label: overlay 输出标签名
+
+    Returns:
+        overlay 滤镜字符串；title_png_path 为空/文件不存在时返回 None（降级到 drawtext）
+    """
+    import os
+
+    if not title_png_path or not os.path.isfile(title_png_path):
+        return None
+    if not title_config or not isinstance(title_config, dict):
+        return None
+
+    return f"{base_label}{title_input_label}overlay=0:0[{output_label}]"
+
+
 # ── B-roll 叠加滤镜 ─────────────────────────────────────────────────────────
 
 
@@ -573,7 +631,7 @@ def build_broll_overlay_filter(
     video_duration: float,
     output_width: int = DEFAULT_OUTPUT_WIDTH,
     output_height: int = DEFAULT_OUTPUT_HEIGHT,
-) -> str:
+) -> tuple[str, str | None]:
     """构建 B-roll 叠加滤镜链。
 
     支持两种模式:
@@ -581,121 +639,182 @@ def build_broll_overlay_filter(
     - pip: 在对口型视频上叠加画中画 B-roll
 
     Args:
-        b_roll_segments: B-roll 片段配置列表
+        b_roll_segments: B-roll 片段配置列表（原始顺序，决定 FFmpeg -i 输入顺序）
         video_duration: 对口型视频总时长(秒)
-        output_width: 输出宽度
-        output_height: 输出高度
+        output_width: 输出宽度（默认 1280；AI 数字人竖屏传 720）
+        output_height: 输出高度（默认 720；AI 数字人竖屏传 1280）
 
     Returns:
-        FFmpeg filter_complex 滤镜字符串片段
+        (filter_complex_str, final_label)
+        - filter_complex_str: filter_complex 片段字符串（末尾无分号）
+        - final_label: 最终输出 pad 标签名，如 "vout"；无 B-roll 时返回 None
     """
     if not b_roll_segments:
-        return ""
+        return "", None
+
+    # 建立原始列表下标 → FFmpeg 输入下标的映射：
+    # cmd 中 [0:v] 是主视频，随后按 b_roll_segments 原始顺序追加 -i，
+    # 因此第 i 个 segment 的输入是 [{i+1}:v]
+    def _input_label(seg: dict[str, Any]) -> str:
+        # seg 必须来自 b_roll_segments；通过 id() 在原列表中查找
+        for i, s in enumerate(b_roll_segments):
+            if s is seg:
+                return f"[{i + 1}:v]"
+        # fallback: 找不到时不应发生，保守返回
+        return "[1:v]"
 
     parts: list[str] = []
     sorted_segments = sorted(b_roll_segments, key=lambda s: s.get("start_time", 0))
 
-    # 按模式分组处理
+    # 按模式分组
     fullscreen_segments = [s for s in sorted_segments if s.get("mode") == "fullscreen"]
     pip_segments = [s for s in sorted_segments if s.get("mode") == "pip"]
 
+    final_label = None
+
     # ── fullscreen 模式: 切分 + concat ──
     if fullscreen_segments:
-        parts.append(_build_fullscreen_filters(fullscreen_segments, video_duration, output_width, output_height))
+        fs_filter, fs_label = _build_fullscreen_filters(
+            fullscreen_segments, b_roll_segments, video_duration, output_width, output_height, _input_label
+        )
+        parts.append(fs_filter)
+        final_label = fs_label
+    else:
+        fs_label = None
 
     # ── pip 模式: overlay 滤镜 ──
     if pip_segments:
-        for idx, seg in enumerate(pip_segments):
-            start = seg.get("start_time", 0)
-            end = seg.get("end_time", video_duration)
-            scale = seg.get("pip_scale", 0.3)
-            position = seg.get("pip_position", "bottom_right")
-
-            pip_w = int(output_width * scale)
-            pip_h = int(output_height * scale)
-
-            # 位置映射
-            pos_map = {
-                "top_left": "10:10",
-                "top_right": "W-w-10:10",
-                "bottom_left": "10:H-h-10",
-                "bottom_right": "W-w-10:H-h-10",
-                "center": "(W-w)/2:(H-h)/2",
-            }
-            pos_expr = pos_map.get(position, pos_map["bottom_right"])
-
-            broll_input_idx = len(sorted_segments)  # placeholder for input index
-            parts.append(
-                f"[{broll_input_idx + idx}:v]scale={pip_w}:{pip_h}," f"enable='between(t,{start},{end})'[pip{idx}];"
-            )
-            # overlay onto main stream
-            if idx == 0:
-                base_label = "[vout]" if fullscreen_segments else "[0:v]"
-            else:
-                base_label = f"[pip{idx - 1}]"
-            parts.append(f"{base_label}[pip{idx}]overlay={pos_expr}:enable='between(t,{start},{end})'[vout{idx}];")
+        pip_filter, pip_label = _build_pip_filters(
+            pip_segments, output_width, output_height, _input_label, base_label=fs_label
+        )
+        parts.append(pip_filter)
+        final_label = pip_label
 
     result = "".join(parts)
     # 清理末尾多余分号
     if result.endswith(";"):
         result = result[:-1]
-    return result
+    return result, final_label
 
 
 def _build_fullscreen_filters(
-    segments: list[dict[str, Any]],
+    sorted_fs_segments: list[dict[str, Any]],
+    all_segments: list[dict[str, Any]],
     video_duration: float,
     output_width: int,
     output_height: int,
-) -> str:
-    """构建 fullscreen 模式的切分 + concat 滤镜.
+    input_label_fn,
+) -> tuple[str, str]:
+    """构建 fullscreen 模式的切分 + concat 滤镜。
 
-    将对口型视频按 B-roll 时间段切分，然后用 concat 拼接 B-roll 片段。
+    将主视频按 B-roll 时间段切分，然后用 concat 拼接主视频片段和 B-roll 片段。
+
+    Returns:
+        (filter_str, final_label) 其中 final_label 是 concat 输出的 pad 标签
     """
     parts: list[str] = []
     prev_end = 0.0
 
-    for idx, seg in enumerate(segments):
+    # 注意：这里的 idx 是 sorted_fs_segments 中的下标；
+    # 实际 FFmpeg 输入下标必须通过 input_label_fn 查询
+    for idx, seg in enumerate(sorted_fs_segments):
         start = seg.get("start_time", 0)
         end = seg.get("end_time", video_duration)
 
-        # 保持原视频片段（B-roll 之前的部分）
+        # 主视频片段（B-roll 之前）
         if prev_end < start:
             parts.append(f"[0:v]trim=start={prev_end}:end={start},setpts=PTS-STARTPTS[main{idx}];")
 
-        # B-roll 片段：缩放至目标分辨率
+        # B-roll 片段：缩放到输出分辨率并裁到对应时长
+        in_lbl = input_label_fn(seg)
         parts.append(
-            f"[{idx + 1}:v]scale={output_width}:{output_height}"
+            f"{in_lbl}scale={output_width}:{output_height}"
             f":force_original_aspect_ratio=decrease,"
             f"pad={output_width}:{output_height}:(ow-iw)/2:(oh-ih)/2,"
             f"trim=start=0:end={end - start},setpts=PTS-STARTPTS[br{idx}];"
         )
         prev_end = end
 
-    # 尾部片段
+    # 尾部主视频片段
     if prev_end < video_duration:
-        last_idx = len(segments)
+        last_idx = len(sorted_fs_segments)
         parts.append(f"[0:v]trim=start={prev_end}:end={video_duration},setpts=PTS-STARTPTS[main{last_idx}];")
 
     # concat 所有片段
-    segment_labels = []
-    for idx in range(len(segments)):
-        start = segments[idx].get("start_time", 0)
-        if (idx == 0 and segments[0].get("start_time", 0) > 0) or idx > 0:
-            prev_end_prev = segments[idx - 1].get("end_time", 0) if idx > 0 else 0
-            if prev_end_prev < start:
-                segment_labels.append(f"[main{idx}]")
+    segment_labels: list[str] = []
+    for idx, seg in enumerate(sorted_fs_segments):
+        start = seg.get("start_time", 0)
+        # 每段 B-roll 之前是否有主视频片段？
+        has_main_before = (idx == 0 and start > 0) or (
+            idx > 0 and sorted_fs_segments[idx - 1].get("end_time", 0) < start
+        )
+        if has_main_before:
+            segment_labels.append(f"[main{idx}]")
         segment_labels.append(f"[br{idx}]")
-
     if prev_end < video_duration:
-        segment_labels.append(f"[main{len(segments)}]")
+        segment_labels.append(f"[main{len(sorted_fs_segments)}]")
 
+    final_lbl = "vout_fs"
     n = len(segment_labels)
     if n > 0:
         concat_inputs = "".join(segment_labels)
-        parts.append(f"{concat_inputs}concat=n={n}:v=1:a=0[vout];")
+        parts.append(f"{concat_inputs}concat=n={n}:v=1:a=0[{final_lbl}];")
 
-    return "".join(parts)
+    return "".join(parts), final_lbl
+
+
+def _build_pip_filters(
+    pip_segments: list[dict[str, Any]],
+    output_width: int,
+    output_height: int,
+    input_label_fn,
+    base_label: str | None,
+) -> tuple[str, str]:
+    """构建 PIP（画中画）overlay 滤镜链。
+
+    Args:
+        pip_segments: 按时间排序的 pip 片段
+        output_width: 输出宽度
+        output_height: 输出高度
+        input_label_fn: 片段 → 输入标签的映射函数
+        base_label: 前序滤镜链输出的标签（如 fullscreen 的 vout_fs），为 None 则基于 [0:v]
+
+    Returns:
+        (filter_str, final_label)
+    """
+    parts: list[str] = []
+    cur_label = base_label  # 当前叠加到的标签
+
+    pos_map = {
+        "top_left": "10:10",
+        "top_right": "W-w-10:10",
+        "bottom_left": "10:H-h-10",
+        "bottom_right": "W-w-10:H-h-10",
+        "center": "(W-w)/2:(H-h)/2",
+    }
+
+    for idx, seg in enumerate(pip_segments):
+        start = seg.get("start_time", 0)
+        end = seg.get("end_time", 0)
+        scale = seg.get("pip_scale", 0.3)
+        position = seg.get("pip_position", "bottom_right")
+        pos_expr = pos_map.get(position, pos_map["bottom_right"])
+
+        pip_w = max(1, int(output_width * scale))
+        pip_h = max(1, int(output_height * scale))
+        enable_expr = f"enable='between(t,{start},{end})'"
+
+        in_lbl = input_label_fn(seg)
+        pip_scaled = f"pip{idx}"
+        parts.append(f"{in_lbl}scale={pip_w}:{pip_h},{enable_expr}[{pip_scaled}];")
+
+        # overlay onto the current base
+        base = f"[{cur_label}]" if cur_label else "[0:v]"
+        out_lbl = f"vout_pip{idx}" if idx < len(pip_segments) - 1 else "vout"
+        parts.append(f"{base}[{pip_scaled}]overlay={pos_expr}:{enable_expr}[{out_lbl}];")
+        cur_label = out_lbl
+
+    return "".join(parts), cur_label or "vout"
 
 
 def build_cover_extract_command(

requirements.txt

@@ -13,3 +13,7 @@ pytest-cov==6.0.0
 
 # 工具
 python-dotenv==1.0.1
+
+# AI 数字人封面智能选帧（cover_frame_scorer 用 cv2/numpy 做清晰度/亮度/色彩评分）
+numpy==1.26.4
+opencv-python-headless==4.10.0.84

scripts/ci/auto_fix_formatting.py

@@ -1,10 +1,13 @@
 #!/usr/bin/env python3
-"""CI中自动修复代码格式（Python: black + isort | Frontend: prettier），并推送回原分支。
+"""CI中自动修复代码格式（Python: black + isort + ruff | Frontend: prettier），并推送回原分支。
 
 - PR事件：所有PR只要Code Quality因格式问题失败，自动修复并push回源分支
 - Push事件（develop/main）：自动修复并push回原分支，保持主干格式永远正确
 - 防循环：修复commit带 [skip ci-format-check] 标记，检测到该标记则跳过修复
-- 只修格式（black/isort/prettier），ruff逻辑类错误不动
+- black/isort/prettier 修格式；ruff check --fix --unsafe-fixes 自动修复
+  ruff 可修复的 lint 规则（含 F401 未使用 import 等 unsafe fix）
+- ruff 目标范围与 validate_style.sh 的检查范围对齐：apps packages tests
+  （alembic/scripts 不在 ruff 检查范围内，不做修复）
 当code quality检查因格式问题失败时触发。
 """
 
@@ -135,7 +138,7 @@ def get_pr_head_branch(pr_number, api_url, token):
 
 
 def fix_python(target_py_files, scan_mode):
-    """修复 Python 文件格式 (black + isort)"""
+    """修复 Python 文件 (black 格式化 + isort 排序 + ruff lint 自动修复)"""
     if not target_py_files:
         print("没有需要修复的 Python 文件，跳过")
         return
@@ -146,14 +149,48 @@ def fix_python(target_py_files, scan_mode):
     result = run(f"python3 -m black {target_str}", check=False)
     print(result.stdout[-500:] if result.stdout else "")
     if result.returncode != 0:
-        print("black执行失败，但继续尝试isort", file=sys.stderr)
+        print("black执行失败，但继续尝试isort/ruff", file=sys.stderr)
 
     print()
     print("--- isort 排序 ---")
     result = run(f"python3 -m isort {target_str}", check=False)
     print(result.stdout[-500:] if result.stdout else "")
     if result.returncode != 0:
-        print("isort执行失败", file=sys.stderr)
+        print("isort执行失败，继续尝试ruff", file=sys.stderr)
+
+    # ruff lint 自动修复
+    # 与 validate_style.sh 的检查范围对齐：只修 apps/packages/tests
+    # （alembic 在 pyproject.toml 中被 exclude，scripts 不在 ruff 检查范围内）
+    ruff_scopes = ("apps/", "packages/", "tests/")
+    ruff_files = [f for f in target_py_files if f.startswith(ruff_scopes)]
+    if scan_mode != "incremental":
+        ruff_targets = "apps packages tests"
+    elif ruff_files:
+        ruff_targets = " ".join(ruff_files)
+    else:
+        ruff_targets = ""
+
+    if ruff_targets:
+        # ruff 由 style job 的 requirements-dev.txt 安装；不可用时跳过（不阻断 black/isort 的修复）
+        avail = run("python3 -m ruff --version", check=False)
+        if avail.returncode != 0:
+            print("ruff 不可用，跳过 ruff 自动修复", file=sys.stderr)
+        else:
+            print()
+            print("--- ruff lint 自动修复 (--fix --unsafe-fixes) ---")
+            # --unsafe-fixes 用于启用 F401（未使用 import）等 ruff 归类为 unsafe 的自动修复；
+            # 安全性由修复后重跑的完整 CI（单测/构建/staging 健康检查）兜底
+            result = run(
+                f"python3 -m ruff check {ruff_targets} --fix --unsafe-fixes",
+                check=False,
+            )
+            print(result.stdout[-1500:] if result.stdout else "")
+            if result.returncode != 0:
+                # 可能是仍有不可自动修复的 lint 错误（留待 style check 再次拦截），或修复过程出错
+                print("ruff 自动修复后仍有未修复项或执行失败，剩余问题由 style check 继续拦截", file=sys.stderr)
+    else:
+        print()
+        print("增量模式且无 apps/packages/tests 范围内的 Python 变更，跳过 ruff 自动修复")
 
 
 def fix_frontend(target_fe_files, scan_mode, repo_root):
@@ -334,7 +371,7 @@ def main():
     # 提交修复
     run("git clean -fd")
     run("git add -u")
-    run('git commit -m "style: auto-format with black + isort + prettier [skip ci-format-check]"')
+    run('git commit -m "style: auto-format with black + isort + ruff + prettier [skip ci-format-check]"')
 
     # 推送（head_branch已从ensure_git_repo获取）
     print(f"\nPR来源分支: {head_branch}")

tests/unit/domain/test_generated_video.py

@@ -87,28 +87,19 @@ class TestGeneratedVideoCreate:
         assert v.file_url == "http://x/v"
 
     def test_create_empty_project_id(self):
-        """空 project_id 无效."""
-        try:
-            GeneratedVideo.create("", "t1", "v", "http://x/v")
-            assert False
-        except ValueError as e:
-            assert "project_id" in str(e)
+        """空 project_id 允许（AI数字人无项目场景）."""
+        v = GeneratedVideo.create("", "t1", "v", "http://x/v")
+        assert v.project_id == ""
 
     def test_create_whitespace_project_id(self):
-        """纯空白 project_id 无效."""
-        try:
-            GeneratedVideo.create("   ", "t1", "v", "http://x/v")
-            assert False
-        except ValueError as e:
-            assert "project_id" in str(e)
+        """纯空白 project_id 归一化为空串."""
+        v = GeneratedVideo.create("   ", "t1", "v", "http://x/v")
+        assert v.project_id == ""
 
     def test_create_empty_task_id(self):
-        """空 generation_task_id 无效."""
-        try:
-            GeneratedVideo.create("p1", "", "v", "http://x/v")
-            assert False
-        except ValueError as e:
-            assert "generation_task_id" in str(e)
+        """空 generation_task_id 允许."""
+        v = GeneratedVideo.create("p1", "", "v", "http://x/v")
+        assert v.generation_task_id == ""
 
     def test_create_empty_name(self):
         """空 name 无效."""

tests/unit/test_ai_avatar_emotion_tts_lipsync.py

@@ -0,0 +1,365 @@
+"""#1822 情绪/语速透传 + 对口型 TTS 直生 + 智能封面 单元测试.
+
+CI 增量映射:
+  cosyvoice_service.normalize_emotion / payload emotion
+  lipsync_service TTS 直生分支（voice_id+script_text）
+  ai_avatar_cover_service 智能选帧
+"""
+
+import os
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+os.environ.setdefault("JWT_SECRET_KEY", "dev-secret-key-for-testing")
+
+
+# ── 情绪归一化 ──────────────────────────────────────────────────────────
+
+
+def test_normalize_emotion_english_values():
+    from packages.application.cosyvoice_service import normalize_emotion
+
+    assert normalize_emotion("natural") == "natural"
+    assert normalize_emotion("excited") == "excited"
+    assert normalize_emotion("calm") == "calm"
+    assert normalize_emotion("friendly") == "friendly"
+    # 大小写 / 空白容错
+    assert normalize_emotion("  Excited  ") == "excited"
+
+
+def test_normalize_emotion_chinese_values():
+    from packages.application.cosyvoice_service import normalize_emotion
+
+    assert normalize_emotion("自然") == "natural"
+    assert normalize_emotion("兴奋") == "excited"
+    assert normalize_emotion("沉稳") == "calm"
+    assert normalize_emotion("亲切") == "friendly"
+
+
+def test_normalize_emotion_invalid_returns_empty():
+    from packages.application.cosyvoice_service import normalize_emotion
+
+    assert normalize_emotion("") == ""
+    assert normalize_emotion("angry") == ""
+    assert normalize_emotion("喜怒哀乐") == ""
+
+
+# ── CosyVoice payload 携带 emotion + rate ──────────────────────────────
+
+
+def _make_service_with_captured_client(captured: dict):
+    """构造 CosyVoiceService，拦截 post 请求体到 captured['json']."""
+    import httpx as _httpx
+
+    from packages.application import cosyvoice_service as mod
+
+    mock_client = MagicMock(spec=_httpx.Client)
+    resp = MagicMock()
+    resp.status_code = 200
+    resp.json.return_value = {
+        "request_id": "req-1",
+        "output": {"audio": {"url": "https://tts/a.mp3", "duration": 1.0}},
+    }
+    resp.raise_for_status = MagicMock()
+
+    def fake_request(method, url, headers, json, timeout):
+        captured["json"] = json
+        return resp
+
+    mock_client.request.side_effect = fake_request
+
+    with patch.object(mod, "get_shared_settings") as settings_patch:
+        s = MagicMock()
+        s.cosyvoice_api_key = "sk-test"
+        s.cosyvoice_base_url = "https://x/api/v1"
+        s.cosyvoice_model = "cosyvoice-v3-flash"
+        s.cosyvoice_clone_model = "voice-enrollment"
+        s.cosyvoice_format = "mp3"
+        s.cosyvoice_sample_rate = 22050
+        s.cosyvoice_voice = "longxiaochun"
+        settings_patch.return_value = s
+        svc = mod.CosyVoiceService(http_client=mock_client)
+    return svc
+
+
+def test_submit_synthesize_payload_includes_emotion_and_rate():
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="你好", voice_id="v-1", speed=1.5, emotion="兴奋")
+
+    inp = captured["json"]["input"]
+    assert inp["emotion"] == "excited"
+    assert inp["rate"] == 1.5
+
+
+def test_submit_synthesize_payload_omits_emotion_when_empty():
+    captured: dict = {}
+    svc = _make_service_with_captured_client(captured)
+
+    svc.submit_synthesize_task(text="你好", voice_id="v-1")
+
+    assert "emotion" not in captured["json"]["input"]
+
+
+# ── 对口型 TTS 直生分支 ─────────────────────────────────────────────────
+
+
+def _lipsync_service_with_mocks():
+    from app.services.lipsync_service import LipsyncService
+
+    db = MagicMock()
+    client = MagicMock()
+    client.is_available = True
+    client.submit_lipsync.return_value = {
+        "success": True,
+        "task_id": "mk-1",
+        "request_id": "req-1",
+    }
+    cosy = MagicMock()
+    cosy.submit_synthesize_task.return_value = {
+        "audio_url": "https://tts/raw.mp3",
+        "request_id": "tts-req",
+        "audio_duration": 3.0,
+    }
+    svc = LipsyncService(db, client=client, cosyvoice_service=cosy, voice_clone_repo=MagicMock())
+    # _resolve_voice_id 默认原样返回（repo.get 返回 None）
+    svc._voice_clone_repo.get.return_value = None
+    return svc, client, cosy
+
+
+def test_create_job_tts_direct_mode_synthesizes_audio():
+    svc, client, cosy = _lipsync_service_with_mocks()
+
+    with patch("app.services.lipsync_service.tts_synthesize_and_submit") as mock_task:
+        mock_task.apply_async.return_value = MagicMock(id="celery-task-123")
+
+        job = svc.create_job(
+            user_id="user-1",
+            video_url="https://oss/person.mp4",
+            voice_id="cosy-v1",
+            script_text="你好世界",
+            speed=1.2,
+            emotion="兴奋",
+        )
+
+    # v4: TTS 模式下 create_job 返回 tts_processing 状态，dispatch Celery 任务
+    assert job.status == "tts_processing"
+    assert job.emotion == "excited"
+    assert job.speed == 1.2
+    # 不直接调用 CosyVoice（由 Celery 任务处理）
+    cosy.submit_synthesize_task.assert_not_called()
+    # 不直接提交 MediaKit（由 Celery 任务处理）
+    client.submit_lipsync.assert_not_called()
+    # dispatch 了 Celery 任务
+    mock_task.apply_async.assert_called_once()
+
+
+def test_create_job_direct_audio_mode_skips_tts():
+    svc, client, cosy = _lipsync_service_with_mocks()
+
+    job = svc.create_job(
+        user_id="user-1",
+        video_url="https://oss/person.mp4",
+        audio_url="https://oss/ready.mp3",
+    )
+
+    cosy.submit_synthesize_task.assert_not_called()
+    _, submit_kwargs = client.submit_lipsync.call_args
+    assert submit_kwargs["audio_url"] == "https://oss/ready.mp3"
+
+
+def test_create_job_tts_failure_raises():
+    """v4: TTS 模式下 create_job 不再同步失败，而是 dispatch Celery 任务。
+    TTS 合成失败由 Celery 任务内部处理并更新 job 状态。"""
+    svc, client, cosy = _lipsync_service_with_mocks()
+
+    with patch("app.services.lipsync_service.tts_synthesize_and_submit") as mock_task:
+        mock_task.apply_async.return_value = MagicMock(id="celery-task-456")
+
+        job = svc.create_job(
+            user_id="user-1",
+            video_url="https://oss/person.mp4",
+            voice_id="v-1",
+            script_text="文本",
+        )
+
+    # create_job 成功返回 tts_processing，不直接调用 TTS
+    assert job.status == "tts_processing"
+    cosy.submit_synthesize_task.assert_not_called()
+    client.submit_lipsync.assert_not_called()
+    mock_task.apply_async.assert_called_once()
+
+
+# ── refresh 同步中间状态 ────────────────────────────────────────────────
+
+
+def test_refresh_syncs_running_status():
+    from app.services.lipsync_service import LipsyncService
+
+    db = MagicMock()
+    client = MagicMock()
+    client.get_task_status.return_value = {"success": True, "status": "running"}
+    svc = LipsyncService(db, client=client)
+
+    job = MagicMock()
+    job.status = "submitted"
+    job.mediakit_task_id = "mk-1"
+    job.id = "j-1"
+    svc.get_job = MagicMock(return_value=job)
+
+    result = svc.refresh_job_status("j-1", "user-1")
+    assert result.status == "running"
+
+
+# ── 智能封面 ────────────────────────────────────────────────────────────
+
+
+def test_smart_cover_selects_best_frame_and_persists():
+    from app.services import ai_avatar_cover_service as cov
+
+    snapshots = [
+        {"image_url": "https://mk/f0.jpg"},
+        {"image_url": "https://mk/f1.jpg"},
+    ]
+    with (
+        patch("packages.shared.mediakit_client.get_mediakit_client") as mk_patch,
+        patch("packages.shared.cover_frame_scorer.score_frames") as score_patch,
+        patch("httpx.Client") as http_client_cls,
+        patch("packages.shared.storage.get_shared_storage_service") as storage_patch,
+    ):
+        mk = MagicMock()
+        mk.is_available = True
+        mk.extract_frames.return_value = snapshots
+        mk_patch.return_value = mk
+        # score_frames 把 f1 选为最佳
+        score_patch.side_effect = lambda cands: [
+            {"url": "https://mk/f1.jpg", "score": 90.0, "image_path": cands[1]["image_path"]},
+            {"url": "https://mk/f0.jpg", "score": 60.0, "image_path": cands[0]["image_path"]},
+        ]
+        # httpx.Client 连接池 mock
+        client_instance = MagicMock()
+        resp = MagicMock()
+        resp.content = b"IMGDATA"
+        resp.raise_for_status = MagicMock()
+        client_instance.get.return_value = resp
+        client_instance.__enter__ = MagicMock(return_value=client_instance)
+        client_instance.__exit__ = MagicMock(return_value=False)
+        http_client_cls.return_value = client_instance
+
+        storage = MagicMock()
+        storage.public_url = "https://oss.example.com"
+        # video_url 不是自家 OSS，不重签
+        storage.get_download_url.side_effect = lambda url, **kw: f"{url}?signed=1"
+        storage.upload_file.return_value = "https://oss.example.com/cover.jpg"
+        storage_patch.return_value = storage
+
+        url = cov.generate_smart_cover("https://other-host/avatar.mp4", job_id="job-1")
+
+    assert "signed=1" in url or url == "https://oss.example.com/cover.jpg"
+    mk.extract_frames.assert_called_once()
+    score_patch.assert_called_once()
+    # 验证使用了增大的轮询参数
+    call_kwargs = mk.extract_frames.call_args
+    assert call_kwargs.kwargs.get("poll_interval") == 1.0 or call_kwargs[1].get("poll_interval") == 1.0
+    assert call_kwargs.kwargs.get("max_poll_attempts") == 15 or call_kwargs[1].get("max_poll_attempts") == 15
+
+
+def test_smart_cover_returns_empty_when_mediakit_unavailable():
+    from app.services import ai_avatar_cover_service as cov
+
+    with (
+        patch("packages.shared.mediakit_client.get_mediakit_client") as mk_patch,
+        patch("packages.shared.storage.get_shared_storage_service") as storage_patch,
+    ):
+        mk = MagicMock()
+        mk.is_available = False
+        mk_patch.return_value = mk
+        storage = MagicMock()
+        storage.public_url = "https://oss.example.com"
+        storage_patch.return_value = storage
+        url = cov.generate_smart_cover("https://oss.example.com/avatar.mp4")
+    assert url == ""
+
+
+def test_sign_video_url_resigns_own_oss_url():
+    """自家 OSS 私有桶 URL 应被重签为长有效期预签名 URL"""
+    from app.services.ai_avatar_cover_service import _sign_video_url_for_mediakit
+
+    with patch("packages.shared.storage.get_shared_storage_service") as storage_patch:
+        storage = MagicMock()
+        storage.public_url = "https://oss.example.com"
+        storage.get_download_url.return_value = "https://oss.example.com/file.mp4?Expires=xxx&Signature=yyy"
+        storage_patch.return_value = storage
+
+        result = _sign_video_url_for_mediakit("https://oss.example.com/file.mp4")
+
+    assert "Signature=yyy" in result
+    storage.get_download_url.assert_called_once()
+
+
+def test_sign_video_url_skips_external_url():
+    """外部 URL（非自家 OSS）应原样返回，不做重签"""
+    from app.services.ai_avatar_cover_service import _sign_video_url_for_mediakit
+
+    with patch("packages.shared.storage.get_shared_storage_service") as storage_patch:
+        storage = MagicMock()
+        storage.public_url = "https://oss.example.com"
+        storage_patch.return_value = storage
+
+        result = _sign_video_url_for_mediakit("https://external-cdn.com/video.mp4")
+
+    assert result == "https://external-cdn.com/video.mp4"
+    storage.get_download_url.assert_not_called()
+
+
+def test_extract_frames_uses_extended_poll_params():
+    """验证 select_best_cover_frame 使用增大后的轮询参数"""
+    from app.services import ai_avatar_cover_service as cov
+
+    with (
+        patch("packages.shared.mediakit_client.get_mediakit_client") as mk_patch,
+        patch("packages.shared.storage.get_shared_storage_service") as storage_patch,
+    ):
+        mk = MagicMock()
+        mk.is_available = True
+        mk.extract_frames.return_value = [{"image_url": "https://mk/f0.jpg"}]
+        mk_patch.return_value = mk
+
+        storage = MagicMock()
+        storage.public_url = "https://oss.example.com"
+        storage_patch.return_value = storage
+
+        cov.select_best_cover_frame("https://other/avatar.mp4", max_frames=3)
+
+    call_kwargs = mk.extract_frames.call_args
+    assert call_kwargs.kwargs.get("poll_interval") == 1.0 or call_kwargs[1].get("poll_interval") == 1.0
+    assert call_kwargs.kwargs.get("max_poll_attempts") == 15 or call_kwargs[1].get("max_poll_attempts") == 15
+    assert call_kwargs.kwargs.get("max_retries") == 1 or call_kwargs[1].get("max_retries") == 1
+
+
+# ── 渲染 script_id 可选（手动文案直生场景）──────────────────────────────
+
+
+def test_render_request_script_id_optional():
+    import os
+    import sys
+
+    sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", "..", "apps", "api"))
+    from app.schemas.ai_avatar_render import CreateAiAvatarRenderRequest
+
+    # 手动文案直生：不传 script_id 也合法
+    req = CreateAiAvatarRenderRequest(lipsync_job_id="j-1")
+    assert req.script_id == ""
+
+    # 空白被 strip
+    req2 = CreateAiAvatarRenderRequest(lipsync_job_id="j-1", script_id="   ")
+    assert req2.script_id == ""
+
+    # title_config 是单个 dict
+    req3 = CreateAiAvatarRenderRequest(
+        lipsync_job_id="j-1",
+        title_config={"text": "标题", "position": "top", "font_size": 40},
+    )
+    assert req3.title_config["position"] == "top"

tests/unit/test_ai_avatar_render_routes.py

@@ -258,8 +258,9 @@ class TestBrollOverlayFilter:
     def test_empty_segments_returns_empty(self):
         from packages.domain.video_filter_builder import build_broll_overlay_filter
 
-        result = build_broll_overlay_filter([], 30.0)
+        result, label = build_broll_overlay_filter([], 30.0)
         assert result == ""
+        assert label is None
 
     def test_pip_mode_generates_overlay(self):
         from packages.domain.video_filter_builder import build_broll_overlay_filter
@@ -275,8 +276,9 @@ class TestBrollOverlayFilter:
                 "pip_scale": 0.3,
             }
         ]
-        result = build_broll_overlay_filter(segments, 30.0)
+        result, label = build_broll_overlay_filter(segments, 30.0)
         assert "overlay" in result or "scale=" in result
+        assert label == "vout"
 
     def test_fullscreen_mode_generates_concat(self):
         from packages.domain.video_filter_builder import build_broll_overlay_filter
@@ -290,8 +292,9 @@ class TestBrollOverlayFilter:
                 "end_time": 10.0,
             }
         ]
-        result = build_broll_overlay_filter(segments, 30.0)
+        result, label = build_broll_overlay_filter(segments, 30.0)
         assert "trim" in result or "concat" in result
+        assert label == "vout_fs"
 
     def test_cover_extract_command(self):
         from packages.domain.video_filter_builder import build_cover_extract_command
@@ -315,3 +318,120 @@ class TestBrollOverlayFilter:
             "/tmp/cover.jpg",
         )
         assert "scale=" in cmd
+
+
+def _make_mock_auth_user(user_id="user-1"):
+    """构造 AuthenticatedUser：current_user.user.id."""
+    auth = MagicMock()
+    auth.user.id = user_id
+    return auth
+
+
+class TestRenderSmartCoverRoute:
+    """POST /renders/{job_id}/smart-cover — 从成片智能抽封面（步骤②）."""
+
+    def test_smart_cover_job_not_found_returns_404(self):
+        """渲染任务不存在 → 404."""
+        from app.api.routes.ai_avatar_render import generate_render_smart_cover
+        from fastapi import HTTPException
+
+        mock_service = MagicMock()
+        mock_service.get_render_job.return_value = None
+        mock_db = MagicMock()
+        mock_user = _make_mock_auth_user()
+
+        # 函数内部 `from app.services.ai_avatar_render_service import AiAvatarRenderService`
+        with patch("app.services.ai_avatar_render_service.AiAvatarRenderService", return_value=mock_service):
+            with pytest.raises(HTTPException) as exc_info:
+                generate_render_smart_cover(job_id="render-missing", current_user=mock_user, db=mock_db)
+
+        assert exc_info.value.status_code == 404
+        assert "不存在" in exc_info.value.detail
+        mock_service.get_render_job.assert_called_once_with("render-missing", "user-1")
+
+    def test_smart_cover_job_not_completed_returns_400(self):
+        """任务未 completed（如 processing）→ 400."""
+        from app.api.routes.ai_avatar_render import generate_render_smart_cover
+        from fastapi import HTTPException
+
+        mock_service = MagicMock()
+        mock_job = _make_mock_render_job(status="processing", output_video_url="https://oss/video.mp4")
+        mock_service.get_render_job.return_value = mock_job
+        mock_db = MagicMock()
+        mock_user = _make_mock_auth_user()
+
+        with patch("app.services.ai_avatar_render_service.AiAvatarRenderService", return_value=mock_service):
+            with pytest.raises(HTTPException) as exc_info:
+                generate_render_smart_cover(job_id="render-1", current_user=mock_user, db=mock_db)
+
+        assert exc_info.value.status_code == 400
+        assert "先完成视频生成" in exc_info.value.detail
+
+    def test_smart_cover_empty_video_url_returns_400(self):
+        """已 completed 但 output_video_url 为空/空白 → 400."""
+        from app.api.routes.ai_avatar_render import generate_render_smart_cover
+        from fastapi import HTTPException
+
+        mock_service = MagicMock()
+        mock_job = _make_mock_render_job(status="completed", output_video_url="   ")
+        mock_service.get_render_job.return_value = mock_job
+        mock_db = MagicMock()
+        mock_user = _make_mock_auth_user()
+
+        with patch("app.services.ai_avatar_render_service.AiAvatarRenderService", return_value=mock_service):
+            with pytest.raises(HTTPException) as exc_info:
+                generate_render_smart_cover(job_id="render-1", current_user=mock_user, db=mock_db)
+
+        assert exc_info.value.status_code == 400
+        assert "URL 为空" in exc_info.value.detail
+
+    def test_smart_cover_success_updates_db_and_returns_url(self):
+        """抽帧成功 → 更新 job.cover_config / output_cover_url 并 commit，返回 completed."""
+        from app.api.routes.ai_avatar_render import generate_render_smart_cover
+
+        mock_service = MagicMock()
+        mock_job = _make_mock_render_job(
+            status="completed",
+            output_video_url="https://oss/final.mp4",
+        )
+        mock_job.cover_config = {"mode": "manual"}
+        mock_service.get_render_job.return_value = mock_job
+        mock_db = MagicMock()
+        mock_user = _make_mock_auth_user()
+
+        with (
+            patch("app.services.ai_avatar_render_service.AiAvatarRenderService", return_value=mock_service),
+            patch(
+                "app.api.routes.ai_avatar_render.generate_smart_cover", return_value="https://oss/cover.jpg"
+            ) as mock_gen,
+        ):
+            result = generate_render_smart_cover(job_id="render-1", current_user=mock_user, db=mock_db)
+
+        mock_gen.assert_called_once_with("https://oss/final.mp4", job_id="render-1", max_frames=5)
+        assert result.status == "completed"
+        assert result.cover_url == "https://oss/cover.jpg"
+        assert mock_job.output_cover_url == "https://oss/cover.jpg"
+        assert mock_job.cover_config["mode"] == "auto_frame"
+        assert mock_job.cover_config["url"] == "https://oss/cover.jpg"
+        mock_db.commit.assert_called_once()
+
+    def test_smart_cover_extract_failure_returns_fallback_failed(self):
+        """generate_smart_cover 抛异常 → fallback_failed，不抛错不写 DB."""
+        from app.api.routes.ai_avatar_render import generate_render_smart_cover
+
+        mock_service = MagicMock()
+        mock_job = _make_mock_render_job(status="completed", output_video_url="https://oss/final.mp4")
+        mock_service.get_render_job.return_value = mock_job
+        mock_db = MagicMock()
+        mock_user = _make_mock_auth_user()
+
+        with (
+            patch("app.services.ai_avatar_render_service.AiAvatarRenderService", return_value=mock_service),
+            patch("app.api.routes.ai_avatar_render.generate_smart_cover", side_effect=RuntimeError("mediakit down")),
+        ):
+            result = generate_render_smart_cover(job_id="render-1", current_user=mock_user, db=mock_db)
+
+        assert result.status == "fallback_failed"
+        assert result.cover_url == ""
+        # 失败时不写 cover_config / 不 commit
+        mock_db.commit.assert_not_called()

tests/unit/test_ai_avatar_render_service.py

@@ -176,13 +176,23 @@ class TestSchemaValidation:
                 script_id="script-1",
             )
 
-    def test_create_request_empty_script_id(self):
+    def test_create_request_empty_script_id_normalized(self):
+        """script_id 改为可选（TTS 直生场景）：空白值应规范化为空串而非抛错。"""
         from app.schemas.ai_avatar_render import CreateAiAvatarRenderRequest
 
-        with pytest.raises(ValueError, match="script_id 不能为空"):
+        req = CreateAiAvatarRenderRequest(
+            lipsync_job_id="lipsync-1",
+            script_id="  ",
+        )
+        assert req.script_id == ""
+
+    def test_create_request_empty_lipsync_job_id_raises(self):
+        from app.schemas.ai_avatar_render import CreateAiAvatarRenderRequest
+
+        with pytest.raises(ValueError, match="lipsync_job_id 不能为空"):
             CreateAiAvatarRenderRequest(
-                lipsync_job_id="lipsync-1",
-                script_id="  ",
+                lipsync_job_id="  ",
+                script_id="script-1",
             )
 
 
@@ -509,9 +519,167 @@ class TestAiAvatarRenderService:
         # 不应执行渲染逻辑
         mock_db.commit.assert_not_called()
 
+    def test_execute_render_success_creates_clip_record(self):
+        """execute_render 完成后自动创建成片记录到成片库."""
+        from app.services.ai_avatar_render_service import AiAvatarRenderService
+
+        mock_db = _make_mock_db()
+        mock_lipsync_job = _make_mock_lipsync_job(
+            status="completed",
+            output_video_url="https://oss/lipsync.mp4",
+            output_duration=30.0,
+        )
+        mock_job = _make_mock_render_job(
+            job_id="render-ok",
+            status="pending",
+            output_video_url="",
+            output_cover_url="",
+            output_duration=0.0,
+        )
+        mock_filter = MagicMock()
+        mock_filter.first.side_effect = [mock_job, mock_lipsync_job]
+        mock_query = MagicMock()
+        mock_query.filter.return_value = mock_filter
+        mock_db.query.return_value = mock_query
+
+        svc = AiAvatarRenderService(mock_db)
+
+        with (
+            patch.object(svc, "_download_video", return_value="/tmp/video.mp4"),
+            patch.object(svc, "_upload_to_oss", side_effect=lambda path, key: f"https://oss/{key}"),
+            patch("subprocess.run") as mock_run,
+            patch("tempfile.TemporaryDirectory") as tmpdir_mock,
+            patch(
+                "app.services.ai_avatar_cover_service.generate_smart_cover", return_value="https://oss/smart_cover.jpg"
+            ),
+            patch("packages.domain.generated_video.GeneratedVideo.create") as gv_create,
+            patch(
+                "packages.adapters.sqlalchemy_impl.generated_video_repository.SQLAlchemyGeneratedVideoRepository"
+            ) as repo_cls,
+        ):
+            import subprocess as _sp
+
+            mock_run.return_value = _sp.CompletedProcess(args=[], returncode=0, stdout="", stderr="")
+            import tempfile as _tf
+
+            tmpdir_mock.return_value.__enter__ = MagicMock(return_value="/tmp/testdir")
+            tmpdir_mock.return_value.__exit__ = MagicMock(return_value=False)
+
+            mock_clip = MagicMock()
+            mock_clip.id = "clip-001"
+            gv_create.return_value = mock_clip
+            mock_repo = MagicMock()
+            repo_cls.return_value = mock_repo
+
+            svc.execute_render("render-ok")
+
+        assert mock_job.status == "completed"
+        gv_create.assert_called_once()
+        call_kwargs = gv_create.call_args
+        assert "https://oss/" in call_kwargs.kwargs["file_url"]
+        assert call_kwargs.kwargs["user_id"] == "user-1"
+        mock_repo.create.assert_called_once_with(mock_clip)
+
+    def test_execute_render_clip_failure_does_not_affect_render(self):
+        """成片创建失败不影响渲染任务标记为成功."""
+        from app.services.ai_avatar_render_service import AiAvatarRenderService
+
+        mock_db = _make_mock_db()
+        mock_lipsync_job = _make_mock_lipsync_job(
+            status="completed",
+            output_video_url="https://oss/lipsync.mp4",
+            output_duration=30.0,
+        )
+        mock_job = _make_mock_render_job(
+            job_id="render-clip-fail",
+            status="pending",
+            output_video_url="",
+            output_cover_url="",
+            output_duration=0.0,
+        )
+        mock_filter = MagicMock()
+        mock_filter.first.side_effect = [mock_job, mock_lipsync_job]
+        mock_query = MagicMock()
+        mock_query.filter.return_value = mock_filter
+        mock_db.query.return_value = mock_query
+
+        svc = AiAvatarRenderService(mock_db)
+
+        with (
+            patch.object(svc, "_download_video", return_value="/tmp/video.mp4"),
+            patch.object(svc, "_upload_to_oss", side_effect=lambda path, key: f"https://oss/{key}"),
+            patch("subprocess.run") as mock_run,
+            patch("tempfile.TemporaryDirectory") as tmpdir_mock,
+            patch("app.services.ai_avatar_cover_service.generate_smart_cover", side_effect=RuntimeError("DB error")),
+        ):
+            import subprocess as _sp
+
+            mock_run.return_value = _sp.CompletedProcess(args=[], returncode=0, stdout="", stderr="")
+            tmpdir_mock.return_value.__enter__ = MagicMock(return_value="/tmp/testdir")
+            tmpdir_mock.return_value.__exit__ = MagicMock(return_value=False)
+            svc.execute_render("render-clip-fail")
+
+        # 即使成片创建失败，渲染任务仍应标记为 completed
+        assert mock_job.status == "completed"
+
     def test_error_exception_has_code(self):
         from app.services.ai_avatar_render_service import AiAvatarRenderError
 
         err = AiAvatarRenderError("测试错误", code="TestCode")
         assert err.code == "TestCode"
         assert str(err) == "测试错误"
+
+
+class TestAiAvatarRenderCoverPassthrough:
+    """execute_render 中封面透传逻辑（320~329 行）：cover_config 含 url/imageUrl/cover_url 时直接透传到 output_cover_url."""
+
+    def _run_execute(self, mock_job, mock_lipsync_job):
+        """驱动 execute_render 跑到完成阶段的通用脚手架（mock IO 部分）."""
+        from app.services.ai_avatar_render_service import AiAvatarRenderService
+
+        mock_db = _make_mock_db()
+        mock_filter = MagicMock()
+        # query.filter 返回同一个 filter 两次（render_job 查询、lipsync 查询）
+        mock_filter.first.side_effect = [mock_job, mock_lipsync_job]
+        mock_query = MagicMock()
+        mock_query.filter.return_value = mock_filter
+        mock_db.query.return_value = mock_query
+
+        svc = AiAvatarRenderService(mock_db)
+        with (
+            patch.object(svc, "_download_video", return_value="/tmp/video.mp4"),
+            patch.object(svc, "_upload_to_oss", side_effect=lambda path, key: f"https://oss/{key}"),
+            patch("subprocess.run") as mock_run,
+            patch("tempfile.TemporaryDirectory") as tmpdir_mock,
+            patch("app.services.ai_avatar_cover_service.generate_smart_cover", return_value=""),
+            patch("packages.domain.generated_video.GeneratedVideo.create", return_value=MagicMock()),
+            patch(
+                "packages.adapters.sqlalchemy_impl.generated_video_repository.SQLAlchemyGeneratedVideoRepository"
+            ) as repo_cls,
+        ):
+            import subprocess as _sp
+
+            mock_run.return_value = _sp.CompletedProcess(args=[], returncode=0, stdout="", stderr="")
+            import tempfile as _tf
+
+            tmpdir_mock.return_value.__enter__ = MagicMock(return_value="/tmp/testdir")
+            tmpdir_mock.return_value.__exit__ = MagicMock(return_value=False)
+            repo_cls.return_value = MagicMock()
+            svc.execute_render(mock_job.id)
+        return mock_db, mock_job
+
+    def test_cover_url_in_cover_config_passthrough_to_output_cover(self):
+        """cover_config.url 存在 → 透传到 output_cover_url."""
+        mock_job = _make_mock_render_job(job_id="render-cov-1", status="pending")
+        mock_job.cover_config = {"mode": "upload", "url": "https://oss/user-cover.jpg"}
+        mock_lipsync_job = _make_mock_lipsync_job(status="completed", output_duration=10.0)
+        _, job = self._run_execute(mock_job, mock_lipsync_job)
+        assert job.output_cover_url == "https://oss/user-cover.jpg"
+
+    def test_cover_imageurl_fallback_also_passthrough(self):
+        """cover_config.imageUrl（老字段）存在 → 也透传到 output_cover_url."""
+        mock_job = _make_mock_render_job(job_id="render-cov-2", status="pending")
+        mock_job.cover_config = {"mode": "upload", "imageUrl": "https://oss/user-cover2.jpg"}
+        mock_lipsync_job = _make_mock_lipsync_job(status="completed", output_duration=10.0)
+        _, job = self._run_execute(mock_job, mock_lipsync_job)
+        assert job.output_cover_url == "https://oss/user-cover2.jpg"

tests/unit/test_cover_frame_scorer.py

@@ -43,7 +43,7 @@ class TestScoreFrame:
 
     @requires_cv2
     def test_clear_image_high_score(self):
-        """清晰、亮度适中、色彩丰富的图像应得高分."""
+        """清晰、亮度适中、色彩丰富的图像应得较高分."""
         # 创建一个清晰的渐变图像（色彩丰富、亮度适中）
         img = np.zeros((100, 100, 3), dtype=np.uint8)
         for i in range(100):
@@ -53,7 +53,8 @@ class TestScoreFrame:
         from packages.shared.cover_frame_scorer import score_frame
 
         score = score_frame(img)
-        assert 50.0 <= score <= 100.0, f"清晰图像应得高分，实际: {score}"
+        # 渐变图清晰度中等+亮度尚可+色彩有变化，分数应明显高于模糊/全黑/全白
+        assert 40.0 <= score <= 100.0, f"清晰图像应得较高分，实际: {score}"
 
     @requires_cv2
     def test_blurry_image_low_clarity(self):
@@ -76,8 +77,8 @@ class TestScoreFrame:
         from packages.shared.cover_frame_scorer import score_frame
 
         score = score_frame(img)
-        # 全黑：清晰度 0，亮度 0，色彩 0
-        assert score <= 5.0, f"全黑图像应接近 0 分，实际: {score}"
+        # 全黑：清晰度 0，亮度偏离130扣约24分，色彩 0 → 得分约0~7，允许cv2内部微小浮点差异
+        assert score <= 10.0, f"全黑图像应接近 0 分，实际: {score}"
 
     @requires_cv2
     def test_bright_image_low_brightness(self):

tests/unit/test_dedup_v2.py

@@ -180,28 +180,26 @@ class TestDetectKeyframeTimestamps:
 
     def test_cannot_open_video_raises(self):
         """无法打开视频时抛出 RuntimeError."""
-        cv2_mock = _dedup_mod.cv2
         mock_cap = MagicMock()
         mock_cap.isOpened.return_value = False
-        cv2_mock.VideoCapture.return_value = mock_cap
 
         import pytest
 
-        with pytest.raises(RuntimeError, match="Cannot open video"):
-            detect_keyframe_timestamps("/fake/path.mp4")
+        with patch.object(_dedup_mod.cv2, "VideoCapture", return_value=mock_cap):
+            with pytest.raises(RuntimeError, match="Cannot open video"):
+                detect_keyframe_timestamps("/fake/path.mp4")
 
     def test_zero_duration_returns_empty(self):
         """视频时长为 0 时返回空列表."""
-        cv2_mock = _dedup_mod.cv2
         mock_cap = MagicMock()
         mock_cap.isOpened.return_value = True
         # cv2.CAP_PROP_FPS etc. are Mock objects; configure get() to return 0 for frame_count
         mock_cap.get.return_value = 0
         mock_cap.read.return_value = (False, None)
-        cv2_mock.VideoCapture.return_value = mock_cap
 
-        result = detect_keyframe_timestamps("/fake/zero.mp4")
-        assert result == []
+        with patch.object(_dedup_mod.cv2, "VideoCapture", return_value=mock_cap):
+            result = detect_keyframe_timestamps("/fake/zero.mp4")
+            assert result == []
 
     def test_function_signature(self):
         """验证函数签名和默认参数."""

tests/unit/test_generated_video.py

@@ -47,32 +47,35 @@ class TestGeneratedVideoCreate:
         assert video.file_url == "https://example.com/video.mp4"
         assert video.user_id == "user1"
 
-    def test_create_empty_project_id_raises(self):
-        with pytest.raises(ValueError, match="project_id cannot be empty"):
-            GeneratedVideo.create(
-                project_id="",
-                generation_task_id="task1",
-                name="视频",
-                file_url="https://example.com/v.mp4",
-            )
+    def test_create_empty_project_id_allowed(self):
+        """project_id 允许为空（AI数字人等无项目场景）。"""
+        video = GeneratedVideo.create(
+            project_id="",
+            generation_task_id="task1",
+            name="视频",
+            file_url="https://example.com/v.mp4",
+        )
+        assert video.project_id == ""
 
-    def test_create_whitespace_project_id_raises(self):
-        with pytest.raises(ValueError, match="project_id cannot be empty"):
-            GeneratedVideo.create(
-                project_id="   ",
-                generation_task_id="task1",
-                name="视频",
-                file_url="https://example.com/v.mp4",
-            )
+    def test_create_whitespace_project_id_normalized_to_empty(self):
+        """project_id 纯空白会被 strip 为空串，不抛异常。"""
+        video = GeneratedVideo.create(
+            project_id="   ",
+            generation_task_id="task1",
+            name="视频",
+            file_url="https://example.com/v.mp4",
+        )
+        assert video.project_id == ""
 
-    def test_create_empty_generation_task_id_raises(self):
-        with pytest.raises(ValueError, match="generation_task_id cannot be empty"):
-            GeneratedVideo.create(
-                project_id="proj1",
-                generation_task_id="",
-                name="视频",
-                file_url="https://example.com/v.mp4",
-            )
+    def test_create_empty_generation_task_id_allowed(self):
+        """generation_task_id 允许为空（兼容部分异步链路）。"""
+        video = GeneratedVideo.create(
+            project_id="proj1",
+            generation_task_id="",
+            name="视频",
+            file_url="https://example.com/v.mp4",
+        )
+        assert video.generation_task_id == ""
 
     def test_create_empty_name_raises(self):
         with pytest.raises(ValueError, match="name cannot be empty"):

tests/unit/test_generated_video_domain.py

@@ -75,32 +75,35 @@ class TestGeneratedVideoCreate:
         assert video.file_url == "https://example.com/out.mp4"
         assert video.user_id == "user_003"
 
-    def test_create_empty_project_id_raises(self):
-        with pytest.raises(ValueError, match="project_id"):
-            GeneratedVideo.create(
-                project_id="",
-                generation_task_id="t",
-                name="n",
-                file_url="u",
-            )
+    def test_create_empty_project_id_allowed(self):
+        """project_id 允许为空（AI数字人等无项目场景）。"""
+        video = GeneratedVideo.create(
+            project_id="",
+            generation_task_id="t",
+            name="n",
+            file_url="u",
+        )
+        assert video.project_id == ""
 
-    def test_create_whitespace_project_id_raises(self):
-        with pytest.raises(ValueError, match="project_id"):
-            GeneratedVideo.create(
-                project_id="   ",
-                generation_task_id="t",
-                name="n",
-                file_url="u",
-            )
+    def test_create_whitespace_project_id_normalized(self):
+        """project_id 纯空白归一化为空串。"""
+        video = GeneratedVideo.create(
+            project_id="   ",
+            generation_task_id="t",
+            name="n",
+            file_url="u",
+        )
+        assert video.project_id == ""
 
-    def test_create_empty_generation_task_id_raises(self):
-        with pytest.raises(ValueError, match="generation_task_id"):
-            GeneratedVideo.create(
-                project_id="p",
-                generation_task_id="",
-                name="n",
-                file_url="u",
-            )
+    def test_create_empty_generation_task_id_allowed(self):
+        """generation_task_id 允许为空。"""
+        video = GeneratedVideo.create(
+            project_id="p",
+            generation_task_id="",
+            name="n",
+            file_url="u",
+        )
+        assert video.generation_task_id == ""
 
     def test_create_empty_name_raises(self):
         with pytest.raises(ValueError, match="name"):

tests/unit/test_generated_video_verification_code.py

@@ -45,25 +45,25 @@ class TestGeneratedVideo:
         assert video.duplicate_of is None
         assert video.generation_params == {}
 
-    def test_create_empty_project_id_raises(self):
-        """空project_id抛异常."""
-        with pytest.raises(ValueError, match="project_id"):
-            GeneratedVideo.create(
-                project_id="  ",
-                generation_task_id="t1",
-                name="v.mp4",
-                file_url="https://x.com/v.mp4",
-            )
+    def test_create_empty_project_id_allowed(self):
+        """project_id 允许为空（AI数字人场景），空白归一化为空串."""
+        video = GeneratedVideo.create(
+            project_id="  ",
+            generation_task_id="t1",
+            name="v.mp4",
+            file_url="https://x.com/v.mp4",
+        )
+        assert video.project_id == ""
 
-    def test_create_empty_task_id_raises(self):
-        """空generation_task_id抛异常."""
-        with pytest.raises(ValueError, match="generation_task_id"):
-            GeneratedVideo.create(
-                project_id="p1",
-                generation_task_id="",
-                name="v.mp4",
-                file_url="https://x.com/v.mp4",
-            )
+    def test_create_empty_task_id_allowed(self):
+        """generation_task_id 允许为空."""
+        video = GeneratedVideo.create(
+            project_id="p1",
+            generation_task_id="",
+            name="v.mp4",
+            file_url="https://x.com/v.mp4",
+        )
+        assert video.generation_task_id == ""
 
     def test_create_empty_name_raises(self):
         """空name抛异常."""

tests/unit/test_lipsync_routes.py

@@ -35,14 +35,14 @@ def mock_mediakit():
 
 @pytest.fixture
 def mock_cosyvoice():
-    """Mock CosyVoice 服务."""
+    """Mock CosyVoice 服务（v3: service 内部走 submit_synthesize_task，返回 dict）."""
     service = MagicMock()
-    service.synthesize_speech.return_value = MagicMock(
-        audio_url="https://oss.example.com/tts-output.mp3",
-        duration=15.0,
-        file_size=12345,
-        request_id="tts-req-789",
-    )
+    service.submit_synthesize_task.return_value = {
+        "audio_url": "https://oss.example.com/tts-output.mp3",
+        "request_id": "tts-req-789",
+    }
+    # synthesize_speech 保留给直接同步调用场景
+    service.synthesize_speech.return_value = MagicMock(audio_url="https://oss.example.com/tts-output.mp3")
     return service
 
 
@@ -159,7 +159,8 @@ class TestSchemaValidation:
             voice_id="longxiaochun_v3",
             script_text="测试文本",
         )
-        assert req.enable_video_loop is False
+        # AI数字人场景文案长度不可控，默认开启视频循环，防止音频长于视频时被截断
+        assert req.enable_video_loop is True
 
     def test_video_url_strip_query_params(self):
         """视频 URL 含查询参数时，扩展名检查应忽略 ? 后面的部分."""
@@ -172,103 +173,129 @@ class TestSchemaValidation:
         )
         assert "?token=" in req.video_url
 
-    def test_no_audio_url_in_request(self):
-        """#1809: 请求体不应包含 audio_url 字段."""
+    def test_dual_mode_fields_present(self):
+        """v3 契约: 双模式——支持直接音频 audio_url，也支持 TTS 直生 voice_id+script_text."""
+        from app.schemas.lipsync import CreateLipsyncJobRequest
+
+        fields = CreateLipsyncJobRequest.model_fields.keys()
+        # 直接音频模式
+        assert "audio_url" in fields
+        # TTS 直生模式
+        assert "voice_id" in fields
+        assert "script_text" in fields
+        # 语速/情绪透传
+        assert "speed" in fields
+        assert "emotion" in fields
+
+    def test_direct_audio_mode_accepted(self):
+        """v3 契约: 只传 audio_url（直接音频模式）也合法，无需 voice_id/script_text."""
         from app.schemas.lipsync import CreateLipsyncJobRequest
 
         req = CreateLipsyncJobRequest(
             video_url="https://example.com/video.mp4",
-            voice_id="longxiaochun_v3",
-            script_text="测试文本",
+            audio_url="https://example.com/audio.mp3",
         )
-        assert not hasattr(req, "audio_url")
-        fields = req.model_fields.keys()
-        assert "audio_url" not in fields
-        assert "voice_id" in fields
-        assert "script_text" in fields
+        assert req.audio_url == "https://example.com/audio.mp3"
+
+    def test_neither_mode_rejected(self):
+        """v3 契约: audio_url 与 voice_id+script_text 都缺时应报错."""
+        from app.schemas.lipsync import CreateLipsyncJobRequest
+
+        with pytest.raises(ValueError):
+            CreateLipsyncJobRequest(video_url="https://example.com/video.mp4")
 
 
 class TestLipsyncServiceUnit:
     """Service 层单元测试（纯 mock，不依赖数据库）— #1809 更新."""
 
     def test_create_job_success(self, mock_mediakit, mock_cosyvoice):
+        """TTS 直生——v4 异步模式：create_job 只创建 DB 记录 + dispatch Celery 任务."""
         from app.services.lipsync_service import LipsyncService
 
         mock_db = MagicMock()
-        mock_db.add = MagicMock()
-        mock_db.flush = MagicMock()
-        mock_db.commit = MagicMock()
-        mock_db.refresh = MagicMock()
+        mock_repo = MagicMock()
+        mock_repo.get.return_value = None  # 预置音色，原样返回 voice_id
 
-        svc = LipsyncService(mock_db, client=mock_mediakit, cosyvoice_service=mock_cosyvoice)
+        with patch(
+            "app.services.lipsync_service.tts_synthesize_and_submit"
+        ) as mock_task:
+            mock_task.apply_async.return_value = MagicMock(id="celery-task-123")
 
-        job = svc.create_job(
-            user_id="user-1",
-            video_url="https://example.com/video.mp4",
-            voice_id="longxiaochun_v3",
-            script_text="大家好，欢迎来到直播间",
-        )
+            svc = LipsyncService(
+                mock_db,
+                client=mock_mediakit,
+                voice_clone_repo=mock_repo,
+            )
 
-        assert job.status == "submitted"
-        assert job.mediakit_task_id == "mk-task-123"
-        # TTS 应该被调用
-        mock_cosyvoice.synthesize_speech.assert_called_once_with(
-            text="大家好，欢迎来到直播间",
-            voice_id="longxiaochun_v3",
-        )
-        # MediaKit 应该用 TTS 生成的 audio_url
-        mock_mediakit.submit_lipsync.assert_called_once()
-        call_kwargs = mock_mediakit.submit_lipsync.call_args
-        assert call_kwargs.kwargs["audio_url"] == "https://oss.example.com/tts-output.mp3"
+            job = svc.create_job(
+                user_id="user-1",
+                video_url="https://example.com/video.mp4",
+                voice_id="longxiaochun_v3",
+                script_text="大家好，欢迎来到直播间",
+                speed=1.2,
+                emotion="兴奋",
+            )
+
+        # TTS 模式：异步返回，状态为 tts_processing
+        assert job.status == "tts_processing"
+        assert not job.audio_url  # TTS 音频尚未合成（默认空字符串）
+        # 不直接调用 CosyVoice
+        mock_cosyvoice.submit_synthesize_task.assert_not_called()
+        # dispatch 了 Celery 任务
+        mock_task.apply_async.assert_called_once()
+        # job 记录透传字段
+        assert job.speed == 1.2
+        assert job.emotion == "excited"
+        # MediaKit 尚未提交（由 Celery 任务处理）
+        mock_mediakit.submit_lipsync.assert_not_called()
 
     def test_create_job_tts_failure(self, mock_mediakit):
-        """TTS 合成失败时，应创建 failed 记录并抛出 CosyVoiceError."""
+        """v4: TTS 模式下 create_job 不再同步失败，而是 dispatch Celery 任务。
+        TTS 合成失败由 Celery 任务内部处理（见 test_lipsync_speed_optimization.py）。"""
         from app.services.lipsync_service import LipsyncService
 
-        from packages.application.cosyvoice_service import CosyVoiceError
-
-        mock_cosyvoice = MagicMock()
-        mock_cosyvoice.synthesize_speech.side_effect = CosyVoiceError("TTS 服务不可用")
-
         mock_db = MagicMock()
-        mock_db.add = MagicMock()
-        mock_db.flush = MagicMock()
-        mock_db.commit = MagicMock()
-        mock_db.refresh = MagicMock()
+        mock_repo = MagicMock()
+        mock_repo.get.return_value = None
 
-        svc = LipsyncService(mock_db, client=mock_mediakit, cosyvoice_service=mock_cosyvoice)
+        svc = LipsyncService(
+            mock_db,
+            client=mock_mediakit,
+            voice_clone_repo=mock_repo,
+        )
 
-        with pytest.raises(CosyVoiceError, match="TTS 服务不可用"):
-            svc.create_job(
+        with patch(
+            "app.services.lipsync_service.tts_synthesize_and_submit"
+        ) as mock_task:
+            mock_task.apply_async.return_value = MagicMock(id="celery-task-456")
+
+            job = svc.create_job(
                 user_id="user-1",
                 video_url="https://example.com/video.mp4",
                 voice_id="longxiaochun_v3",
                 script_text="测试文本",
             )
 
-        # 不应提交到 MediaKit
+        # TTS 模式下 create_job 成功返回，状态为 tts_processing
+        assert job.status == "tts_processing"
         mock_mediakit.submit_lipsync.assert_not_called()
-        # 应该记录了失败状态
-        added_job = mock_db.add.call_args[0][0]
-        assert added_job.status == "failed"
-        assert "TTS" in added_job.error_message
+        mock_task.apply_async.assert_called_once()
 
-    def test_create_job_api_failure(self, mock_mediakit, mock_cosyvoice):
-        """MediaKit 提交失败."""
+    def test_create_job_api_failure(self, mock_mediakit):
+        """MediaKit 提交失败（直传音频模式同步触发）."""
         from app.services.lipsync_service import LipsyncService
         from app.services.mediakit_client import MediaKitError
 
         mock_mediakit.submit_lipsync.side_effect = MediaKitError("API 调用失败", code="SubmitFailed")
 
         mock_db = MagicMock()
-        svc = LipsyncService(mock_db, client=mock_mediakit, cosyvoice_service=mock_cosyvoice)
+        svc = LipsyncService(mock_db, client=mock_mediakit)
 
         with pytest.raises(MediaKitError, match="API 调用失败"):
             svc.create_job(
                 user_id="user-1",
                 video_url="https://example.com/video.mp4",
-                voice_id="longxiaochun_v3",
-                script_text="测试文本",
+                audio_url="https://example.com/audio.mp3",
             )
 
     def test_get_job_delegates_to_db(self, mock_mediakit, mock_cosyvoice):
@@ -399,78 +426,284 @@ class TestLipsyncServiceUnit:
         assert result.status == "completed"
 
     def test_create_job_stores_tts_audio_url(self, mock_mediakit, mock_cosyvoice):
-        """#1809: 验证 job 的 audio_url 来自 TTS 合成结果."""
+        """v4: TTS 模式下 create_job 返回 tts_processing 状态，audio_url 尚未设置（由 Celery 任务处理）."""
         from app.services.lipsync_service import LipsyncService
 
         mock_db = MagicMock()
-        mock_db.add = MagicMock()
-        mock_db.flush = MagicMock()
-        mock_db.commit = MagicMock()
-        mock_db.refresh = MagicMock()
+        mock_repo = MagicMock()
+        mock_repo.get.return_value = None
 
-        svc = LipsyncService(mock_db, client=mock_mediakit, cosyvoice_service=mock_cosyvoice)
+        with patch(
+            "app.services.lipsync_service.tts_synthesize_and_submit"
+        ) as mock_task:
+            mock_task.apply_async.return_value = MagicMock(id="celery-task-789")
 
+            svc = LipsyncService(
+                mock_db,
+                client=mock_mediakit,
+                voice_clone_repo=mock_repo,
+            )
+            job = svc.create_job(
+                user_id="user-1",
+                video_url="https://example.com/video.mp4",
+                voice_id="my-clone-voice",
+                script_text="这是一段测试文本",
+            )
+
+        # TTS 模式下 create_job 返回 tts_processing 状态
+        assert job.status == "tts_processing"
+        # audio_url 尚未设置（由 Celery 任务异步处理），模型默认为空字符串
+        assert not job.audio_url
+        mock_task.apply_async.assert_called_once()
+
+    def test_create_job_direct_audio_skips_tts(self, mock_mediakit, mock_cosyvoice):
+        """v3: 直接音频模式（传 audio_url）不触发 TTS，原样把 audio_url 提交 MediaKit."""
+        from app.services.lipsync_service import LipsyncService
+
+        mock_db = MagicMock()
+        svc = LipsyncService(
+            mock_db,
+            client=mock_mediakit,
+            cosyvoice_service=mock_cosyvoice,
+            voice_clone_repo=MagicMock(),
+        )
         job = svc.create_job(
             user_id="user-1",
             video_url="https://example.com/video.mp4",
-            voice_id="my-clone-voice",
-            script_text="这是一段测试文本",
+            audio_url="https://example.com/direct-audio.mp3",
         )
 
-        # job.audio_url 应该是 TTS 返回的 URL
-        assert job.audio_url == "https://oss.example.com/tts-output.mp3"
+        mock_cosyvoice.submit_synthesize_task.assert_not_called()
+        call_kwargs = mock_mediakit.submit_lipsync.call_args
+        assert call_kwargs.kwargs["audio_url"] == "https://example.com/direct-audio.mp3"
+        assert job.audio_url == "https://example.com/direct-audio.mp3"
 
 
 class TestErrorHandling:
-    """#1809 补充：错误返回 400 而非 500."""
+    """v3: 音色解析与错误码在 service 层处理，路由层做 HTTP 状态码映射."""
 
-    def test_voice_id_resolve_failure_returns_400(self, mock_mediakit, mock_cosyvoice):
-        """voice_clone_repo 查询异常时返回 400 而非 500."""
-        from app.api.routes.lipsync import _resolve_voice_id
-        from fastapi import HTTPException
+    def test_voice_id_resolve_forbidden(self, mock_mediakit, mock_cosyvoice):
+        """v3: 克隆音色属于他人时 service._resolve_voice_id 抛 VoiceForbidden（路由映射 403）."""
+        from app.services.lipsync_service import LipsyncService
+        from app.services.mediakit_client import MediaKitError
 
+        mock_db = MagicMock()
         mock_repo = MagicMock()
-        mock_repo.get.side_effect = Exception("DB connection error")
+        other_profile = MagicMock()
+        other_profile.user_id = "user-other"
+        other_profile.voice_id = "cv-voice-1"
+        mock_repo.get.return_value = other_profile
 
-        with pytest.raises(HTTPException) as exc_info:
-            _resolve_voice_id("bad-voice-id", "user-1", mock_repo)
-        assert exc_info.value.status_code == 400
-        assert "voice_id" in str(exc_info.value.detail)
+        svc = LipsyncService(
+            mock_db,
+            client=mock_mediakit,
+            cosyvoice_service=mock_cosyvoice,
+            voice_clone_repo=mock_repo,
+        )
 
-    def test_create_job_value_error_returns_400(self, mock_mediakit):
-        """ValueError（参数无效）返回 400 而非 500."""
-        from app.services.lipsync_service import LipsyncService
-
-        mock_cosyvoice = MagicMock()
-        mock_cosyvoice.synthesize_speech.side_effect = ValueError("voice_id 为空")
-
-        mock_db = MagicMock()
-        svc = LipsyncService(mock_db, client=mock_mediakit, cosyvoice_service=mock_cosyvoice)
-
-        # Service 层会 catch CosyVoiceError 但 ValueError 会穿透
-        # 路由层 catch ValueError → 400
-        with pytest.raises(ValueError):
+        with pytest.raises(MediaKitError) as exc_info:
             svc.create_job(
                 user_id="user-1",
                 video_url="https://example.com/video.mp4",
-                voice_id="",
-                script_text="test",
+                voice_id="clone-profile-id",
+                script_text="测试",
             )
+        assert exc_info.value.code == "VoiceForbidden"
+        mock_mediakit.submit_lipsync.assert_not_called()
 
-    def test_create_job_unexpected_exception_returns_400(self, mock_mediakit):
-        """未预期的异常应被路由层捕获返回 400 而非 500."""
+    def test_voice_id_resolve_not_ready(self, mock_mediakit, mock_cosyvoice):
+        """v3: 克隆音色尚未生成 voice_id 时抛 VoiceNotReady（路由映射 400）."""
         from app.services.lipsync_service import LipsyncService
-
-        mock_cosyvoice = MagicMock()
-        mock_cosyvoice.synthesize_speech.side_effect = RuntimeError("unexpected")
+        from app.services.mediakit_client import MediaKitError
 
         mock_db = MagicMock()
-        svc = LipsyncService(mock_db, client=mock_mediakit, cosyvoice_service=mock_cosyvoice)
+        mock_repo = MagicMock()
+        profile = MagicMock()
+        profile.user_id = "user-1"
+        profile.voice_id = ""  # 克隆未完成
+        mock_repo.get.return_value = profile
 
-        with pytest.raises(RuntimeError):
+        svc = LipsyncService(
+            mock_db,
+            client=mock_mediakit,
+            cosyvoice_service=mock_cosyvoice,
+            voice_clone_repo=mock_repo,
+        )
+
+        with pytest.raises(MediaKitError) as exc_info:
             svc.create_job(
                 user_id="user-1",
                 video_url="https://example.com/video.mp4",
-                voice_id="test-voice",
+                voice_id="clone-profile-id",
+                script_text="测试",
+            )
+        assert exc_info.value.code == "VoiceNotReady"
+
+    def test_tts_value_error_mapped_to_invalid_param(self, mock_mediakit):
+        """v4: TTS 模式下 create_job 不再同步调用 CosyVoice，
+        而是 dispatch Celery 任务。ValueError 由 Celery 任务内部处理。"""
+        from app.services.lipsync_service import LipsyncService
+
+        mock_cosyvoice = MagicMock()
+        mock_cosyvoice.submit_synthesize_task.side_effect = ValueError("voice_id 为空")
+
+        mock_db = MagicMock()
+        mock_repo = MagicMock()
+        mock_repo.get.return_value = None
+
+        svc = LipsyncService(
+            mock_db,
+            client=mock_mediakit,
+            voice_clone_repo=mock_repo,
+        )
+
+        with patch(
+            "app.services.lipsync_service.tts_synthesize_and_submit"
+        ) as mock_task:
+            mock_task.apply_async.return_value = MagicMock(id="celery-task-789")
+
+            # TTS 模式下 create_job 不再同步失败
+            job = svc.create_job(
+                user_id="user-1",
+                video_url="https://example.com/video.mp4",
+                voice_id="some-voice",
                 script_text="test",
             )
+
+        # 确认返回 tts_processing 状态
+        assert job.status == "tts_processing"
+        # TTS 合成由 Celery 任务处理，不直接调用 CosyVoice
+        mock_cosyvoice.submit_synthesize_task.assert_not_called()
+        mock_task.apply_async.assert_called_once()
+
+    def test_missing_both_inputs_raises_invalid_input(self, mock_mediakit, mock_cosyvoice):
+        """v3: 既无 audio_url 又无 voice_id+script_text 时抛 InvalidInput（路由映射 400）."""
+        from app.services.lipsync_service import LipsyncService
+        from app.services.mediakit_client import MediaKitError
+
+        mock_db = MagicMock()
+        svc = LipsyncService(
+            mock_db,
+            client=mock_mediakit,
+            cosyvoice_service=mock_cosyvoice,
+            voice_clone_repo=MagicMock(),
+        )
+
+        with pytest.raises(MediaKitError) as exc_info:
+            svc.create_job(
+                user_id="user-1",
+                video_url="https://example.com/video.mp4",
+            )
+        assert exc_info.value.code == "InvalidInput"
+        mock_cosyvoice.submit_synthesize_task.assert_not_called()
+        mock_mediakit.submit_lipsync.assert_not_called()
+
+
+class TestSignMediaUrl403Fix:
+    """#1839 私有桶 OSS URL 重签：MediaKit GPU worker 拉取裸/过期 URL 会 403.
+
+    - 自家 OSS 的裸 public_url / 已过期短预签名 → 重签 7 天长有效期
+    - 外部临时 URL（CosyVoice/MediaKit）→ 原样透传
+    - 签名异常 → 降级原 URL，不阻断
+    """
+
+    OSS_PUBLIC_BASE = "https://xiaoxia-autocut.oss-cn-hangzhou.aliyuncs.com"
+
+    def _svc(self, mock_mediakit, mock_cosyvoice):
+        from app.services.lipsync_service import LipsyncService
+
+        return LipsyncService(
+            MagicMock(),
+            client=mock_mediakit,
+            cosyvoice_service=mock_cosyvoice,
+            voice_clone_repo=MagicMock(),
+        )
+
+    def test_own_oss_unsigned_url_gets_resigned(self, mock_mediakit, mock_cosyvoice):
+        """裸 public_url（不带签名，私有桶匿名 403）必须被重签."""
+        from app.services.lipsync_service import MEDIAKIT_URL_TTL_SECONDS
+
+        svc = self._svc(mock_mediakit, mock_cosyvoice)
+        raw = f"{self.OSS_PUBLIC_BASE}/lipsync-tts/user-1/audio.mp3"
+        signed = raw + "?Expires=999&Signature=abc&OSSAccessKeyId=key"
+
+        storage = MagicMock()
+        storage.public_url = self.OSS_PUBLIC_BASE
+        storage.get_download_url.return_value = signed
+        with patch("app.services.lipsync_service.get_shared_storage_service", return_value=storage):
+            out = svc._sign_media_url(raw)
+
+        assert out == signed
+        storage.get_download_url.assert_called_once_with(raw, expires_seconds=MEDIAKIT_URL_TTL_SECONDS)
+        assert MEDIAKIT_URL_TTL_SECONDS == 7 * 24 * 3600
+
+    def test_own_oss_expired_presign_gets_resigned(self, mock_mediakit, mock_cosyvoice):
+        """已带过期签名的旧预签名 URL 也要抽 key 后重签（不把旧 query 带进新签名）."""
+        svc = self._svc(mock_mediakit, mock_cosyvoice)
+        old = f"{self.OSS_PUBLIC_BASE}/avatar/video.mp4?Expires=111&Signature=old"
+        fresh = f"{self.OSS_PUBLIC_BASE}/avatar/video.mp4?Expires=999&Signature=fresh"
+
+        storage = MagicMock()
+        storage.public_url = self.OSS_PUBLIC_BASE
+        storage.get_download_url.return_value = fresh
+        with patch("app.services.lipsync_service.get_shared_storage_service", return_value=storage):
+            out = svc._sign_media_url(old)
+
+        assert out == fresh
+        # 传给 get_download_url 的是原始完整 URL（内部抽 key），有效期 7 天
+        called_url = storage.get_download_url.call_args.args[0]
+        assert called_url == old
+
+    def test_external_url_passthrough(self, mock_mediakit, mock_cosyvoice):
+        """CosyVoice/MediaKit 外部临时链接不处理，原样透传."""
+        svc = self._svc(mock_mediakit, mock_cosyvoice)
+        external = "https://cv-tts.cosyvoice.aliyuncs.com/output/x.mp3"
+
+        storage = MagicMock()
+        storage.public_url = self.OSS_PUBLIC_BASE
+        with patch("app.services.lipsync_service.get_shared_storage_service", return_value=storage):
+            out = svc._sign_media_url(external)
+
+        assert out == external
+        storage.get_download_url.assert_not_called()
+
+    def test_empty_url_returns_empty(self, mock_mediakit, mock_cosyvoice):
+        svc = self._svc(mock_mediakit, mock_cosyvoice)
+        assert svc._sign_media_url("") == ""
+
+    def test_sign_error_falls_back_to_raw(self, mock_mediakit, mock_cosyvoice):
+        """签名抛异常时降级返回原 URL，不阻断对口型提交."""
+        svc = self._svc(mock_mediakit, mock_cosyvoice)
+        raw = f"{self.OSS_PUBLIC_BASE}/lipsync-tts/u/a.mp3"
+
+        storage = MagicMock()
+        storage.public_url = self.OSS_PUBLIC_BASE
+        storage.get_download_url.side_effect = RuntimeError("oss down")
+        with patch("app.services.lipsync_service.get_shared_storage_service", return_value=storage):
+            out = svc._sign_media_url(raw)
+
+        assert out == raw
+
+    def test_create_job_resigns_oss_urls_before_submit(self, mock_mediakit, mock_cosyvoice):
+        """端到端：create_job 提交 MediaKit 前，自家 OSS 的 video_url 必须是重签后的长签名 URL."""
+        svc = self._svc(mock_mediakit, mock_cosyvoice)
+        raw_video = f"{self.OSS_PUBLIC_BASE}/avatar/person.mp4"
+        signed_video = raw_video + "?Expires=999&Signature=fresh"
+        raw_audio = f"{self.OSS_PUBLIC_BASE}/direct/audio.mp3"
+        signed_audio = raw_audio + "?Expires=999&Signature=afresh"
+
+        storage = MagicMock()
+        storage.public_url = self.OSS_PUBLIC_BASE
+        storage.get_download_url.side_effect = lambda u, expires_seconds=0: (
+            signed_video if u == raw_video else signed_audio
+        )
+        with patch("app.services.lipsync_service.get_shared_storage_service", return_value=storage):
+            svc.create_job(
+                user_id="user-1",
+                video_url=raw_video,
+                audio_url=raw_audio,
+            )
+
+        kw = mock_mediakit.submit_lipsync.call_args.kwargs
+        assert kw["video_url"] == signed_video
+        assert kw["audio_url"] == signed_audio

tests/unit/test_lipsync_speed_optimization.py

@@ -0,0 +1,334 @@
+"""AI 数字人口型视频生成速度优化 — 单元测试.
+
+验证两个优化点：
+1. FFmpeg 编码 preset 从 fast 改为 veryfast（提速 30~50%）
+2. TTS 合成从同步改为 Celery 异步任务（API 响应从 6~35s 降到 <1s）
+
+Issue: lipsync-speed-optimization
+"""
+
+import os
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+os.environ.setdefault("JWT_SECRET_KEY", "dev-secret-key-for-testing")
+
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# 优化1: FFmpeg 编码提速 — preset veryfast
+# ═══════════════════════════════════════════════════════════════════════════════
+
+
+class TestFFmpegPresetOptimization:
+    """验证 FFmpeg 编码命令从 -preset fast 改为 -preset veryfast."""
+
+    def test_preset_is_veryfast(self):
+        """_build_ffmpeg_command 输出必须包含 -preset veryfast."""
+        from app.services.ai_avatar_render_service import AiAvatarRenderService
+
+        svc = AiAvatarRenderService.__new__(AiAvatarRenderService)
+        cmd = svc._build_ffmpeg_command(
+            input_video="https://example.com/video.mp4",
+            b_roll_segments=[],
+            filter_complex="",
+            final_label=None,
+            output_path="/tmp/output.mp4",
+        )
+        # cmd 现在是 list[str]；preset 与值是相邻两个元素
+        assert "-preset" in cmd, f"期望包含 -preset，实际命令: {cmd}"
+        preset_idx = cmd.index("-preset")
+        assert cmd[preset_idx + 1] == "veryfast", f"期望 veryfast，实际: {cmd}"
+
+    def test_preset_veryfast_with_filter(self):
+        """带滤镜场景下也必须使用 veryfast."""
+        from app.services.ai_avatar_render_service import AiAvatarRenderService
+
+        svc = AiAvatarRenderService.__new__(AiAvatarRenderService)
+        cmd = svc._build_ffmpeg_command(
+            input_video="https://example.com/video.mp4",
+            b_roll_segments=[],
+            filter_complex="overlay=0:0",
+            final_label="[v]",
+            output_path="/tmp/output.mp4",
+        )
+        assert "-preset" in cmd
+        assert cmd[cmd.index("-preset") + 1] == "veryfast"
+        assert "-filter_complex" in cmd
+
+    def test_preset_not_fast(self):
+        """确保不再使用旧的 -preset fast."""
+        from app.services.ai_avatar_render_service import AiAvatarRenderService
+
+        svc = AiAvatarRenderService.__new__(AiAvatarRenderService)
+        cmd = svc._build_ffmpeg_command(
+            input_video="https://example.com/video.mp4",
+            b_roll_segments=[],
+            filter_complex="",
+            final_label=None,
+            output_path="/tmp/output.mp4",
+        )
+        # 确保是 veryfast 而不是 fast
+        assert "-preset" in cmd
+        preset_idx = cmd.index("-preset")
+        assert cmd[preset_idx + 1] == "veryfast"
+        # 禁止 fast 单独作为 preset 值（veryfast 包含 "fast" 子串，不影响）
+        assert cmd[preset_idx + 1] != "fast"
+
+
+# ═══════════════════════════════════════════════════════════════════════════════
+# 优化2: TTS 合成 Celery 异步化
+# ═══════════════════════════════════════════════════════════════════════════════
+
+
+def _make_service_with_mocks():
+    """构造 LipsyncService 测试实例及 mock 依赖."""
+    from app.services.lipsync_service import LipsyncService
+
+    db = MagicMock()
+    client = MagicMock()
+    client.is_available = True
+    client.submit_lipsync.return_value = {
+        "success": True,
+        "task_id": "mk-1",
+        "request_id": "req-1",
+    }
+    cosy = MagicMock()
+    cosy.submit_synthesize_task.return_value = {
+        "audio_url": "https://tts/raw.mp3",
+        "request_id": "tts-req",
+        "audio_duration": 3.0,
+    }
+    svc = LipsyncService(db, client=client, cosyvoice_service=cosy, voice_clone_repo=MagicMock())
+    # _resolve_voice_id 默认原样返回（repo.get 返回 None）
+    svc._voice_clone_repo.get.return_value = None
+    return svc, client, cosy
+
+
+class TestCreateJobAsyncTTS:
+    """验证 TTS 模式改为 Celery 异步后的行为."""
+
+    def test_tts_mode_returns_tts_processing_status(self):
+        """TTS 模式下 create_job 立即返回，状态为 tts_processing."""
+        svc, client, cosy = _make_service_with_mocks()
+
+        with patch("app.services.lipsync_service.tts_synthesize_and_submit") as mock_task:
+            mock_task.apply_async = MagicMock()
+
+            job = svc.create_job(
+                user_id="user-1",
+                video_url="https://example.com/video.mp4",
+                voice_id="longxiaochun_v3",
+                script_text="大家好",
+                speed=1.0,
+                emotion="",
+            )
+
+        assert job.status == "tts_processing"
+
+    def test_tts_mode_dispatches_celery_task(self):
+        """TTS 模式必须 dispatch Celery 异步任务."""
+        svc, client, cosy = _make_service_with_mocks()
+
+        with patch("app.services.lipsync_service.tts_synthesize_and_submit") as mock_task:
+            mock_task.apply_async = MagicMock()
+
+            svc.create_job(
+                user_id="user-1",
+                video_url="https://example.com/video.mp4",
+                voice_id="v-1",
+                script_text="测试文本",
+            )
+
+        mock_task.apply_async.assert_called_once()
+        call_kwargs = mock_task.apply_async.call_args
+        args = call_kwargs.kwargs.get("args") or call_kwargs[1].get("args", call_kwargs[0][0] if call_kwargs[0] else ())
+        assert args[1] == "user-1"  # user_id
+        assert args[2] == "v-1"  # voice_id
+        assert args[3] == "测试文本"  # script_text
+
+    def test_tts_mode_celery_dispatch_failure_marks_job_failed(self):
+        """Celery dispatch 失败时，job 标为 failed 并写入 error_message，前端轮询能直接看到错误."""
+        svc, client, cosy = _make_service_with_mocks()
+
+        with patch("app.services.lipsync_service.tts_synthesize_and_submit") as mock_task:
+            mock_task.apply_async = MagicMock(side_effect=Exception("Celery broker down"))
+
+            job = svc.create_job(
+                user_id="user-1",
+                video_url="https://example.com/video.mp4",
+                voice_id="v-1",
+                script_text="测试文本",
+            )
+
+        # job 已创建且状态标为 failed
+        assert job is not None
+        assert job.status == "failed"
+        assert "Celery 任务投递失败" in job.error_message
+        assert job.error_code == "AsyncDispatchFailed"
+        # MediaKit 未被调用
+        client.submit_lipsync.assert_not_called()
+
+    def test_tts_mode_voice_validation_still_sync(self):
+        """TTS 模式下音色校验仍在 HTTP 请求中同步执行."""
+        from app.services.mediakit_client import MediaKitError
+
+        svc, client, cosy = _make_service_with_mocks()
+        # 模拟音色属于其他用户
+        other_profile = MagicMock()
+        other_profile.user_id = "user-other"
+        svc._voice_clone_repo.get.return_value = other_profile
+
+        with patch("app.tasks.lipsync_tts.tts_synthesize_and_submit"):
+            with pytest.raises(MediaKitError) as exc:
+                svc.create_job(
+                    user_id="user-1",
+                    video_url="https://example.com/video.mp4",
+                    voice_id="clone-profile-id",
+                    script_text="测试",
+                )
+        assert exc.value.code == "VoiceForbidden"
+
+    def test_tts_mode_missing_input_raises_immediately(self):
+        """缺少 voice_id 或 script_text 时立即报错，不 dispatch Celery 任务."""
+        from app.services.mediakit_client import MediaKitError
+
+        svc, client, cosy = _make_service_with_mocks()
+
+        with patch("app.tasks.lipsync_tts.tts_synthesize_and_submit") as mock_task:
+            mock_task.delay = MagicMock()
+
+            with pytest.raises(MediaKitError) as exc:
+                svc.create_job(
+                    user_id="user-1",
+                    video_url="https://example.com/video.mp4",
+                    # 缺少 voice_id 和 script_text
+                )
+            assert exc.value.code == "InvalidInput"
+
+        # Celery 任务未被 dispatch
+        mock_task.delay.assert_not_called()
+        # TTS 和 MediaKit 均未调用
+        cosy.submit_synthesize_task.assert_not_called()
+        client.submit_lipsync.assert_not_called()
+
+
+class TestCreateJobDirectAudio:
+    """验证直接音频模式不受异步化影响."""
+
+    def test_direct_audio_still_submits_synchronously(self):
+        """直接音频模式仍然同步提交 MediaKit，状态为 submitted."""
+        svc, client, cosy = _make_service_with_mocks()
+
+        with patch("app.tasks.lipsync_tts.tts_synthesize_and_submit") as mock_task:
+            mock_task.delay = MagicMock()
+
+            job = svc.create_job(
+                user_id="user-1",
+                video_url="https://example.com/video.mp4",
+                audio_url="https://example.com/audio.mp3",
+            )
+
+        assert job.status == "submitted"
+        assert job.mediakit_task_id == "mk-1"
+        client.submit_lipsync.assert_called_once()
+        # TTS Celery 任务不应被调用
+        mock_task.delay.assert_not_called()
+
+    def test_direct_audio_skips_tts(self):
+        """直接音频模式不调用 CosyVoice TTS."""
+        svc, client, cosy = _make_service_with_mocks()
+
+        job = svc.create_job(
+            user_id="user-1",
+            video_url="https://example.com/video.mp4",
+            audio_url="https://example.com/audio.mp3",
+        )
+
+        cosy.submit_synthesize_task.assert_not_called()
+        call_kwargs = client.submit_lipsync.call_args
+        assert call_kwargs.kwargs["audio_url"] == "https://example.com/audio.mp3"
+
+
+class TestCancelJobTtsProcessing:
+    """验证 tts_processing 状态的任务可以被取消."""
+
+    def test_cancel_tts_processing(self):
+        """tts_processing 状态的任务可以成功取消."""
+        svc, client, cosy = _make_service_with_mocks()
+
+        mock_job = MagicMock()
+        mock_job.status = "tts_processing"
+        mock_job.id = "job-1"
+        svc.get_job = MagicMock(return_value=mock_job)
+
+        result = svc.cancel_job("job-1", "user-1")
+        assert result.status == "cancelled"
+
+    def test_cancel_pending_still_works(self):
+        """pending 状态仍可取消."""
+        svc, client, cosy = _make_service_with_mocks()
+
+        mock_job = MagicMock()
+        mock_job.status = "pending"
+        mock_job.id = "job-1"
+        svc.get_job = MagicMock(return_value=mock_job)
+
+        result = svc.cancel_job("job-1", "user-1")
+        assert result.status == "cancelled"
+
+    def test_cancel_submitted_still_works(self):
+        """submitted 状态仍可取消."""
+        svc, client, cosy = _make_service_with_mocks()
+
+        mock_job = MagicMock()
+        mock_job.status = "submitted"
+        mock_job.id = "job-1"
+        svc.get_job = MagicMock(return_value=mock_job)
+
+        result = svc.cancel_job("job-1", "user-1")
+        assert result.status == "cancelled"
+
+
+class TestCreateJobCommitOrder:
+    """验证事务顺序修复：create_job 必须先 commit 再发 Celery 任务，避免 worker 消费时 job 不可见。"""
+
+    def test_commit_called_before_apply_async_in_tts_mode(self):
+        """TTS 模式：db.commit() 必须在 apply_async() 之前调用，防止 worker 查不到 job 永远卡在 tts_processing。"""
+        svc, client, cosy = _make_service_with_mocks()
+        call_order: list[str] = []
+
+        def track_commit():
+            call_order.append("commit")
+
+        def track_apply_async(*args, **kwargs):
+            call_order.append("apply_async")
+
+        svc.db.commit.side_effect = track_commit
+
+        with patch("app.services.lipsync_service.tts_synthesize_and_submit") as mock_task:
+            mock_task.apply_async = MagicMock(side_effect=track_apply_async)
+            svc.create_job(
+                user_id="user-1",
+                video_url="https://example.com/video.mp4",
+                voice_id="v-1",
+                script_text="测试",
+            )
+
+        # 至少有一次 commit 在 apply_async 之前
+        assert "commit" in call_order, "db.commit 必须被调用"
+        assert "apply_async" in call_order, "apply_async 必须被调用"
+        assert call_order.index("commit") < call_order.index(
+            "apply_async"
+        ), f"事务顺序错误：commit 必须在 apply_async 之前，实际顺序 {call_order}"
+
+    def test_job_not_found_retry_mechanism_exists(self):
+        """worker 侧 job not found 必须有重试机制（self.retry），而不是静默 return。"""
+        import inspect
+
+        from app.tasks.lipsync_tts import tts_synthesize_and_submit
+
+        source = inspect.getsource(tts_synthesize_and_submit.run)
+        assert (
+            "self.retry" in source or "retry" in source
+        ), "tts_synthesize_and_submit 在 job not found 时必须重试，防止静默失败"

tests/unit/test_lipsync_tts.py

@@ -0,0 +1,628 @@
+"""AI 数字人口型 TTS Celery 异步任务 — 单元测试.
+
+覆盖 lipsync_tts.py 的全部主要分支：
+- Job 不存在/cancelled/正常/异常路径
+- TTS 合成、音频下载、OSS 上传、MediaKit 提交
+- CosyVoiceError/ValueError/MediaKitError/顶层异常等错误码
+"""
+
+import os
+import sys
+import types
+from types import ModuleType
+from unittest.mock import MagicMock, patch
+
+os.environ.setdefault("JWT_SECRET_KEY", "dev-secret-key-for-testing")
+
+
+class _FakeQuery:
+    """模拟 SQLAlchemy query.filter().first() 链式调用."""
+
+    def __init__(self, job):
+        self._job = job
+
+    def filter(self, *args, **kwargs):
+        return self
+
+    def first(self):
+        return self._job
+
+
+def _make_fake_job(**kwargs):
+    """构造可 setattr 的 job 记录."""
+    job = MagicMock()
+    job.id = kwargs.get("job_id", "job-1")
+    job.user_id = kwargs.get("user_id", "user-1")
+    job.status = kwargs.get("status", "tts_processing")
+    job.audio_url = kwargs.get("audio_url", "")
+    job.video_url = kwargs.get("video_url", "https://oss/video.mp4")
+    job.mediakit_task_id = kwargs.get("mediakit_task_id", "")
+    job.enable_video_loop = kwargs.get("enable_video_loop", False)
+    job.error_code = ""
+    job.error_message = ""
+    job.submitted_at = None
+    job.updated_at = None
+    return job
+
+
+def _build_session(job):
+    """构造 mock DB session + factory. 返回 (session, factory)."""
+    session = MagicMock()
+    session.query.return_value = _FakeQuery(job)
+    session.commit = MagicMock()
+    session.close = MagicMock()
+    factory = MagicMock(return_value=session)
+    return session, factory
+
+
+def _apply_all_patches(
+    *,
+    job=None,
+    cosyvoice_service=None,
+    cosyvoice_side_effect=None,
+    cosyvoice_error=None,
+    download_bytes=b"AUDIO",
+    download_error=None,
+    storage=None,
+    mk_client=None,
+    mk_submit_return=None,
+    mk_submit_error=None,
+):
+    """统一构造测试需要的 patch 列表.
+
+    lipsync_tts.run() 在函数体内部懒 import 多个模块，通过 sys.modules 注入
+    伪造包路径避免真实导入；对存在的模块用 patch() 替换返回值/side_effect。
+    """
+    # SessionLocal 通过懒探测获取（Worker 用 worker_app.db，API 用 app.db），
+    # 测试环境里两个模块都能被真实导入，必须同时 mock 保证用的是 fake session。
+    fake_app_db = ModuleType("app.db")
+    fake_worker_db = ModuleType("worker_app.db")
+    session, factory = _build_session(job)
+    fake_app_db.SessionLocal = factory
+    fake_worker_db.SessionLocal = factory
+
+    patches = [
+        patch.dict(sys.modules, {"app.db": fake_app_db, "worker_app.db": fake_worker_db}),
+        patch(
+            "app.tasks.lipsync_tts._sign_media_url",
+            side_effect=lambda url: url + "?signed" if url else url,
+        ),
+    ]
+
+    # CosyVoice
+    if cosyvoice_service is not None:
+        cosy_instance = cosyvoice_service
+    else:
+        cosy_instance = MagicMock()
+        if cosyvoice_side_effect is not None:
+            cosy_instance.submit_synthesize_task.side_effect = cosyvoice_side_effect
+        elif cosyvoice_error is not None:
+            cosy_instance.submit_synthesize_task.side_effect = cosyvoice_error
+        else:
+            cosy_instance.submit_synthesize_task.return_value = {"audio_url": "https://tts/raw.mp3"}
+    patches.append(patch("packages.application.cosyvoice_service.CosyVoiceService", return_value=cosy_instance))
+
+    # safe_download_bytes
+    if download_error is not None:
+        patches.append(patch("packages.shared.url_security.safe_download_bytes", side_effect=download_error))
+    else:
+        patches.append(patch("packages.shared.url_security.safe_download_bytes", return_value=download_bytes))
+
+    # Storage
+    if storage is None:
+        storage = MagicMock()
+        storage.public_url = "https://oss.example.com"
+        storage.upload_file.return_value = "https://oss.example.com/tts.mp3"
+    patches.append(patch("packages.shared.storage.get_shared_storage_service", return_value=storage))
+
+    # MediaKit client
+    if mk_client is not None:
+        patches.append(patch("app.services.mediakit_client.get_mediakit_client", return_value=mk_client))
+    else:
+        client = MagicMock()
+        if mk_submit_error is not None:
+            client.submit_lipsync.side_effect = mk_submit_error
+        else:
+            client.submit_lipsync.return_value = mk_submit_return or {"task_id": "mk-1"}
+        patches.append(patch("app.services.mediakit_client.get_mediakit_client", return_value=client))
+
+    return session, patches
+
+
+class TestTtsSynthesizeAndSubmit:
+    """测试 Celery 任务 tts_synthesize_and_submit.run 的所有分支."""
+
+    def test_job_not_found_returns_early(self):
+        """Job 不存在 → 日志报错直接返回，不抛异常."""
+        from app.tasks.lipsync_tts import tts_synthesize_and_submit
+
+        session, patches = _apply_all_patches(job=None)
+        entered = [p.__enter__() for p in patches]
+        try:
+            tts_synthesize_and_submit.run("missing-job", "user-1", "v1", "你好", 1.0, "")
+        finally:
+            for p in reversed(patches):
+                p.__exit__(None, None, None)
+
+        session.commit.assert_not_called()
+        session.close.assert_called_once()
+
+    def test_cancelled_job_skipped(self):
+        """Job 已 cancelled → 跳过不处理，不调用 TTS/MediaKit."""
+        from app.tasks.lipsync_tts import tts_synthesize_and_submit
+
+        job = _make_fake_job(status="cancelled")
+        session, patches = _apply_all_patches(job=job)
+        entered = [p.__enter__() for p in patches]
+        try:
+            tts_synthesize_and_submit.run("job-1", "user-1", "v1", "你好", 1.0, "")
+        finally:
+            for p in reversed(patches):
+                p.__exit__(None, None, None)
+
+        # cancelled 不应 commit，不应触发 TTS/MediaKit
+        session.commit.assert_not_called()
+        session.close.assert_called_once()
+
+    def test_happy_path_tts_to_mediakit(self):
+        """完整正常流程：TTS 合成 → OSS 上传 → 签名 → 提交 MediaKit → submitted."""
+        from app.tasks.lipsync_tts import tts_synthesize_and_submit
+
+        job = _make_fake_job()
+        mk_client = MagicMock()
+        mk_client.submit_lipsync.return_value = {"task_id": "mk-999"}
+        session, patches = _apply_all_patches(job=job, mk_client=mk_client)
+        entered = [p.__enter__() for p in patches]
+        try:
+            tts_synthesize_and_submit.run("job-1", "user-1", "v1", "你好世界", 1.0, "happy")
+        finally:
+            for p in reversed(patches):
+                p.__exit__(None, None, None)
+
+        assert job.status == "submitted"
+        assert job.mediakit_task_id == "mk-999"
+        assert job.error_code == ""
+        mk_client.submit_lipsync.assert_called_once()
+        call_kwargs = mk_client.submit_lipsync.call_args.kwargs
+        assert call_kwargs["client_token"] == "job-1"
+        # CosyVoice 临时 URL 经 _sign_media_url 透传（mock 统一追加 ?signed），
+        # 自家 OSS 才会被重签，外部 URL 原样透传；job.audio_url 存原始临时 URL
+        assert call_kwargs["audio_url"] == "https://tts/raw.mp3?signed"
+        assert job.audio_url == "https://tts/raw.mp3"
+        session.commit.assert_called()
+        session.close.assert_called_once()
+
+    def test_cosyvoice_error_marks_tts_synthesis_failed(self):
+        """CosyVoiceError → failed, error_code=TTSSynthesisFailed."""
+        from app.tasks.lipsync_tts import tts_synthesize_and_submit
+
+        from packages.application.cosyvoice_service import CosyVoiceError
+
+        job = _make_fake_job()
+        session, patches = _apply_all_patches(job=job, cosyvoice_error=CosyVoiceError("tts boom"))
+        entered = [p.__enter__() for p in patches]
+        try:
+            tts_synthesize_and_submit.run("job-1", "user-1", "v1", "你好", 1.0, "")
+        finally:
+            for p in reversed(patches):
+                p.__exit__(None, None, None)
+
+        assert job.status == "failed"
+        assert job.error_code == "TTSSynthesisFailed"
+        session.close.assert_called_once()
+
+    def test_value_error_marks_tts_invalid_param(self):
+        """ValueError（参数错误）→ failed, error_code=TTSInvalidParam."""
+        from app.tasks.lipsync_tts import tts_synthesize_and_submit
+
+        job = _make_fake_job()
+        session, patches = _apply_all_patches(job=job, cosyvoice_side_effect=ValueError("bad param"))
+        entered = [p.__enter__() for p in patches]
+        try:
+            tts_synthesize_and_submit.run("job-1", "user-1", "v1", "你好", -1.0, "")
+        finally:
+            for p in reversed(patches):
+                p.__exit__(None, None, None)
+
+        assert job.status == "failed"
+        assert job.error_code == "TTSInvalidParam"
+        session.close.assert_called_once()
+
+    def test_no_audio_url_marks_tts_no_audio(self):
+        """TTS 返回空 audio_url → failed, error_code=TTSNoAudio."""
+        from app.tasks.lipsync_tts import tts_synthesize_and_submit
+
+        job = _make_fake_job()
+        cosy = MagicMock()
+        cosy.submit_synthesize_task.return_value = {"audio_url": ""}
+        session, patches = _apply_all_patches(job=job, cosyvoice_service=cosy)
+        entered = [p.__enter__() for p in patches]
+        try:
+            tts_synthesize_and_submit.run("job-1", "user-1", "v1", "你好", 1.0, "")
+        finally:
+            for p in reversed(patches):
+                p.__exit__(None, None, None)
+
+        assert job.status == "failed"
+        assert job.error_code == "TTSNoAudio"
+        session.close.assert_called_once()
+
+    def test_oss_upload_failure_falls_back_to_temp_url(self):
+        """OSS 上传失败 → 回退临时 URL，仍然 submitted."""
+        from app.tasks.lipsync_tts import tts_synthesize_and_submit
+
+        job = _make_fake_job()
+        storage = MagicMock()
+        storage.public_url = "https://oss.example.com"
+        storage.upload_file.side_effect = RuntimeError("oss down")
+        mk_client = MagicMock()
+        mk_client.submit_lipsync.return_value = {"task_id": "mk-7"}
+        session, patches = _apply_all_patches(job=job, storage=storage, mk_client=mk_client)
+        entered = [p.__enter__() for p in patches]
+        try:
+            tts_synthesize_and_submit.run("job-1", "user-1", "v1", "你好", 1.0, "")
+        finally:
+            for p in reversed(patches):
+                p.__exit__(None, None, None)
+
+        # 上传失败后 audio_url 回退为临时 TTS URL，仍继续提交到 MediaKit
+        assert job.audio_url == "https://tts/raw.mp3"
+        assert job.status == "submitted"
+        assert job.mediakit_task_id == "mk-7"
+        mk_client.submit_lipsync.assert_called_once()
+        session.close.assert_called_once()
+
+    def test_mediakit_error_marks_mediakit_unavailable(self):
+        """MediaKit 提交失败 → failed, error_code=MediaKitUnavailable."""
+        from app.services.mediakit_client import MediaKitError
+        from app.tasks.lipsync_tts import tts_synthesize_and_submit
+
+        job = _make_fake_job()
+        mk_err = MediaKitError("mk down", code="MediaKitUnavailable")
+        session, patches = _apply_all_patches(job=job, mk_submit_error=mk_err)
+        entered = [p.__enter__() for p in patches]
+        try:
+            tts_synthesize_and_submit.run("job-1", "user-1", "v1", "你好", 1.0, "")
+        finally:
+            for p in reversed(patches):
+                p.__exit__(None, None, None)
+
+        assert job.status == "failed"
+        assert job.error_code == "MediaKitUnavailable"
+        session.close.assert_called_once()
+
+    def test_top_level_exception_marks_async_task_error(self):
+        """顶层未预期异常 → failed, error_code=AsyncTaskError."""
+        from app.tasks.lipsync_tts import tts_synthesize_and_submit
+
+        job = _make_fake_job()
+        fake_app_db = ModuleType("app.db")
+        fake_worker_db = ModuleType("worker_app.db")
+        session, factory = _build_session(job)
+        fake_app_db.SessionLocal = factory
+        fake_worker_db.SessionLocal = factory
+
+        # CosyVoiceService 在 __init__ 抛 RuntimeError（非 CosyVoiceError/ValueError）
+        fake_cosy_mod = ModuleType("packages.application.cosyvoice_service")
+
+        class _CosyVoiceErrorForTest(Exception):
+            pass
+
+        class _BoomService:
+            def __init__(self):
+                raise RuntimeError("top-level boom")
+
+        fake_cosy_mod.CosyVoiceError = _CosyVoiceErrorForTest
+        fake_cosy_mod.CosyVoiceService = _BoomService
+
+        with patch.dict(
+            sys.modules,
+            {
+                "app.db": fake_app_db,
+                "worker_app.db": fake_worker_db,
+                "packages.application.cosyvoice_service": fake_cosy_mod,
+            },
+        ):
+            tts_synthesize_and_submit.run("job-1", "user-1", "v1", "你好", 1.0, "")
+
+        assert job.status == "failed"
+        assert job.error_code == "AsyncTaskError"
+        session.close.assert_called()
+
+
+class TestSignMediaUrl:
+    """覆盖模块内 _sign_media_url 的所有分支（CI 增量覆盖率需要）."""
+
+    def test_empty_url_returns_empty(self):
+        from app.tasks.lipsync_tts import _sign_media_url
+
+        assert _sign_media_url("") == ""
+        assert _sign_media_url(None) is None
+
+    def test_own_oss_url_signed(self):
+        """自家 OSS URL → 调用 storage.get_download_url 签名."""
+        from app.tasks.lipsync_tts import _sign_media_url
+
+        fake_storage = MagicMock()
+        fake_storage.public_url = "https://oss.example.com/"
+        fake_storage.get_download_url.return_value = "https://oss.example.com/a?sig=xyz"
+
+        with patch("packages.shared.storage.get_shared_storage_service", return_value=fake_storage):
+            result = _sign_media_url("https://oss.example.com/lipsync/a.mp3")
+
+        assert result == "https://oss.example.com/a?sig=xyz"
+        fake_storage.get_download_url.assert_called_once()
+
+    def test_external_url_passthrough(self):
+        """外部 URL（不是自家 OSS host）→ 原样透传，不签名."""
+        from app.tasks.lipsync_tts import _sign_media_url
+
+        fake_storage = MagicMock()
+        fake_storage.public_url = "https://oss.example.com/"
+
+        with patch("packages.shared.storage.get_shared_storage_service", return_value=fake_storage):
+            result = _sign_media_url("https://tts.example.com/raw.mp3")
+
+        assert result == "https://tts.example.com/raw.mp3"
+        fake_storage.get_download_url.assert_not_called()
+
+    def test_storage_exception_falls_back(self):
+        """storage 调用异常 → 降级原样返回，不抛错."""
+        from app.tasks.lipsync_tts import _sign_media_url
+
+        with patch(
+            "packages.shared.storage.get_shared_storage_service",
+            side_effect=RuntimeError("storage down"),
+        ):
+            result = _sign_media_url("https://oss.example.com/a.mp3")
+
+        assert result == "https://oss.example.com/a.mp3"
+
+    def test_no_public_url_passthrough(self):
+        """storage.public_url 为空 → 原样透传."""
+        from app.tasks.lipsync_tts import _sign_media_url
+
+        fake_storage = MagicMock()
+        fake_storage.public_url = ""
+
+        with patch("packages.shared.storage.get_shared_storage_service", return_value=fake_storage):
+            result = _sign_media_url("https://anything.example.com/a.mp3")
+
+        assert result == "https://anything.example.com/a.mp3"
+        fake_storage.get_download_url.assert_not_called()
+
+
+class TestPersistOutputVideoTask:
+    """persist_output_video_task：下载 MediaKit 临时视频 → 上传自有 OSS → 更新 DB."""
+
+    def _make_persist_job(self, **kwargs):
+        job = MagicMock()
+        job.id = kwargs.get("job_id", "job-1")
+        job.user_id = kwargs.get("user_id", "user-1")
+        job.output_video_url = kwargs.get("output_video_url", "https://temp.mk/output.mp4")
+        job.updated_at = None
+        return job
+
+    def _persist_patches(self, *, job, video_bytes=b"FAKEMP4", download_side_effect=None, upload_url=None):
+        """统一 patch：SessionLocal、httpx.Client、storage、_sign_media_url."""
+        fake_app_db = ModuleType("app.db")
+        fake_worker_db = ModuleType("worker_app.db")
+        session, factory = _build_session(job)
+        fake_app_db.SessionLocal = factory
+        fake_worker_db.SessionLocal = factory
+
+        # httpx.Client 上下文管理器
+        fake_response = MagicMock()
+        fake_response.content = video_bytes
+        fake_response.raise_for_status = MagicMock()
+        fake_client = MagicMock()
+        fake_client.get.return_value = fake_response
+        fake_client_cm = MagicMock()
+        fake_client_cm.__enter__ = MagicMock(return_value=fake_client)
+        fake_client_cm.__exit__ = MagicMock(return_value=False)
+        FakeHttpxClient = MagicMock(return_value=fake_client_cm)
+        if download_side_effect is not None:
+            fake_client.get.side_effect = download_side_effect
+
+        # storage
+        storage = MagicMock()
+        storage.public_url = "https://oss.example.com/"
+        storage.upload_file.return_value = upload_url or "https://oss.example.com/lipsync-outputs/user-1/job-1.mp4"
+        # _sign_media_url 内部会调 storage.get_download_url，必须mock返回字符串
+        _upload_url = upload_url or "https://oss.example.com/lipsync-outputs/user-1/job-1.mp4"
+        storage.get_download_url.return_value = _upload_url + "?signed"
+
+        fake_httpx = ModuleType("httpx")
+        fake_httpx.Client = FakeHttpxClient
+
+        patches = [
+            patch.dict(
+                sys.modules,
+                {"app.db": fake_app_db, "worker_app.db": fake_worker_db, "httpx": fake_httpx},
+            ),
+            patch("packages.shared.storage.get_shared_storage_service", return_value=storage),
+            patch("app.tasks.lipsync_tts._sign_media_url", side_effect=lambda url: url + "?signed" if url else url),
+        ]
+        return session, fake_client, storage, patches
+
+    def test_success_download_upload_updates_db(self):
+        """正常路径：下载 temp_url → 上传 OSS → 签名 → 写回 DB commit."""
+        from app.tasks.lipsync_tts import persist_output_video_task
+
+        job = self._make_persist_job(output_video_url="https://temp.mk/x.mp4")
+        session, fake_client, storage, patches = self._persist_patches(
+            job=job, video_bytes=b"VIDEODATA", upload_url="https://oss.example.com/lipsync-outputs/user-1/job-1.mp4"
+        )
+        entered = [p.__enter__() for p in patches]
+        try:
+            persist_output_video_task("job-1", "user-1", "https://temp.mk/x.mp4")
+        finally:
+            for p in reversed(patches):
+                p.__exit__(None, None, None)
+
+        fake_client.get.assert_called_once_with("https://temp.mk/x.mp4")
+        storage.upload_file.assert_called_once()
+        call_args = storage.upload_file.call_args.args
+        # 上传的 key 必须是 lipsync-outputs/{user_id}/{job_id}.mp4
+        assert call_args[1] == "lipsync-outputs/user-1/job-1.mp4"
+        # upload_file 返回永久 URL，再被 _sign_media_url 追加 ?signed
+        assert job.output_video_url == "https://oss.example.com/lipsync-outputs/user-1/job-1.mp4?signed"
+        assert job.updated_at is not None
+        session.commit.assert_called_once()
+        session.close.assert_called_once()
+
+    def test_download_failure_keeps_temp_url_no_commit(self):
+        """下载失败（raise）→ 记录 warning、保留 temp_url、不抛异常."""
+        from app.tasks.lipsync_tts import persist_output_video_task
+
+        job = self._make_persist_job(output_video_url="https://temp.mk/x.mp4")
+        session, fake_client, storage, patches = self._persist_patches(
+            job=job, download_side_effect=RuntimeError("network down")
+        )
+        entered = [p.__enter__() for p in patches]
+        try:
+            persist_output_video_task("job-1", "user-1", "https://temp.mk/x.mp4")
+        finally:
+            for p in reversed(patches):
+                p.__exit__(None, None, None)
+
+        storage.upload_file.assert_not_called()
+        # output_video_url 保持原值（temp_url）
+        assert job.output_video_url == "https://temp.mk/x.mp4"
+        # 内层 except 不会 commit
+        # 注：若内部发生 commit 说明测试失败
+        session.close.assert_called_once()
+
+    def test_empty_temp_url_skips_persist(self):
+        """temp_url 为空 → 直接返回，不下载不上传."""
+        from app.tasks.lipsync_tts import persist_output_video_task
+
+        job = self._make_persist_job(output_video_url="")
+        session, fake_client, storage, patches = self._persist_patches(job=job)
+        entered = [p.__enter__() for p in patches]
+        try:
+            persist_output_video_task("job-1", "user-1", "")
+        finally:
+            for p in reversed(patches):
+                p.__exit__(None, None, None)
+
+        fake_client.get.assert_not_called()
+        storage.upload_file.assert_not_called()
+        session.commit.assert_not_called()
+        session.close.assert_called_once()
+
+    def test_job_not_found_returns_early(self):
+        """DB 中找不到 job → 直接返回，不抛错."""
+        from app.tasks.lipsync_tts import persist_output_video_task
+
+        session, fake_client, storage, patches = self._persist_patches(job=None)
+        entered = [p.__enter__() for p in patches]
+        try:
+            persist_output_video_task("missing", "user-1", "https://temp.mk/x.mp4")
+        finally:
+            for p in reversed(patches):
+                p.__exit__(None, None, None)
+
+        fake_client.get.assert_not_called()
+        storage.upload_file.assert_not_called()
+        session.commit.assert_not_called()
+        session.close.assert_called_once()
+
+
+class TestLipsyncServiceRefreshCompletedAsyncPersist:
+    """refresh_job_status 在 completed 分支异步转存的单元测试（补 0% 覆盖的 316~335 行）."""
+
+    def test_refresh_completed_dispatches_persist_task(self):
+        """completed 分支：设置 temp_url → commit → dispatch persist_output_video_task.apply_async."""
+        from app.services.lipsync_service import LipsyncService
+
+        mock_job = MagicMock()
+        mock_job.id = "job-1"
+        mock_job.user_id = "user-1"
+        mock_job.mediakit_task_id = "mk-1"
+        mock_job.status = "submitted"
+        mock_job.output_video_url = ""
+        mock_job.output_duration = 0.0
+
+        mock_db = MagicMock()
+        mock_query = MagicMock()
+        mock_filter = MagicMock()
+        mock_filter.first.return_value = mock_job
+        mock_query.filter.return_value = mock_filter
+        mock_db.query.return_value = mock_query
+
+        mock_client = MagicMock()
+        mock_client.get_task_status.return_value = {
+            "status": "completed",
+            "result": {"video_url": "https://temp.mk/out.mp4", "duration": 25.5},
+        }
+
+        fake_persist_task = MagicMock()
+        svc = LipsyncService(mock_db, client=mock_client, cosyvoice_service=MagicMock())
+        with patch.dict("sys.modules", {}):
+            # 直接 patch 懒 import 路径
+            with patch("app.tasks.lipsync_tts.persist_output_video_task", fake_persist_task, create=False):
+                # 但懒 import 发生在函数内部 from app.tasks.lipsync_tts import persist_output_video_task
+                # 通过 patch sys.modules 的方式提供
+                import sys as _sys
+
+                fake_mod = MagicMock()
+                fake_mod.persist_output_video_task = fake_persist_task
+                _sys.modules["app.tasks.lipsync_tts"] = fake_mod
+                try:
+                    result = svc.refresh_job_status("job-1", "user-1")
+                finally:
+                    _sys.modules.pop("app.tasks.lipsync_tts", None)
+
+        assert result.status == "completed"
+        assert result.output_video_url == "https://temp.mk/out.mp4"
+        assert result.output_duration == 25.5
+        mock_db.commit.assert_called()
+        # 必须在 commit 之后 dispatch
+        fake_persist_task.apply_async.assert_called_once()
+        kwargs = fake_persist_task.apply_async.call_args.kwargs
+        assert kwargs["args"] == ("job-1", "user-1", "https://temp.mk/out.mp4")
+
+    def test_refresh_completed_dispatch_exception_does_not_break_return(self):
+        """apply_async 抛异常（如 Celery 不可用）→ 捕获 warning，仍返回 completed job."""
+        from app.services.lipsync_service import LipsyncService
+
+        mock_job = MagicMock()
+        mock_job.id = "job-2"
+        mock_job.user_id = "user-1"
+        mock_job.mediakit_task_id = "mk-2"
+        mock_job.status = "submitted"
+        mock_job.output_video_url = ""
+        mock_job.output_duration = 0.0
+
+        mock_db = MagicMock()
+        mock_query = MagicMock()
+        mock_filter = MagicMock()
+        mock_filter.first.return_value = mock_job
+        mock_query.filter.return_value = mock_filter
+        mock_db.query.return_value = mock_query
+
+        mock_client = MagicMock()
+        mock_client.get_task_status.return_value = {
+            "status": "completed",
+            "result": {"video_url": "https://temp.mk/out2.mp4", "duration": 10.0},
+        }
+
+        fake_persist_task = MagicMock()
+        fake_persist_task.apply_async.side_effect = ConnectionError("celery down")
+
+        svc = LipsyncService(mock_db, client=mock_client, cosyvoice_service=MagicMock())
+        import sys as _sys
+
+        fake_mod = MagicMock()
+        fake_mod.persist_output_video_task = fake_persist_task
+        _sys.modules["app.tasks.lipsync_tts"] = fake_mod
+        try:
+            result = svc.refresh_job_status("job-2", "user-1")
+        finally:
+            _sys.modules.pop("app.tasks.lipsync_tts", None)
+
+        # 即便 dispatch 失败，主流程不受影响：仍然返回 completed + temp_url
+        assert result.status == "completed"
+        assert result.output_video_url == "https://temp.mk/out2.mp4"
+        fake_persist_task.apply_async.assert_called_once()