fix(backend): #1896 修复字体映射补充开源字体文件

问题：DRAWTEXT_FONT_MAP把所有前端字体（思源宋体/苹方/微软雅黑/楷体/华康俪金黑）全部映射到 NotoSansSC，导致无论选什么字体成片都渲染为思源黑体；FONT_NAME_MAP 将楷体错误映射到 Noto Serif CJK SC（宋体）。

修复：
- packages/domain/video_filter_builder.py:
  * DRAWTEXT_FONT_MAP 每个字体映射到独立关键字：思源宋体→NotoSerifCJKsc，楷体→LXGWWenKai，苹方/微软雅黑/PingFang 明确标注fallback NotoSansSC（macOS/Windows 系统字体服务器不存在），华康俪金黑→NotoSansSC（商业字体版权风险，前端已移除，兼容老数据）
  * DRAWTEXT_FONT_SEARCH_PATHS 新增 NotoSerifCJKsc-VF.otf 和 LXGWWenKai-Regular.ttf 路径
- packages/domain/ass_subtitle_builder.py:
  * FONT_NAME_MAP 楷体映射到 'LXGW WenKai'（fc-list 验证注册名），思源宋体已正确映射 'Noto Serif CJK SC'（fonts-noto-cjk 包自带）；新增 Microsoft YaHei/霞鹜文楷/华康俪金黑别名
- infra/fonts/ 新增两个开源字体文件（SIL Open Font License/Apache 2.0 免费可商用）：
  * NotoSerifCJKsc-VF.otf（53MB，思源宋体可变字体，含多字重）
  * LXGWWenKai-Regular.ttf（25MB，霞鹜文楷开源楷体，v1.522）
- infra/docker/worker-base.Dockerfile + api-base.Dockerfile:
  * 保留 Serif .ttc（宋体fallback），仅删除 Sans .ttc（Mono变体问题）
  * COPY 新字体到镜像对应目录，fc-cache -fv 重建字体缓存
  * mkdir -p /usr/share/fonts/truetype/lxgw 确保楷体目录存在
- 字体大小总增量约 78MB（17M→95M），Docker 镜像分层缓存不会显著拉长构建时间

验证：
- 视频合成/ASS字幕/字幕滤镜三条渲染路径字体映射均独立正确
- fc-list 在本地实测可正确识别 'LXGW WenKai' / 'Noto Serif CJK SC' 注册名
- 苹方/微软雅黑等系统字体通过 NotoSansSC fallback 正常显示中文
- 存量数据（老数据选华康俪金黑/苹方等）继续渲染不中断
- 899 个字幕/滤镜/合成相关单测全通过，190 个writeback/filter/compose单测全通过

.gitea/workflows/ci-pipeline.yml

@@ -196,7 +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:
@@ -827,9 +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=""
@@ -1025,15 +1022,22 @@ jobs:
         BRANCH_TAG="${IMAGE_FULL}:${GITHUB_REF_NAME}"
         CACHE_REF="${REGISTRY}/${{ matrix.cache_name }}:${GITHUB_REF_NAME}"
 
-        EXTRA_BUILD_ARGS="APP_VERSION=\"${GITHUB_SHA}\""
-        if [ "${{ matrix.service }}" = "web" ]; then
-          EXTRA_BUILD_ARGS="$EXTRA_BUILD_ARGS NGINX_CONF=infra/docker/nginx-staging.conf"
+        # develop 分支额外推送 :dev tag（Watchtower 监听的滚动更新 tag）
+        if [ "${GITHUB_REF_NAME}" = "develop" ]; then
+          DEV_TAG="${IMAGE_FULL}:dev"
+          EXTRA_TAGS="$BRANCH_TAG $DEV_TAG"
+          PUSHED_TAGS_SUMMARY="${BRANCH_TAG} + ${DEV_TAG}"
+        else
+          EXTRA_TAGS="$BRANCH_TAG"
+          PUSHED_TAGS_SUMMARY="${BRANCH_TAG}"
         fi
 
+        EXTRA_BUILD_ARGS="APP_VERSION=\"${GITHUB_SHA}\""
+
         NO_CACHE_FLAG=""
         for i in 1 2 3; do
           echo "=== Docker build 尝试 $i/3 (${{ matrix.service_display }}) ==="
-          if EXTRA_TAGS="$BRANCH_TAG" bash scripts/ci/docker_build_push.sh $NO_CACHE_FLAG ${{ matrix.dockerfile }} "${IMAGE_TAG}" "${CACHE_REF}" $EXTRA_BUILD_ARGS; then
+          if EXTRA_TAGS="$EXTRA_TAGS" bash scripts/ci/docker_build_push.sh $NO_CACHE_FLAG ${{ matrix.dockerfile }} "${IMAGE_TAG}" "${CACHE_REF}" $EXTRA_BUILD_ARGS; then
             echo "✅ Docker build 成功"
             break
           fi
@@ -1046,7 +1050,7 @@ jobs:
           fi
         done
 
-        echo "${{ matrix.service_display }} image pushed: ${IMAGE_TAG} (+ ${BRANCH_TAG})"
+        echo "${{ matrix.service_display }} image pushed: ${IMAGE_TAG} (+ ${PUSHED_TAGS_SUMMARY})"
 
     - name: Job duration summary
       if: always()
@@ -1250,9 +1254,11 @@ jobs:
         ACR_PASSWORD: ${{ secrets.ACR_PASSWORD }}
       run: |
         set -eux
-        staging_host="${STAGING_SSH_HOST:-47.98.113.167}"
+        # CI runner (act_runner) 部署在 116 staging 本机（116.62.226.203 公网 22 未开放），
+        # 默认走 127.0.0.1:22 本机 SSH，避免跨机网络依赖；可通过 secrets 覆盖。
+        staging_host="${STAGING_SSH_HOST:-127.0.0.1}"
         staging_user="${STAGING_SSH_USER:-root}"
-        staging_port="${STAGING_SSH_PORT:-22222}"
+        staging_port="${STAGING_SSH_PORT:-22}"
         echo "Host: $staging_host"
         echo "Port: $staging_port"
 
@@ -1476,7 +1482,6 @@ jobs:
       - validate-security
       - validate-python
       - unit-tests
-      - frontend-lint
       - frontend-unit-test
     if: github.event_name == 'push' && github.ref_name == 'main' && !failure() && !cancelled()
     strategy:
@@ -1567,9 +1572,6 @@ 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=""
@@ -2133,3 +2135,4 @@ jobs:
           START_TIME=""
           [ -f /tmp/ci_job_start_time ] && START_TIME=$(cat /tmp/ci_job_start_time)
           curl -sfH "Authorization: token ${GITHUB_TOKEN:-$GITEA_TOKEN}" -o /tmp/_ci_trace.py "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/ci_trace_report.py?ref=${GITHUB_SHA}" 2>/dev/null && python3 /tmp/_ci_trace.py --service xiaoxia-saas-ci --status $STATUS --start-time "$START_TIME" || true
+# CI retry trigger

.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

EMPTY_COMMIT.txt

@@ -0,0 +1 @@
+retrigger3

README.md

@@ -263,3 +263,4 @@ pytest --cov=packages --cov-report=html
 ---
 
 **License**: MIT
+<!-- CI trigger: 1788229339 -->

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 +11,14 @@
 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,
-    SmartCoverRequest,
+    FinalizeRenderResponse,
     SmartCoverResponse,
 )
 from app.services.ai_avatar_cover_service import generate_smart_cover
@@ -77,10 +78,16 @@ 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 ─────────────────────────────────────────────────
@@ -172,38 +179,146 @@ 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 /smart-cover — 智能获取封面（MediaKit 抽帧 + 评分选帧）────────
+# ── POST /{job_id}/smart-cover — 从最终成片智能抽封面（步骤②）────────
 
 
-@router.post("/smart-cover", response_model=SmartCoverResponse)
-def generate_avatar_smart_cover(
-    body: SmartCoverRequest,
+@router.post("/{job_id}/smart-cover", response_model=SmartCoverResponse)
+def generate_render_smart_cover(
+    job_id: str,
     current_user: AuthenticatedUser = Depends(get_current_user),
-) -> SmartCoverResponse:
-    """智能获取数字人视频封面.
+    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）.
 
-    复用智能剪辑的 MediaKit 抽帧 + 质量评分选最佳帧逻辑（非 FFmpeg 简单截帧），
-    并将选中帧转存到自家 OSS，返回非临时的封面公网 URL。
-
-    前端「智能获取封面」按钮可直接调用本接口；不依赖渲染任务完成。
+    - 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
     """
-    video_url = (body.video_url or "").strip()
-    if not video_url.startswith(("http://", "https://")):
-        raise HTTPException(status_code=400, detail="video_url 必须是合法的 HTTP/HTTPS 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 = ""
 
-    cover_url = generate_smart_cover(video_url, max_frames=body.max_frames)
     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",
         )
-    logger.info("智能封面生成成功: user=%s", current_user.user.id)
+
+    # 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")
+
+# ── POST /{job_id}/finalize — 封面选定后正式入库成片库 ────────────────────
+
+
+@router.post("/{job_id}/finalize", response_model=FinalizeRenderResponse)
+def finalize_render_job(
+    job_id: str,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    db: Session = Depends(get_db_session),
+):
+    """用户完成封面选择后，将视频正式保存到成片库.
+
+    - 必须等渲染任务 completed 后才可调用
+    - 如果已通过 smart-cover/custom-cover 设置了封面，会自动带上
+    - 返回成片库视频ID
+    - 幂等：已 finalize 的任务重复调用会返回 existing 记录
+    """
+    from app.services.ai_avatar_render_service import AiAvatarRenderError, 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")
+
+    # 幂等检查（通过 generation_task_id=job_id 识别，finalize_job 内部也做了一次，这里提前返回简化）
+    from packages.adapters.sqlalchemy_impl.models import GeneratedVideoModel
+
+    existing = (
+        db.query(GeneratedVideoModel)
+        .filter(
+            GeneratedVideoModel.user_id == current_user.user.id,
+            GeneratedVideoModel.generation_task_id == job_id,
+        )
+        .first()
+    )
+    if existing is not None:
+        return FinalizeRenderResponse(
+            video_id=existing.id,
+            cover_url=existing.thumbnail_url or "",
+            status="already_finalized",
+        )
+
+    try:
+        video = svc.finalize_job(job_id, current_user.user.id)
+        return FinalizeRenderResponse(
+            video_id=video.id,
+            cover_url=video.thumbnail_url or job.output_cover_url or "",
+            status="success",
+        )
+    except AiAvatarRenderError as exc:
+        status_map = {
+            "RenderJobNotFound": 404,
+            "RenderNotCompleted": 400,
+            "OutputVideoMissing": 400,
+        }
+        raise HTTPException(
+            status_code=status_map.get(exc.code, 400),
+            detail=str(exc),
+        ) from exc
+    except Exception as exc:
+        logger.error("渲染任务finalize失败: job_id=%s err=%s", job_id, exc, exc_info=True)
+        raise HTTPException(status_code=500, detail=f"保存到成片库失败: {str(exc)}") from exc

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_cover.py

@@ -766,7 +766,12 @@ def generate_cover(
             storage_svc = get_shared_storage_service()
             mk_client = get_mediakit_client()
             # 从 plan.config 读取完整标题样式，E2 从源素材抽帧时叠加（源素材本身无标题）
+            # #1901 统一读 "title"，兼容老数据 "title_config"
             _e2_title_cfg = (plan.config or {}).get("title", {}) or {}
+            if not isinstance(_e2_title_cfg, dict) or not (_e2_title_cfg.get("text") or "").strip():
+                _alt = (plan.config or {}).get("title_config", {}) or {}
+                if isinstance(_alt, dict):
+                    _e2_title_cfg = _alt
             if not isinstance(_e2_title_cfg, dict):
                 _e2_title_cfg = {}
             _e2_title_text = (_e2_title_cfg.get("text", "") or "").strip() if _e2_title_cfg.get("enabled", True) else ""

apps/api/app/api/routes/generation_tasks.py

@@ -58,27 +58,10 @@ def _variant_value(values: list[str], index: int, fallback: str = "") -> str:
 
 
 def _query_voice_durations(db: Session, voice_ids: list[str]) -> list[float]:
-    """批量查询配音素材时长（秒），#1749 配音时长分配用。
+    """[已下沉] 路由层兼容别名 → app.services.generation_common.query_voice_durations。"""
+    from app.services.generation_common import query_voice_durations
 
-    逐项 try/float 硬化：MagicMock/异常/缺失 → 0.0（无配音不分配，不阻断）。
-    """
-    ids = [v for v in dict.fromkeys(voice_ids or []) if v]
-    if not ids:
-        return []
-    try:
-        from packages.adapters.sqlalchemy_impl.models import AssetModel
-
-        rows = db.query(AssetModel.id, AssetModel.duration).filter(AssetModel.id.in_(ids)).all()
-        dur_map: dict[str, float] = {}
-        for row in rows:
-            try:
-                dur_map[row[0]] = float(row[1] or 0.0)
-            except (TypeError, ValueError):
-                dur_map[row[0]] = 0.0
-        return [dur_map.get(v, 0.0) for v in ids]
-    except Exception:
-        logger.warning("[生成任务] 配音时长查询失败（按无配音处理，不阻断）", exc_info=True)
-        return [0.0 for _ in ids]
+    return query_voice_durations(db, voice_ids)
 
 
 def _to_generation_task_response(task) -> GenerationTaskResponse:
@@ -184,61 +167,10 @@ def _writeback_edit_plan_config(
     title_config: dict | None,
     db: Session,
 ) -> None:
-    """任务入队成功后，回写 EditPlan.config：generation_task_id + title_config。
+    """[已下沉] 路由层兼容别名 → app.services.generation_common.writeback_edit_plan_config。"""
+    from app.services.generation_common import writeback_edit_plan_config
 
-    用 merge 方式更新，不整体覆盖 config，避免丢失其他字段。
-    失败只记日志，不影响任务创建。
-    """
-    if not plan_id:
-        return
-    try:
-        from packages.adapters.sqlalchemy_impl.models import EditPlanModel
-
-        plan_model = db.query(EditPlanModel).filter(EditPlanModel.id == plan_id).first()
-        if plan_model is None:
-            logger.warning("[生成任务] 回写plan.config失败: plan不存在 plan_id=%s", plan_id)
-            return
-
-        current_config = plan_model.config if isinstance(plan_model.config, dict) else {}
-        merged = dict(current_config)
-        merged["generation_task_id"] = task_id
-
-        # 检查标题是否发生变化，如果变化则清除 cover 字段强制重新生成封面
-        if title_config:
-            old_title_config = merged.get("title_config", {}) or {}
-            old_title_text = (old_title_config.get("text") or "").strip()
-            new_title_text = (title_config.get("text") or "").strip()
-            if old_title_text != new_title_text:
-                # 标题变化，清除旧封面
-                if "cover" in merged:
-                    del merged["cover"]
-                    logger.info(
-                        "[生成任务] 标题变化，清除旧封面: plan_id=%s old_title=%s new_title=%s",
-                        plan_id,
-                        old_title_text,
-                        new_title_text,
-                    )
-            merged["title_config"] = title_config
-
-        plan_model.config = merged
-        db.commit()
-        logger.info(
-            "[生成任务] 回写plan.config成功: plan_id=%s task_id=%s keys=%s",
-            plan_id,
-            task_id,
-            list(merged.keys()),
-        )
-    except Exception as e:
-        logger.warning(
-            "[生成任务] 回写plan.config异常（不影响任务创建）: plan_id=%s error=%s",
-            plan_id,
-            e,
-            exc_info=True,
-        )
-        try:
-            db.rollback()
-        except Exception:
-            pass
+    return writeback_edit_plan_config(plan_id, task_id, title_config, db)
 
 
 def _resolve_project_and_library(
@@ -489,25 +421,14 @@ def create_generation_task(
         # 各变体配音时长（查询硬化：异常 → 0.0 不阻断）
         voice_durations = _query_voice_durations(db, variant_voices)
 
-        # 解析批量源 plan：优先前端传入；否则按 template_id + user 查最新（与单任务兜底同源）
-        batch_source_plan_id = request.source_edit_plan_id
-        if not batch_source_plan_id and request.template_id:
-            try:
-                from packages.adapters.sqlalchemy_impl.models import EditPlanModel
+        # 解析批量源 plan：优先前端传入；否则按 template_id + user 查最新（公共函数）
+        from app.services.generation_common import resolve_latest_plan_by_template
 
-                _latest = (
-                    db.query(EditPlanModel)
-                    .filter(
-                        EditPlanModel.template_id == request.template_id,
-                        EditPlanModel.created_by_user_id == user_id,
-                    )
-                    .order_by(EditPlanModel.created_at.desc())
-                    .first()
-                )
-                if _latest:
-                    batch_source_plan_id = _latest.id
-            except Exception:
-                logger.warning("[生成任务] 批量源 plan 解析失败", exc_info=True)
+        batch_source_plan_id = (
+            request.source_edit_plan_id
+            or resolve_latest_plan_by_template(db, template_id=request.template_id, user_id=user_id)
+            or ""
+        )
 
         if not batch_source_plan_id and not request.variant_plan_ids:
             # 无任何可用源 plan：批量变体无从选片，明确报错，严禁静默共用/同源
@@ -552,7 +473,12 @@ def create_generation_task(
                 ) from clone_err
             variant_plan_ids.append(_plan0.id)
 
-            # 变体 1..N-1 独立选片
+            # #1855 P0：批次区间避让表，从变体0实际clips构建初始值（公共函数）
+            from app.services.generation_common import collect_plan_segments as _collect_segments
+
+            _batch_segments = _collect_segments(_plan0.id, _plan_svc._clip_repo)
+
+            # 变体 1..N-1 独立选片（传入累积batch_segments做素材区间避让）
             for task_index in range(1, count):
                 variant = None
                 last_err: Exception | None = None
@@ -564,6 +490,7 @@ def create_generation_task(
                             created_by_user_id=user_id,
                             name_suffix=f"批量{task_index + 1}",
                             voice_duration=voice_durations[task_index] if task_index < len(voice_durations) else 0.0,
+                            batch_segments=_batch_segments,
                         )
                         break
                     except ValueError as ve:
@@ -595,7 +522,16 @@ def create_generation_task(
                     ) from last_err
                 variant_plan_ids.append(variant.id)
 
-        # ③ 配音时长分配（回传 plan / clone 变体0 均需幂等分配；reselect 已在选片时分配）
+                # #1855 P0：把新变体的clips区间追加到batch_segments，供下一变体避让
+                try:
+                    _new_segs = _collect_segments(variant.id, _plan_svc._clip_repo)
+                    for _aid, _ivs in _new_segs.items():
+                        _batch_segments.setdefault(_aid, []).extend(_ivs)
+                except Exception:
+                    logger.exception("[生成任务] 变体%d 区间收集失败（不阻断）", task_index)
+
+        # ③ 配音时长分配（回传 plan / clone 变体0 均需幂等分配；reselect 已在选片时分配，
+        #    #1855：apply_voice_duration_to_plan 已内置幂等判断，重复调用安全）
         for _vi, _pid in enumerate(variant_plan_ids):
             _vd = voice_durations[_vi] if _vi < len(voice_durations) else 0.0
             if _vd > 0:
@@ -616,24 +552,13 @@ def create_generation_task(
             )
             _single_vd: list[float] = _query_voice_durations(db, _voices)
             _single_dur = _single_vd[0] if _single_vd else 0.0
-            _single_plan = request.source_edit_plan_id
-            if not _single_plan and request.template_id:
-                try:
-                    from packages.adapters.sqlalchemy_impl.models import EditPlanModel
+            from app.services.generation_common import resolve_latest_plan_by_template
 
-                    _latest = (
-                        db.query(EditPlanModel)
-                        .filter(
-                            EditPlanModel.template_id == request.template_id,
-                            EditPlanModel.created_by_user_id == user_id,
-                        )
-                        .order_by(EditPlanModel.created_at.desc())
-                        .first()
-                    )
-                    if _latest:
-                        _single_plan = _latest.id
-                except Exception:
-                    logger.warning("[生成任务] 单任务源 plan 解析失败", exc_info=True)
+            _single_plan = (
+                request.source_edit_plan_id
+                or resolve_latest_plan_by_template(db, template_id=request.template_id, user_id=user_id)
+                or ""
+            )
             if _single_dur > 0 and _single_plan:
                 from app.services.edit_plan_service import EditPlanService
 

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

@@ -90,25 +90,12 @@ def create_variant_plans(
     except VariantVoiceError as exc:
         raise HTTPException(status_code=400, detail=str(exc)) from exc
 
-    # 解析源 plan：显式传入优先；否则按 template_id + user 查最新
+    # 解析源 plan：显式传入优先；否则按 template_id + user 查最新（公共函数）
+    from app.services.generation_common import resolve_latest_plan_by_template
+
     source_plan_id = request.source_edit_plan_id.strip()
     if not source_plan_id and request.template_id.strip():
-        try:
-            from packages.adapters.sqlalchemy_impl.models import EditPlanModel
-
-            _latest = (
-                db.query(EditPlanModel)
-                .filter(
-                    EditPlanModel.template_id == request.template_id.strip(),
-                    EditPlanModel.created_by_user_id == user_id,
-                )
-                .order_by(EditPlanModel.created_at.desc())
-                .first()
-            )
-            if _latest:
-                source_plan_id = _latest.id
-        except Exception:
-            logger.warning("[variant-plans] 源 plan 解析失败", exc_info=True)
+        source_plan_id = resolve_latest_plan_by_template(db, template_id=request.template_id, user_id=user_id) or ""
 
     if not source_plan_id:
         raise HTTPException(
@@ -122,7 +109,7 @@ 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 +130,7 @@ 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
@@ -14,18 +15,20 @@ 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.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, BackgroundTasks, Depends, HTTPException, Query
 from sqlalchemy.orm import Session
 
-from packages.application.cosyvoice_service import CosyVoiceError, CosyVoiceService
-
 logger = logging.getLogger(__name__)
 
 router = APIRouter()
@@ -34,13 +37,10 @@ router = APIRouter()
 def _get_service(
     db: Session = Depends(get_db_session),
     voice_clone_repo=Depends(get_voice_clone_profile_repository),
-    cosyvoice_service: CosyVoiceService = Depends(get_cosyvoice_service),
 ) -> LipsyncService:
-    # voice_clone_repo 用于克隆音色 profile 解析；cosyvoice_service 用于 TTS 直生
-    # （TTS 合成、音色解析、错误码归一化都在 LipsyncService 内部完成）
+    # voice_clone_repo 用于克隆音色 profile 解析
     return LipsyncService(
         db,
-        cosyvoice_service=cosyvoice_service,
         voice_clone_repo=voice_clone_repo,
     )
 
@@ -56,15 +56,20 @@ def create_lipsync_job(
 ):
     """提交对口型任务.
 
-    #1809/#1822: 前端传 {video_url, voice_id, script_text, speed?, emotion?}，
-    后端内部解析音色、调 TTS 合成音频、转存 OSS，再提交 MediaKit；
-    也支持直接传 {video_url, audio_url}。
+    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）。
     """
     try:
         job = svc.create_job(
             user_id=current_user.user.id,
             video_url=body.video_url,
             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,
@@ -73,23 +78,13 @@ def create_lipsync_job(
             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:
-        # TTS 合成失败 / 音色无权访问 → 400/403；MediaKit 提交失败 → 502
         status_code = 502
         if exc.code in ("VoiceForbidden",):
             status_code = 403
         elif exc.code in ("InvalidInput", "TTSInvalidParam", "VoiceNotReady"):
             status_code = 400
-        elif exc.code == "TTSSynthesisFailed":
-            status_code = 502
         raise HTTPException(
             status_code=status_code,
             detail={
@@ -99,7 +94,6 @@ def create_lipsync_job(
             },
         ) 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,
@@ -109,6 +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 ─────────────────────────────────────────────────
 
 
@@ -147,11 +187,7 @@ def get_lipsync_job(
     current_user: AuthenticatedUser = Depends(get_current_user),
     svc: LipsyncService = Depends(_get_service),
 ):
-    """获取对口型任务详情.
-
-    非终态任务：先返回 DB 缓存，挂后台刷新（下次轮询拿到新状态），
-    避免 MediaKit 慢响应阻塞前端轮询。
-    """
+    """获取对口型任务详情."""
     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")
@@ -187,13 +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）."""
+    """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/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)
 
@@ -277,20 +277,25 @@ 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)
 
 
+_ALLOWED_PREVIEW_EMOTIONS = {"", "natural", "excited", "calm", "friendly"}
+
+
 @router.get("/{clone_id}/preview", response_model=VoiceClonePreviewResponse)
 def get_voice_clone_preview(
     clone_id: str,
     text: str = Query("", description="自定义试听文本，为空则使用默认示例"),
+    speed: float = Query(1.0, ge=0.5, le=2.0, description="语速，0.5-2.0，默认 1.0"),
+    emotion: str = Query("", description="情绪：natural/excited/calm/friendly，空字符串为默认自然"),
     authenticated_user: AuthenticatedUser = Depends(get_current_user),
     repository: SQLAlchemyVoiceCloneProfileRepository = Depends(get_voice_clone_profile_repository),
     cosyvoice: CosyVoiceService = Depends(get_cosyvoice_service),
@@ -298,11 +303,17 @@ def get_voice_clone_preview(
     """获取克隆音色试听音频（实时 TTS 合成）。
 
     - 克隆音色必须处于 ready 状态
-    - 使用默认试听文本时，结果缓存 7 天
-    - 可传入自定义 text 参数试听不同文本
+    - 使用默认试听文本时，结果缓存 7 天（仅默认 text+speed=1.0+emotion=空 组合缓存）
+    - 可传入自定义 text/speed/emotion 试听不同效果
     """
     import time
 
+    if emotion not in _ALLOWED_PREVIEW_EMOTIONS:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail=f"不支持的 emotion 值: {emotion}，可选: natural/excited/calm/friendly 或留空",
+        )
+
     use_case = GetVoiceCloneUseCase(repository)
     try:
         profile = use_case.execute(clone_id, authenticated_user.user.id)
@@ -315,8 +326,8 @@ def get_voice_clone_preview(
             detail=f"Voice clone is not ready (current status: {profile.status})",
         )
 
-    # 有自定义文本时不缓存
-    use_cache = not text.strip()
+    # 仅默认试听文本 + 默认 speed + 默认 emotion 时使用缓存
+    use_cache = (not text.strip()) and abs(speed - 1.0) < 1e-6 and (not emotion)
 
     if use_cache and clone_id in _clone_preview_cache:
         audio_url, duration, file_size, cached_text, cached_at = _clone_preview_cache[clone_id]
@@ -337,12 +348,15 @@ def get_voice_clone_preview(
             text=preview_text,
             voice_id=profile.voice_id,
             format="mp3",
-            speed=1.0,
+            speed=speed,
+            emotion=emotion,
         )
     except CosyVoiceError as e:
         raise HTTPException(status_code=502, detail=f"TTS 合成失败: {e}") from e
+    except ValueError as e:
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=str(e)) from e
 
-    # 缓存（仅默认试听文本）
+    # 缓存（仅默认参数组合）
     if use_cache:
         _clone_preview_cache[clone_id] = (
             result.audio_url,

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

@@ -52,7 +52,9 @@ class CreateAiAvatarRenderRequest(BaseModel):
     lipsync_job_id: str = Field(..., description="对口型任务 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")
 
@@ -67,7 +69,6 @@ class CreateAiAvatarRenderRequest(BaseModel):
     @field_validator("script_id")
     @classmethod
     def validate_script_id(cls, v: str) -> str:
-        # script_id 可选：手动输入文案（TTS 直生）场景不关联文案库条目
         return (v or "").strip()
 
 
@@ -109,16 +110,17 @@ class AiAvatarRenderProgressResponse(BaseModel):
     error_message: str
 
 
-class SmartCoverRequest(BaseModel):
-    """智能封面请求 — MediaKit 抽帧 + 质量评分选最佳帧."""
-
-    video_url: str = Field(..., description="数字人视频 URL（对口型/渲染成片）")
-    max_frames: int = Field(5, ge=1, le=10, description="抽帧数量（默认 5）")
-
-
 class SmartCoverResponse(BaseModel):
-    """智能封面响应."""
+    """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）")
+
+
+class FinalizeRenderResponse(BaseModel):
+    """封面选好后点「完成」，正式入库成片库的响应."""
+
+    video_id: str = Field(..., description="成片库视频ID")
+    cover_url: str = Field("", description="封面URL")
+    status: str = Field("success", description="success/already_finalized")

apps/api/app/schemas/lipsync.py

@@ -1,9 +1,12 @@
-"""对口型 API Schema 定义 — #1796 / #1809 / #1822.
+"""对口型 API Schema 定义 — #1796 / #1809 / #1822 / #1845（配音前置）.
 
-支持两种输入模式（二选一）：
-1. TTS 直生模式（推荐）：传 voice_id + script_text（+ speed/emotion），
-   后端内部先调 CosyVoice 合成音频，再提交 MediaKit 对口型。
+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
@@ -33,6 +36,7 @@ class LipsyncJobResponse(BaseModel):
     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
@@ -45,15 +49,19 @@ class LipsyncJobResponse(BaseModel):
 class CreateLipsyncJobRequest(BaseModel):
     """创建对口型任务请求.
 
-    两种模式（二选一）：
-    - TTS 直生：voice_id + script_text 必填（+ 可选 speed/emotion）；audio_url 留空。
+    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）")
 
-    # 模式 2：直接音频
+    # 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）")
@@ -61,7 +69,9 @@ class CreateLipsyncJobRequest(BaseModel):
     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(False, description="音频长于视频时是否循环画面")
+    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）")
 
     @model_validator(mode="after")
@@ -72,15 +82,16 @@ class CreateLipsyncJobRequest(BaseModel):
         if not video.startswith(("http://", "https://")):
             raise ValueError("video_url 必须是 HTTP/HTTPS URL")
         lower = video.lower().split("?")[0]
-        if not lower.endswith(".mp4"):
-            raise ValueError("video_url 仅支持 MP4 格式")
+        allowed_video_exts = (".mp4", ".mov", ".m4v", ".webm", ".avi", ".mkv", ".3gp")
+        if not any(lower.endswith(ext) for ext in allowed_video_exts):
+            raise ValueError("video_url 格式不支持，仅支持: " + ", ".join(allowed_video_exts))
 
         has_audio = bool((self.audio_url or "").strip())
         has_tts = bool((self.voice_id or "").strip()) and bool((self.script_text or "").strip())
 
         if not has_audio and not has_tts:
             raise ValueError(
-                "必须提供驱动音频：要么传 audio_url（直接音频模式），"
+                "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）"
             )
 
@@ -98,3 +109,23 @@ class CreateLipsyncJobRequest(BaseModel):
             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/services/ai_avatar_cover_service.py

@@ -1,10 +1,13 @@
-"""AI 数字人封面服务 — 复用智能剪辑的 MediaKit 抽帧 + 质量评分选最佳帧.
+"""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。
 """
 
@@ -15,22 +18,56 @@ 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=2s × max_poll=30 → 最长 60s（与 mediakit_client 默认值/lipsync 轮询保持一致，防止合成视频下载+抽帧超时）
+COVER_POLL_INTERVAL = 2.0
+COVER_MAX_POLL_ATTEMPTS = 30
 
-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.
+# 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
 
-    Args:
-        video_url: 可公网访问的视频 URL
-        max_frames: 抽帧数量
+# 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
 
-    Returns:
-        最佳帧图片 URL；失败返回空字符串
+
+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
@@ -40,13 +77,19 @@ def select_best_cover_frame(video_url: str, *, max_frames: int = 5) -> str:
             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=2.0,
-            max_poll_attempts=5,
-            max_retries=0,
+            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])
@@ -55,24 +98,25 @@ def select_best_cover_frame(video_url: str, *, max_frames: int = 5) -> str:
         if len(snapshots) == 1:
             return snapshots[0].get("image_url") or snapshots[0].get("url") or ""
 
-        # 下载各帧评分
         import httpx
 
         candidates = []
-        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 = httpx.get(url, timeout=15, follow_redirects=True)
-                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:
-                candidates.append({"image_path": None, "url": url, "score": 0.0})
+        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 ""
@@ -81,7 +125,6 @@ def select_best_cover_frame(video_url: str, *, max_frames: int = 5) -> str:
         best = scored[0] if scored else None
         best_url = best.get("url", "") if best else ""
 
-        # 清理临时文件
         for c in candidates:
             p = c.get("image_path")
             if p:
@@ -102,16 +145,15 @@ def select_best_cover_frame(video_url: str, *, max_frames: int = 5) -> str:
         return ""
 
 
-def persist_cover_to_oss(frame_url: str, *, job_id: str = "", prefix: str = "ai-avatar/covers") -> str:
-    """下载帧图并转存到 OSS，返回公网封面 URL.
+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）.
 
-    Args:
-        frame_url: MediaKit 返回的临时帧图 URL
-        job_id: 关联任务 ID（用于 OSS key 命名）
-        prefix: OSS key 前缀
-
-    Returns:
-        OSS 公网 URL；失败回退原始 frame_url
+    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 ""
@@ -119,27 +161,33 @@ def persist_cover_to_oss(frame_url: str, *, job_id: str = "", prefix: str = "ai-
     try:
         import httpx
 
-        resp = httpx.get(frame_url, timeout=30, follow_redirects=True)
-        resp.raise_for_status()
-        if not resp.content:
-            return frame_url
+        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
+            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)
-        return public_url or frame_url
+        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
@@ -151,10 +199,15 @@ def persist_cover_to_oss(frame_url: str, *, job_id: str = "", prefix: str = "ai-
                 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，返回封面公网 URL.
+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。
 
-    供独立封面接口与渲染管线复用。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:

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
 
@@ -196,9 +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:
@@ -228,27 +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()
@@ -257,57 +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
 
-                # 封面：优先复用智能剪辑的 MediaKit 抽帧 + 质量评分选最佳帧；
-                # MediaKit 不可用时回退到 FFmpeg 已按 cover_config 抽取的 cover_path
-                smart_cover_url = ""
-                if output_video_url:
-                    try:
-                        from app.services.ai_avatar_cover_service import (
-                            generate_smart_cover,
-                        )
-
-                        smart_cover_url = generate_smart_cover(output_video_url, job_id=job_id, max_frames=5)
-                    except Exception:
-                        logger.warning("智能封面(MediaKit)失败，回退 FFmpeg 封面 job_id=%s", job_id, exc_info=True)
-
-                if smart_cover_url:
-                    job.output_cover_url = smart_cover_url
-                elif 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
@@ -322,18 +399,105 @@ class AiAvatarRenderService:
             self.db.commit()
             logger.info("渲染任务完成: %s", job_id)
 
+            # 7. 渲染完成，停留在「待选封面」状态：不自动入库。
+            #    用户在前端选好封面、点「完成」后，由 /{job_id}/finalize 接口显式入库。
+            logger.info("渲染任务完成，等待用户选择封面后入库: job_id=%s", job_id)
+
         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 _persist_to_library(self, job: AiAvatarRenderJob, cover_url: Optional[str] = None):
+        """将渲染结果写入成片库，返回 GeneratedVideo 领域对象.
+
+        Args:
+            job: 渲染任务（必须 status=completed 且 output_video_url 非空）
+            cover_url: 可选的封面 URL 覆盖（finalize 时传入即优先使用，否则取 job.output_cover_url）
+        """
+        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 用 render_job_id 便于反查。
+        clip_project_id = (job.project_id or "").strip() or "ai_avatar"
+        clip_generation_task_id = job.id
+        effective_cover = (cover_url or "").strip() if cover_url else (job.output_cover_url or "").strip()
+        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=effective_cover or None,
+            generation_params={
+                "source": "ai_avatar_render",
+                "render_job_id": job.id,
+            },
+        )
+        video_repo = SQLAlchemyGeneratedVideoRepository(self.db)
+        video_repo.create(clip)
+        logger.info("[数字人渲染] 成片已入库: clip_id=%s render_job=%s", clip.id, job.id)
+        return clip
+
+    def finalize_job(self, job_id: str, user_id: str, cover_url: Optional[str] = None):
+        """用户在前端点「完成」后调用：将已 completed 的渲染任务正式入库到成片库.
+
+        - 必须 status=completed 才可调用
+        - cover_url 若传入则优先使用并回写 job.output_cover_url；否则使用 job.output_cover_url（smart-cover/custom-cover 已写入）
+        - 幂等：已入库则返回已存在的 GeneratedVideo
+        """
+        from packages.adapters.sqlalchemy_impl.generated_video_repository import (
+            SQLAlchemyGeneratedVideoRepository,
+        )
+        from packages.adapters.sqlalchemy_impl.models import GeneratedVideoModel
+
+        job = self.get_render_job(job_id, user_id)
+        if job is None:
+            raise AiAvatarRenderError("渲染任务不存在", code="RenderJobNotFound")
+        if job.status != "completed":
+            raise AiAvatarRenderError(f"渲染任务未完成（当前状态: {job.status}），无法入库", code="RenderNotCompleted")
+        if not (job.output_video_url or "").strip():
+            raise AiAvatarRenderError("渲染成片视频 URL 为空，无法入库", code="OutputVideoMissing")
+
+        # 幂等检查：已入库直接返回现有记录（通过 generation_task_id=job_id 识别，
+        # 因为入库时 generation_task_id 被设置为 render_job_id 自身）
+        existing = (
+            self.db.query(GeneratedVideoModel)
+            .filter(
+                GeneratedVideoModel.user_id == user_id,
+                GeneratedVideoModel.generation_task_id == job_id,
+            )
+            .first()
+        )
+        if existing is not None:
+            logger.info("[数字人渲染] finalize 幂等命中，返回已存在记录: clip_id=%s job_id=%s", existing.id, job_id)
+            return SQLAlchemyGeneratedVideoRepository(self.db).get(existing.id)
+
+        # 传入 cover_url 时回写到 job
+        if cover_url and cover_url.strip():
+            job.output_cover_url = cover_url.strip()
+            # 同步更新 cover_config，保持 smart-cover 路径一致
+            if isinstance(job.cover_config, dict):
+                job.cover_config = {**job.cover_config, "mode": "auto_frame", "url": cover_url.strip()}
+            job.updated_at = datetime.now(timezone.utc)
+            self.db.commit()
+
+        return self._persist_to_library(job, cover_url=cover_url)
 
     def _download_video(self, url: str) -> str:
         """下载视频到临时文件."""
@@ -351,32 +515,132 @@ 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/edit_plan_service.py

@@ -473,6 +473,7 @@ class EditPlanService:
         name_suffix: str = "变体",
         voice_duration: float = 0.0,
         rng=None,
+        batch_segments: dict[str, list[tuple[float, float]]] | None = None,
     ) -> EditPlan:
         """为批量变体生成独立 plan：完整重跑单视频选片流程（#1743）。
 
@@ -489,6 +490,8 @@ class EditPlanService:
             created_by_user_id: 新 plan 归属用户。
             name_suffix: plan 名后缀。
             rng: 可选随机数（测试注入种子）。
+            batch_segments: 可选，外部传入的批次内已使用素材区间（前序变体避让用）。
+                传入时作为初始避让对象；未传则保持原逻辑从源 plan clips 自建（向后兼容）。
 
         Raises:
             ValueError: 源 plan 不存在/无片段、素材池为空或时长全未知。
@@ -537,16 +540,26 @@ class EditPlanService:
             voice = float(voice_duration or 0.0)
         except (TypeError, ValueError):
             voice = 0.0
+        rhythm_template_for_reselect = None
+        if source.config:
+            rhythm_template_for_reselect = source.config.get("rhythm_template")
+
         if voice > 0 and source_clips_data:
             from packages.domain.voice_duration_planner import plan_clip_durations
 
             _effects: list[str | None] = [c.get("transition_effect") for c in source_clips_data]
             _tdurs: list[float] = [float(c.get("transition_duration") or 0.0) for c in source_clips_data]
+            # #1855 P0：先占位durations为空dict，真正查durations在后面pool_ids确定后执行；
+            # plan_clip_durations 的 asset_durations 参数在该函数中仅作最大段长钳制，
+            # 这里先不依赖它（durations 还没查），传 None 让planner用默认策略；
+            # 真正的asset_durations会在后面 clips_data 生成时传入 reselect_clips_for_variant
             target_durations = plan_clip_durations(
                 len(source_clips_data),
                 voice,
                 transition_effects=_effects,
                 transition_durations=_tdurs,
+                rhythm_template=rhythm_template_for_reselect,
+                asset_durations=None,
             )
             if target_durations:
                 for _c, _d in zip(source_clips_data, target_durations, strict=False):
@@ -582,12 +595,18 @@ class EditPlanService:
             created_by_user_id=created_by_user_id or (source.created_by_user_id or ""),
         )
 
-        # 批次内区间：以源 plan（变体 0）片段为初始避让对象
-        batch_segments: dict[str, list[tuple[float, float]]] = {}
-        for c in clips:
-            if c.asset_id and float(c.duration or 0) > 0:
-                st = float(c.start_time or 0.0)
-                batch_segments.setdefault(c.asset_id, []).append((st, st + float(c.duration)))
+        # 批次内区间：外部传入时使用外部传入（含前序变体已用区间）；
+        # 否则保持原逻辑从源 plan clips 自建（向后兼容）
+        if batch_segments is not None:
+            batch_segments_resolved: dict[str, list[tuple[float, float]]] = {
+                k: list(v) for k, v in batch_segments.items()
+            }
+        else:
+            batch_segments_resolved = {}
+            for c in clips:
+                if c.asset_id and float(c.duration or 0) > 0:
+                    st = float(c.start_time or 0.0)
+                    batch_segments_resolved.setdefault(c.asset_id, []).append((st, st + float(c.duration)))
 
         clips_data = reselect_clips_for_variant(
             source_clips_data,
@@ -595,7 +614,7 @@ class EditPlanService:
             asset_durations=durations,
             asset_scene_points=scene_points,
             historical_used_segments=historical,
-            batch_segments=batch_segments,
+            batch_segments=batch_segments_resolved,
             target_durations=target_durations,
             rng=rng,
         )
@@ -767,6 +786,17 @@ class EditPlanService:
         if plan is None:
             return None
 
+        # #1855 P0：幂等判断——如果已成功分配过且当前 total_duration 已接近 voice_duration，直接返回
+        try:
+            existing_mark = None
+            if plan.config:
+                existing_mark = plan.config.get("voice_duration_applied")
+            cur_total = float(plan.total_duration or 0.0)
+            if existing_mark is not None and abs(existing_mark - voice) < 1e-6 and abs(cur_total - voice) < 0.5:
+                return plan
+        except Exception:
+            pass
+
         clips: List[EditPlanClip] = []
         skip, page = 0, 500
         while True:
@@ -838,6 +868,10 @@ class EditPlanService:
         )
         try:
             plan.total_duration = net
+            # #1855 P0：写入幂等标记，避免二次调用时只重分配 duration 不重算 start_time
+            new_cfg = dict(plan.config or {})
+            new_cfg["voice_duration_applied"] = voice
+            plan.config = new_cfg
             db = self._clip_repo.session
             db.commit()
         except Exception:
@@ -878,6 +912,69 @@ class EditPlanService:
         rng = rng or _random.Random()
         plan_ids: list[str] = []
 
+        # #1855 P0：先确定片段数 clip_count（用于节奏模板生成长度匹配）
+        from packages.domain.bgm_pool import allocate_bgm_pool_for_variants
+        from packages.domain.variant_plan_selector import (
+            generate_pixel_perturbation,
+            generate_visual_perturbation,
+        )
+        from packages.domain.voice_duration_planner import RHYTHM_TEMPLATES, adapt_template_length
+
+        clip_count = 0
+        # 从源 plan 获取片段数（分页读，避免关系加载问题）
+        _sclips: list = []
+        _sk, _pg = 0, 500
+        while True:
+            _b = self._clip_repo.list_by_plan(source_plan_id, skip=_sk, limit=_pg)
+            if not _b:
+                break
+            _sclips.extend(_b)
+            if len(_b) < _pg:
+                break
+            _sk += _pg
+        clip_count = len(_sclips)
+
+        # 预先生成所有 N 个变体的节奏模板/BGM/扰动参数（时机提前到选片前写入config）
+        rhythm_templates_for_variants: list = []
+        for _idx in range(count):
+            if clip_count > 0:
+                variant_seed = rng.randint(0, 999999)
+                _tpl = adapt_template_length(RHYTHM_TEMPLATES[variant_seed % len(RHYTHM_TEMPLATES)], clip_count)
+                rhythm_templates_for_variants.append(_tpl)
+            else:
+                rhythm_templates_for_variants.append(None)
+
+        source_bgm_config: dict = {}
+        source_plan = self.get_plan(source_plan_id)
+        if source_plan and source_plan.config:
+            source_bgm_config = source_plan.config.get("bgm", {}) or {}
+        variant_seeds_for_bgm = [rng.randint(0, 999999) for _ in range(count)]
+        bgm_pool_assignments = allocate_bgm_pool_for_variants(source_bgm_config, variant_seeds_for_bgm)
+
+        def _build_variant_config_update(idx: int) -> dict:
+            """构建单个变体的 config 更新（节奏模板/BGM/视觉/像素扰动）。"""
+            upd: dict = {}
+            try:
+                perturbation = generate_visual_perturbation(rng)
+                if idx == 0:
+                    perturbation["hflip"] = False
+                upd["visual_perturbation"] = perturbation
+            except Exception:
+                logger.exception("变体 %d 视觉扰动生成失败（不阻断）", idx)
+            try:
+                pixel_pert = generate_pixel_perturbation(rng)
+                upd["pixel_perturbation"] = pixel_pert
+            except Exception:
+                logger.exception("变体 %d 像素扰动生成失败（不阻断）", idx)
+            rt = rhythm_templates_for_variants[idx] if idx < len(rhythm_templates_for_variants) else None
+            if rt is not None:
+                upd["rhythm_template"] = rt
+            if idx < len(bgm_pool_assignments):
+                existing_bgm = dict((source_plan.config or {}).get("bgm", {}) or {})
+                existing_bgm.update(bgm_pool_assignments[idx])
+                upd["bgm"] = existing_bgm
+            return upd
+
         # 变体 0：clone（片段结构同源 plan，起点重算），不污染源 plan
         plan0 = self.clone_plan_for_variant(
             source_plan_id,
@@ -890,6 +987,15 @@ class EditPlanService:
                 v0_voice = float(voice_durations[0] or 0.0)
             except (TypeError, ValueError):
                 v0_voice = 0.0
+
+        # #1855 P0：在配音分配前先写入变体0的节奏模板/扰动/BGM，确保 apply_voice_duration_to_plan 能读到 rhythm_template
+        try:
+            _cfg0 = _build_variant_config_update(0)
+            if _cfg0:
+                self.update_plan_config(plan0.id, _cfg0)
+        except Exception:
+            logger.exception("变体0 配置写入失败（不阻断）: plan=%s", plan0.id)
+
         if v0_voice > 0:
             try:
                 self.apply_voice_duration_to_plan(plan0.id, v0_voice)
@@ -897,7 +1003,12 @@ class EditPlanService:
                 logger.exception("变体0 配音分配失败（不阻断）: plan=%s", plan0.id)
         plan_ids.append(plan0.id)
 
-        # 变体 1..N-1：独立选片
+        # #1855 P0：批次内素材区间避让表——从变体0实际落库的clips构建初始值（公共函数）
+        from app.services.generation_common import collect_plan_segments as _collect_plan_segments
+
+        batch_segments_acc: dict[str, list[tuple[float, float]]] = _collect_plan_segments(plan0.id, self._clip_repo)
+
+        # 变体 1..N-1：独立选片（传入累积的 batch_segments 做区间避让）
         for i in range(1, count):
             voice = 0.0
             if voice_durations and i < len(voice_durations):
@@ -905,6 +1016,14 @@ class EditPlanService:
                     voice = float(voice_durations[i] or 0.0)
                 except (TypeError, ValueError):
                     voice = 0.0
+
+            # #1855 P0：在reselect前先为"变体i"准备配置更新——但reselect内部复制的是source.config，
+            # 所以每个变体独立的节奏模板需要在reselect后单独写入config；
+            # 但 plan_clip_durations 用的是 source.config.rhythm_template（即源plan的节奏模板），
+            # 为了让每个变体在选片阶段就使用自己的节奏模板分配段长，这里采用：
+            #   - reselect 仍使用源 plan 的 rhythm_template（保持片段骨架一致）
+            #   - 选片完成后立即写入该变体自己的 rhythm_template/扰动/BGM 到config，
+            #     后续不再二次 apply_voice_duration_to_plan（由幂等标记跳过）
             variant = self.reselect_plan_for_variant(
                 source_plan_id,
                 candidate_asset_ids,
@@ -912,73 +1031,26 @@ class EditPlanService:
                 name_suffix=f"变体{i + 1}",
                 voice_duration=voice,
                 rng=rng,
+                batch_segments=batch_segments_acc,
             )
+
+            # 选片完成后写入该变体的独立配置（节奏模板/扰动/BGM）
+            try:
+                _cfgi = _build_variant_config_update(i)
+                if _cfgi:
+                    self.update_plan_config(variant.id, _cfgi)
+            except Exception:
+                logger.exception("变体 %d 配置写入失败（不阻断）: plan=%s", i, variant.id)
+
             plan_ids.append(variant.id)
 
-        # #1764：为每个变体生成独立节奏模板（让批量视频片段时长分布不同）
-        from packages.domain.voice_duration_planner import RHYTHM_TEMPLATES, adapt_template_length
-
-        clip_count = 0
-        if voice_durations and len(voice_durations) > 0:
-            # 从源 plan 获取片段数
-            source_plan = self.get_plan(source_plan_id)
-            if source_plan and hasattr(source_plan, "clips"):
-                clip_count = len(list(source_plan.clips)) if source_plan.clips else 0
-
-        rhythm_templates_for_variants = []
-        if clip_count > 0:
-            for idx in range(len(plan_ids)):
-                # 每个变体用不同的 seed 选择节奏模板
-                variant_seed = rng.randint(0, 999999)
-                template = adapt_template_length(RHYTHM_TEMPLATES[variant_seed % len(RHYTHM_TEMPLATES)], clip_count)
-                rhythm_templates_for_variants.append(template)
-                logger.info("变体 %d 节奏模板: plan=%s template=%s", idx, plan_ids[idx], template)
-
-        # #1767：BGM 池差异化分配（让批量变体使用不同 BGM / 段落 / 音量）
-        from packages.domain.bgm_pool import allocate_bgm_pool_for_variants
-
-        source_bgm_config = {}
-        source_plan = self.get_plan(source_plan_id)
-        if source_plan and source_plan.config:
-            source_bgm_config = source_plan.config.get("bgm", {}) or {}
-
-        variant_seeds_for_bgm = [rng.randint(0, 999999) for _ in plan_ids]
-        bgm_pool_assignments = allocate_bgm_pool_for_variants(source_bgm_config, variant_seeds_for_bgm)
-
-        # 为每个变体生成独立视觉扰动参数（让批量视频画面本身更不同）
-        from packages.domain.variant_plan_selector import generate_visual_perturbation
-
-        for idx, pid in enumerate(plan_ids):
+            # #1855 P0：把当前新变体的 clips 区间追加到 batch_segments，供下一变体避让
             try:
-                perturbation = generate_visual_perturbation(rng)
-                # 变体 0 不做 hflip（保持预览 plan 原始画面方向）
-                if idx == 0:
-                    perturbation["hflip"] = False
-                config_update = {"visual_perturbation": perturbation}
-                # #1764：写入节奏模板
-                if idx < len(rhythm_templates_for_variants):
-                    config_update["rhythm_template"] = rhythm_templates_for_variants[idx]
-                # #1765：写入像素级扰动滤镜
-                from packages.domain.variant_plan_selector import generate_pixel_perturbation
-
-                pixel_pert = generate_pixel_perturbation(rng)
-                config_update["pixel_perturbation"] = pixel_pert
-                # #1767：写入 BGM 池分配（覆盖 bgm 配置中的 preset_id / audio_offset / volume_adjust_db）
-                if idx < len(bgm_pool_assignments):
-                    existing_bgm = dict((source_plan.config or {}).get("bgm", {}) or {})
-                    existing_bgm.update(bgm_pool_assignments[idx])
-                    config_update["bgm"] = existing_bgm
-                self.update_plan_config(pid, config_update)
-                logger.info(
-                    "变体 %d 视觉扰动+像素扰动+BGM池: plan=%s vis=%s pix=%s bgm=%s",
-                    idx,
-                    pid,
-                    perturbation,
-                    pixel_pert,
-                    bgm_pool_assignments[idx] if idx < len(bgm_pool_assignments) else None,
-                )
+                _new_segs = _collect_plan_segments(variant.id, self._clip_repo)
+                for _aid, _ivs in _new_segs.items():
+                    batch_segments_acc.setdefault(_aid, []).extend(_ivs)
             except Exception:
-                logger.exception("变体 %d 视觉扰动生成失败（不阻断）: plan=%s", idx, pid)
+                logger.exception("变体 %d 区间收集失败（不阻断）: plan=%s", i, variant.id)
 
         # 标记所有变体 plan 的 clips 为 ready（已分配素材+起点，语义上就是 ready）
         for pid in plan_ids:

apps/api/app/services/generation_common.py

@@ -0,0 +1,189 @@
+"""智能剪辑公共服务辅助函数（从 route 层下沉）。
+
+集中管理：
+- query_voice_durations：批量查询配音素材时长
+- writeback_edit_plan_config：任务入队后回写 EditPlan.config
+- collect_plan_segments：分页读取 plan clips 构建素材区间表（变体避让用）
+- resolve_latest_plan_by_template：按 template_id + user_id 查最新 EditPlan
+
+设计原则：
+- 无副作用的纯查询 / 幂等写回；失败一律不阻断主流程（记日志 + 返回安全默认值）
+- 不依赖 FastAPI / HTTPException，便于 service 层和 worker 复用
+"""
+
+from __future__ import annotations
+
+import logging
+from typing import Any, Optional
+
+from sqlalchemy.orm import Session
+
+logger = logging.getLogger(__name__)
+
+
+def query_voice_durations(db: Session, voice_ids: list[str]) -> list[float]:
+    """批量查询配音素材时长（秒），#1749 配音时长分配用。
+
+    逐项 try/float 硬化：MagicMock/异常/缺失 → 0.0（无配音不分配，不阻断）。
+
+    #1855 P0修复：不再对 voice_ids 去重，保持与调用方传入顺序/长度一致，
+    允许同配音id多次出现时返回相同时长（支持"同配音N变体"的时长对齐）。
+    """
+    raw_ids = list(voice_ids or [])
+    if not raw_ids:
+        return []
+    unique_ids: list[str] = []
+    _seen: set[str] = set()
+    for v in raw_ids:
+        if v and v not in _seen:
+            _seen.add(v)
+            unique_ids.append(v)
+    if not unique_ids:
+        return [0.0 for _ in raw_ids]
+    try:
+        from packages.adapters.sqlalchemy_impl.models import AssetModel
+
+        rows = db.query(AssetModel.id, AssetModel.duration).filter(AssetModel.id.in_(unique_ids)).all()
+        dur_map: dict[str, float] = {}
+        for row in rows:
+            try:
+                dur_map[row[0]] = float(row[1] or 0.0)
+            except (TypeError, ValueError):
+                dur_map[row[0]] = 0.0
+        return [dur_map.get(v, 0.0) if v else 0.0 for v in raw_ids]
+    except Exception:
+        logger.warning("[generation_common] 配音时长查询失败（按无配音处理，不阻断）", exc_info=True)
+        return [0.0 for _ in raw_ids]
+
+
+def writeback_edit_plan_config(
+    plan_id: str,
+    task_id: str,
+    title_config: dict | None,
+    db: Session,
+) -> None:
+    """任务入队成功后，回写 EditPlan.config：generation_task_id + title_config。
+
+    用 merge 方式更新，不整体覆盖 config，避免丢失其他字段。
+    失败只记日志，不影响任务创建。
+    """
+    if not plan_id:
+        return
+    try:
+        from packages.adapters.sqlalchemy_impl.models import EditPlanModel
+
+        plan_model = db.query(EditPlanModel).filter(EditPlanModel.id == plan_id).first()
+        if plan_model is None:
+            logger.warning("[generation_common] 回写plan.config失败: plan不存在 plan_id=%s", plan_id)
+            return
+
+        current_config = plan_model.config if isinstance(plan_model.config, dict) else {}
+        merged = dict(current_config)
+        merged["generation_task_id"] = task_id
+
+        if title_config:
+            # #1901 统一字段名为 "title"（worker sync_configs_to_plan 写的是 "title"）
+            # 先读取新旧两个 key，判断标题文字是否变化
+            old_title_cfg = merged.get("title", {}) or {}
+            if not isinstance(old_title_cfg, dict) or not (old_title_cfg.get("text") or "").strip():
+                old_title_cfg = merged.get("title_config", {}) or {}
+            old_title_text = (old_title_cfg.get("text") or "").strip() if isinstance(old_title_cfg, dict) else ""
+            new_title_text = (title_config.get("text") or "").strip()
+            if old_title_text != new_title_text:
+                if "cover" in merged:
+                    del merged["cover"]
+                    logger.info(
+                        "[generation_common] 标题变化，清除旧封面: plan_id=%s old_title=%s new_title=%s",
+                        plan_id,
+                        old_title_text,
+                        new_title_text,
+                    )
+            # 字段名归一化（font_size→size, font_preset→font, font_color→color），与 worker sync_configs_to_plan 保持一致
+            normalized = dict(title_config)
+            if "font_size" in normalized and "size" not in normalized:
+                normalized["size"] = normalized["font_size"]
+            if "font_preset" in normalized and "font" not in normalized:
+                normalized["font"] = normalized["font_preset"]
+            if "font_color" in normalized and "color" not in normalized:
+                normalized["color"] = normalized["font_color"]
+            merged["title"] = normalized
+            # 清掉旧 key，避免双字段并存
+            merged.pop("title_config", None)
+
+        plan_model.config = merged
+        db.commit()
+        logger.info(
+            "[generation_common] 回写plan.config成功: plan_id=%s task_id=%s keys=%s",
+            plan_id,
+            task_id,
+            list(merged.keys()),
+        )
+    except Exception as e:
+        logger.warning(
+            "[generation_common] 回写plan.config异常（不影响任务创建）: plan_id=%s error=%s",
+            plan_id,
+            e,
+            exc_info=True,
+        )
+        try:
+            db.rollback()
+        except Exception:
+            pass
+
+
+def collect_plan_segments(
+    plan_id: str,
+    clip_repo: Any,
+    *,
+    page_size: int = 500,
+) -> dict[str, list[tuple[float, float]]]:
+    """分页读取 plan 所有 clips，构建 {asset_id: [(start, end), ...]} 素材区间表。
+
+    用于 #1855 P0 批次内素材区间避让（变体间素材片段重叠控制）。
+    """
+    segs: dict[str, list[tuple[float, float]]] = {}
+    sk, pg = 0, page_size
+    while True:
+        batch = clip_repo.list_by_plan(plan_id, skip=sk, limit=pg)
+        if not batch:
+            break
+        for c in batch:
+            if c.asset_id and float(c.duration or 0) > 0:
+                st = float(c.start_time or 0.0)
+                segs.setdefault(c.asset_id, []).append((st, st + float(c.duration)))
+        if len(batch) < pg:
+            break
+        sk += pg
+    return segs
+
+
+def resolve_latest_plan_by_template(
+    db: Session,
+    *,
+    template_id: str,
+    user_id: str,
+) -> Optional[str]:
+    """按 template_id + user_id 查找最新的 EditPlan.id（模板兜底用）。找不到返回 None。"""
+    if not (template_id or "").strip():
+        return None
+    try:
+        from packages.adapters.sqlalchemy_impl.models import EditPlanModel
+
+        latest = (
+            db.query(EditPlanModel)
+            .filter(
+                EditPlanModel.template_id == template_id.strip(),
+                EditPlanModel.created_by_user_id == user_id,
+            )
+            .order_by(EditPlanModel.created_at.desc())
+            .first()
+        )
+        return latest.id if latest else None
+    except Exception:
+        logger.warning(
+            "[generation_common] 按template查找最新plan失败: template=%s user=%s",
+            template_id,
+            user_id,
+            exc_info=True,
+        )
+        return None

apps/api/app/services/lipsync_service.py

@@ -1,8 +1,12 @@
-"""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/查询对口型任务
-- 双输入模式：TTS 直生（voice_id + script_text，内部先合成音频转存 OSS）或直接音频（audio_url）
+- 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（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）
@@ -25,17 +29,23 @@ 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, 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 排队 + 拉取可能延迟，私有桶裸 URL 或 1 小时短预签名都会 403，故统一重签长有效期。
 MEDIAKIT_URL_TTL_SECONDS = 7 * 24 * 3600
 
 
@@ -139,6 +149,100 @@ class LipsyncService:
             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 ──────────────────────────────────────────────────────────
 
     def create_job(
@@ -147,45 +251,46 @@ class LipsyncService:
         user_id: str,
         video_url: str,
         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 = False,
+        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.
 
-        两种输入模式：
-        - TTS 直生：voice_id + script_text（audio_url 留空），后端先合成音频
-        - 直接音频：提供 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:
-            MediaKitError: TTS 合成或 MediaKit 提交失败
+            MediaKitError: 参数校验失败或 MediaKit 提交失败
         """
-        # 0. TTS 直生模式：先合成音频（在创建 DB 记录之前完成，失败直接抛出）
-        if not audio_url:
+        # 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)
+
+        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",
                 )
-            # 预合成：用临时 job_id 命名 OSS 对象
-            pre_job_id = str(uuid.uuid4())
-            audio_url = self._synthesize_and_persist_audio(
-                user_id=user_id,
-                job_id=pre_job_id,
-                voice_id=voice_id,
-                script_text=script_text,
-                speed=speed,
-                emotion=emotion,
-            )
-
-        # #1839: 私有桶 OSS 的裸 URL / 即将过期的短预签名会让 MediaKit GPU worker 拉取时 403，
-        # 提交前统一重签长有效期；外部临时 URL（CosyVoice/MediaKit）原样透传。
-        video_url = self._sign_media_url(video_url)
-        if audio_url:
-            audio_url = self._sign_media_url(audio_url)
+            # 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())
@@ -199,34 +304,136 @@ class LipsyncService:
             voice_id=voice_id or "",
             script_text=script_text or "",
             speed=speed,
-            emotion=normalize_emotion(emotion),
-            status="pending",
+            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]:
@@ -260,11 +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
@@ -289,11 +492,25 @@ class LipsyncService:
         if mk_status == STATUS_COMPLETED:
             result = status_data.get("result", {})
             job.status = STATUS_COMPLETED
-            output_url = result.get("video_url", "")
-            # MediaKit 输出为临时 URL，转存自家 OSS 防止过期（失败则回退临时 URL）
-            job.output_video_url = self._persist_output_video(output_url, job_id, user_id)
+            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"
@@ -301,7 +518,6 @@ class LipsyncService:
             job.error_code = error.get("code", "TaskFailed")
             job.completed_at = datetime.now(timezone.utc)
         else:
-            # 中间状态（running/processing/queued 等）同步到 DB，避免前端永远卡在 submitted
             if isinstance(mk_status, str) and mk_status:
                 job.status = mk_status
         job.updated_at = datetime.now(timezone.utc)
@@ -310,10 +526,7 @@ class LipsyncService:
         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，不影响任务完成。
-        """
+        """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:
@@ -333,39 +546,33 @@ class LipsyncService:
             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，供 MediaKit 拉取 / 前端播放。
-
-        - 裸 public_url（upload_file 返回，不带签名）→ 私有桶匿名访问 403，重签。
-        - 已带签名但即将过期的 URL（如前端 1h 预签名）→ 抽 storage_key 后重签。
-        - 外部 URL（CosyVoice/MediaKit 临时链接，非本桶 host）→ 原样透传。
-        - 任何异常都降级原样返回，不阻断主流程。
-        """
+        """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  # 无法判定归属，保守透传
+                return url
             own_host = urlparse(public_base).netloc.lower()
             host = urlparse(url).netloc.lower()
             if not own_host or host != own_host:
-                return url  # 非自家 OSS（外部临时链接），不处理
+                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:  # noqa: BLE001 - 签名失败不阻断，降级原 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/generation/variantPlans.ts

@@ -26,6 +26,10 @@ export interface BatchVariantPlansRequest {
   count: number
   /** 源剪辑计划 ID：优先取预览/草稿关联的 plan；不传由后端按 template_id+user 兜底最新 plan */
   source_edit_plan_id?: string
+  /** 统一配音 ID（共用配音模式）；独立配音模式不传，改传 voice_library_ids */
+  voice_library_id?: string
+  /** 独立配音 ID 列表（长度=count，按变体序号一一对应）；共用配音模式不传 */
+  voice_library_ids?: string[]
 }
 
 /** 单个变体的计划片段 */
@@ -36,6 +40,8 @@ export interface VariantPlan {
   plan_id: string
   /** 该变体的真实片段（顺序/素材/起点与正式成片一致） */
   clips: EditPlanClip[]
+  /** 该变体实际配音时长（秒），用于前端预览按配音时长对齐音画；后端暂未返回时缺省 */
+  voice_duration?: number
 }
 
 /** 批量变体计划响应 */

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

@@ -16,6 +16,7 @@ export interface TTSSynthesizeRequest {
   voice_id?: string
   output_name?: string
   language?: string
+  emotion?: string
   speed?: number
   voice_model?: string
   voice_clone_profile_id?: string
@@ -103,6 +104,7 @@ export interface TTSPreviewRequest {
   voice_id: string
   speed?: number
   pitch?: number
+  language?: string
   emotion?: string // 情绪参数：natural/excited/calm/friendly
 }
 

apps/web/src/api/voice-clone/clones.ts

@@ -96,9 +96,12 @@ export const retryVoiceClone = async (id: string): Promise<VoiceCloneProfile> =>
 export const getVoiceClonePreview = async (
   cloneId: string,
   text?: string,
+  options?: { speed?: number; emotion?: string },
 ): Promise<VoiceClonePreviewResponse> => {
   const searchParams = new URLSearchParams()
   if (text) searchParams.set("text", text)
+  if (options?.speed !== undefined) searchParams.set("speed", String(options.speed))
+  if (options?.emotion) searchParams.set("emotion", options.emotion)
   const qs = searchParams.toString()
   const response = await apiClient.get<VoiceClonePreviewResponse>(
     `/voice-clones/${cloneId}/preview${qs ? `?${qs}` : ""}`,

apps/web/src/index.css

@@ -25,15 +25,6 @@ body {
   background-color: var(--bg-secondary);
 }
 
-/* ── 自定义字体 ── */
-@font-face {
-  font-family: "华康俪金黑";
-  src: url("/fonts/DFLiJinHei-W8.ttf") format("truetype");
-  font-weight: 700;
-  font-style: normal;
-  font-display: swap;
-}
-
 /* 滚动条 - V21 样式 */
 ::-webkit-scrollbar {
   width: 8px;

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

@@ -484,7 +484,6 @@
   margin-bottom: 12px;
 }
 
-
 .aa-lipsync-preview {
   width: 100%;
   max-width: 240px;
@@ -550,21 +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 {
@@ -572,6 +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;
@@ -1250,3 +1266,19 @@
   min-width: 300px;
 }
 
+/* 渲染完成后的封面确认区 */
+.aa-finalize-section {
+  display: flex;
+  flex-direction: column;
+  align-items: center;
+  padding: 16px 0 8px;
+}
+
+.aa-finalize-cover {
+  width: 100%;
+  max-width: 240px;
+  aspect-ratio: 9/16;
+  border-radius: 12px;
+  overflow: hidden;
+  position: relative;
+}

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

@@ -2,7 +2,7 @@
  * 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[]> => {
@@ -34,17 +34,28 @@ export const getAssetById = async (id: string): Promise<{ file_url?: string; id:
   return response.data
 }
 
-/* ── 对口型（模式A：TTS 直生，后端内部合成音频；不要先调 TTS 拿 audio_url） ── */
+/* ── 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: {
-  /** 人物视频 URL（MP4）；由素材 id 经 getAssetById 拿 file_url，禁止传 video_asset_id */
+  /** 人物视频 URL（MP4）；由素材 id 经 getAssetById 拿 file_url */
   video_url: string
-  /** 音色 ID（预置音色 或 克隆音色 profile UUID，后端会解析） */
-  voice_id: string
-  /** 要合成的文案（手动输入或文案库内容） */
-  script_text: string
-  /** 语速 0.5~2.0，默认 1.0 */
+  /** 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 */
+  /** 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
@@ -53,21 +64,27 @@ export const createLipsyncJob = async (data: {
   return response.data
 }
 
-export const getLipsyncJob = async (id: string): Promise<LipsyncJob> => {
-  const response = await apiClient.get<LipsyncJob>(`/lipsync/jobs/${id}`)
+/* ── #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
 }
 
-/* ── 智能封面（MediaKit 抽帧 + 质量评分选最佳帧，独立于渲染任务） ── */
-export const generateSmartCover = async (
-  video_url: string,
-  max_frames = 5,
-): Promise<{ cover_url: string; status: string; message: string }> => {
-  const response = await apiClient.post<{ cover_url: string; status: string; message: string }>(
-    "/ai-avatar/render/smart-cover",
-    { video_url, max_frames },
-    { timeout: 60000 },
-  )
+export const getLipsyncJob = async (id: string): Promise<LipsyncJob> => {
+  const response = await apiClient.get<LipsyncJob>(`/lipsync/jobs/${id}`, { timeout: 60000 })
   return response.data
 }
 
@@ -80,15 +97,35 @@ export const submitRender = async (data: {
   cover_config?: Record<string, unknown>
   project_id?: 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
+}
+
+/* ── 封面选定后点「完成」正式入库（POST /ai-avatar/render/{job_id}/finalize） ── */
+export const finalizeRenderJob = (renderId: string) =>
+  apiClient.post<{ video_id: string; cover_url: string; status: string }>(
+    `/ai-avatar/render/${renderId}/finalize`,
+  )

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/ModalCoverSelect.tsx

@@ -0,0 +1,59 @@
+/**
+ * AI数字人 — 封面选择弹窗
+ * 渲染完成后由主页面唤起，内部用 PanelCoverAndGenerate（select-cover 变体）提供
+ * 智能抽帧 + 自定义上传 + 预览 + 确定按钮。
+ */
+import React from "react"
+import type { AiAvatarCoverConfig, RenderJob } from "../types"
+import PanelCoverAndGenerate from "./PanelCoverAndGenerate"
+
+interface ModalCoverSelectProps {
+  open: boolean
+  onClose: () => void
+  renderJob: RenderJob | null
+  coverConfig: AiAvatarCoverConfig
+  onCoverConfigChange: (partial: Partial<AiAvatarCoverConfig>) => void
+  onGenerateRenderSmartCover: (renderId: string) => Promise<{ cover_url: string; message?: string }>
+  onUploadCover?: (file: File) => void
+  onCoverSelected: (coverUrl: string) => void
+}
+
+const ModalCoverSelect: React.FC<ModalCoverSelectProps> = ({
+  open,
+  onClose,
+  renderJob,
+  coverConfig,
+  onCoverConfigChange,
+  onGenerateRenderSmartCover,
+  onUploadCover,
+  onCoverSelected,
+}) => {
+  if (!open) return null
+
+  return (
+    <div className="aa-modal-overlay" onClick={onClose}>
+      <div className="aa-modal" onClick={(e) => e.stopPropagation()} style={{ maxWidth: 480 }}>
+        <div className="aa-modal__header">
+          <span className="aa-modal__title">选择封面</span>
+          <button type="button" className="aa-modal__close" onClick={onClose} aria-label="关闭">
+            ×
+          </button>
+        </div>
+        <div className="aa-modal__body" style={{ padding: 20 }}>
+          <PanelCoverAndGenerate
+            variant="select-cover"
+            coverConfig={coverConfig}
+            onCoverConfigChange={onCoverConfigChange}
+            renderJob={renderJob}
+            onGenerateRenderSmartCover={onGenerateRenderSmartCover}
+            onUploadCover={onUploadCover}
+            onClose={onClose}
+            onCoverSelected={onCoverSelected}
+          />
+        </div>
+      </div>
+    </div>
+  )
+}
+
+export default ModalCoverSelect

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

@@ -1,35 +1,45 @@
 /**
- * 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 / 封面选择弹窗内容：
+ * - variant="setup"（默认）：分辨率 / 配置摘要 / 「开始生成视频」按钮，用于主页面步骤2配置阶段；
+ *   渲染完成后仍内嵌封面预览与按钮，方便不打开弹窗直接操作。
+ * - variant="select-cover"：只渲染封面选择区（智能获取封面 + 自定义上传 + 预览），
+ *   用于 ModalCoverSelect 弹窗中；传 onClose 时底部显示「确定」按钮。
  *
- * 注意：v3 已删除"画面插入模式"，本面板不包含该选项。
+ * 封面一律从最终成片（已叠加标题/B-roll）抽帧，本面板不再叠加标题。
  */
-import React, { useRef } from "react"
-import type { AiAvatarCoverConfig } from "../types"
+import React, { useRef, useState } from "react"
+import type { AiAvatarCoverConfig, RenderJob } from "../types"
+
+type PanelVariant = "setup" | "select-cover"
 
 interface PanelCoverAndGenerateProps {
+  variant?: PanelVariant
   coverConfig: AiAvatarCoverConfig
   onCoverConfigChange: (partial: Partial<AiAvatarCoverConfig>) => void
-  resolution: string
-  onResolutionChange: (r: string) => void
-  isGenerating: boolean
-  onGenerate: () => void
-  /** 智能获取封面（MediaKit 选帧） */
-  onSmartCover: () => void
-  smartCoverLoading: boolean
-  canSmartCover: boolean
-  /** 配置汇总信息 */
-  summary: {
+  resolution?: string
+  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
+  /** 弹窗关闭回调（传入则表示在弹窗中使用，底部显示「确定」按钮） */
+  onClose?: () => void
+  /** 封面选好（智能抽帧/自定义上传成功）后通知父组件，参数为封面 URL */
+  onCoverSelected?: (coverUrl: string) => void
+  /** 配置汇总信息（仅 variant="setup" 使用） */
+  summary?: {
     videoName: string | null
     voiceName: string | null
     scriptLength: number
     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"
   }
 }
 
@@ -48,53 +58,116 @@ const LIPSYNC_STATUS_LABEL: Record<string, { text: string; cls: string }> = {
 }
 
 const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
+  variant = "setup",
   coverConfig,
   onCoverConfigChange,
-  resolution,
+  resolution = "720p",
   onResolutionChange,
-  isGenerating,
+  isGenerating = false,
   onGenerate,
-  onSmartCover,
-  smartCoverLoading,
-  canSmartCover,
+  renderJob,
+  onGenerateRenderSmartCover,
+  onUploadCover,
+  onClose,
+  onCoverSelected,
   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 = () => {
     uploadInputRef.current?.click()
   }
 
+  const _applyCoverUrl = (url: string, mode: "upload" | "auto_frame") => {
+    const partial: Partial<AiAvatarCoverConfig> = {
+      mode,
+      thumbnail_url: url,
+    }
+    if (mode === "auto_frame") {
+      partial.smart_cover_url = url
+    } else {
+      partial.upload_url = url
+    }
+    onCoverConfigChange(partial)
+    onCoverSelected?.(url)
+  }
+
   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）
+    if (onUploadCover) {
+      onUploadCover(file)
+      e.target.value = ""
+      return
+    }
+    // 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 })
-    // 允许重复选择同一文件
+    _applyCoverUrl(url, "upload")
     e.target.value = ""
   }
 
-  /** 智能获取封面（调后端 MediaKit 抽帧评分选最佳帧，#1822） */
-  const handleSmartCover = () => {
-    onCoverConfigChange({ mode: "auto_frame" })
-    onSmartCover()
+  /** 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) {
+        _applyCoverUrl(res.cover_url, "auto_frame")
+      } 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 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
 
-  const canGenerate = summary.lipsyncStatus === "completed" && !isGenerating
+  /** 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)
 
-  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" />
+  /** 封面区占位文字 */
+  const coverPlaceholder = isRenderCompleted ? "暂无封面" : "视频生成后可选择封面"
+
+  /** 配置摘要中的封面状态标签 */
+  const coverSummaryNode = (() => {
+    if (!summary) return null
+    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>
+  })()
+
+  // ── 封面选择区（两种 variant 共用） ─────────────────────────────────
+  const coverSection = (
+    <div className="aa-cover-section" style={{ marginTop: variant === "select-cover" ? 0 : 16 }}>
+      <div className="aa-label" style={{ marginBottom: 8 }}>
+        {variant === "select-cover" ? "选择封面" : "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">暂无封面</span>
+          <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">
@@ -102,8 +175,8 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
           type="button"
           className={`aa-btn aa-btn--ghost${coverConfig.mode === "auto_frame" ? " active" : ""}`}
           onClick={handleSmartCover}
-          disabled={smartCoverLoading || !canSmartCover}
-          title={canSmartCover ? "基于对口型成片智能选帧" : "请先完成对口型生成"}
+          disabled={!canSmartCover}
+          title={isRenderCompleted ? "从成片智能选帧" : "请先生成视频"}
         >
           {smartCoverLoading ? "⏳ 智能选帧中…" : "🎬 智能获取封面"}
         </button>
@@ -111,6 +184,8 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
           type="button"
           className={`aa-btn aa-btn--ghost${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>
@@ -122,14 +197,36 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
           onChange={handleFileChange}
         />
       </div>
+    </div>
+  )
 
+  // ── select-cover 变体：只渲染封面区 + 弹窗确定按钮 ──
+  if (variant === "select-cover") {
+    return (
+      <div className="aa-cover-generate">
+        {coverSection}
+        {onClose && (
+          <div style={{ marginTop: 16, display: "flex", justifyContent: "flex-end" }}>
+            <button type="button" className="aa-btn aa-btn--primary" onClick={onClose}>
+              确定
+            </button>
+          </div>
+        )}
+      </div>
+    )
+  }
+
+  // ── setup 变体：分辨率 / 配置摘要 / 生成按钮（渲染完成后内嵌封面区） ──
+  return (
+    <div className="aa-cover-generate">
       {/* 分辨率选择 */}
       <div className="aa-form-field">
         <label className="aa-label">分辨率</label>
         <select
           className="aa-select"
           value={resolution}
-          onChange={(e) => onResolutionChange(e.target.value)}
+          onChange={(e) => onResolutionChange?.(e.target.value)}
+          disabled={isGenerating}
         >
           {RESOLUTION_OPTIONS.map((opt) => (
             <option key={opt.value} value={opt.value}>
@@ -144,7 +241,7 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
         <div className="aa-config-summary">
           <div className="aa-config-summary__row">
             <span>出镜视频</span>
-            {summary.videoName ? (
+            {summary?.videoName ? (
               <span className="aa-config-summary__value">{summary.videoName}</span>
             ) : (
               <span className="aa-config-summary__empty">未配置</span>
@@ -152,7 +249,7 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
           </div>
           <div className="aa-config-summary__row">
             <span>音色</span>
-            {summary.voiceName ? (
+            {summary?.voiceName ? (
               <span className="aa-config-summary__value">{summary.voiceName}</span>
             ) : (
               <span className="aa-config-summary__empty">未配置</span>
@@ -160,7 +257,7 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
           </div>
           <div className="aa-config-summary__row">
             <span>文案</span>
-            {summary.scriptLength > 0 ? (
+            {summary && summary.scriptLength > 0 ? (
               <span className="aa-config-summary__value">{summary.scriptLength} 字</span>
             ) : (
               <span className="aa-config-summary__empty">未配置</span>
@@ -177,12 +274,12 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
           <div className="aa-config-summary__row">
             <span>B-roll 画面插入</span>
             <span className="aa-config-summary__value">
-              {summary.brollCount > 0 ? `${summary.brollCount} 段` : "无"}
+              {summary && summary.brollCount > 0 ? `${summary.brollCount} 段` : "无"}
             </span>
           </div>
           <div className="aa-config-summary__row">
             <span>标题</span>
-            {summary.hasTitle ? (
+            {summary?.hasTitle ? (
               <span className="aa-config-summary__value">已设置</span>
             ) : (
               <span className="aa-config-summary__empty">未配置</span>
@@ -190,11 +287,7 @@ 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>
 
@@ -207,11 +300,16 @@ const PanelCoverAndGenerate: React.FC<PanelCoverAndGenerateProps> = ({
         >
           {isGenerating ? "⏳ 生成中..." : "🚀 开始生成视频"}
         </button>
-        {summary.lipsyncStatus !== "completed" && !isGenerating && (
+        {summary?.lipsyncStatus !== "completed" && !isGenerating && (
           <div style={{ marginTop: 8, fontSize: 11, color: "#8c8ca1", textAlign: "center" }}>
             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>
     </div>
   )

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

@@ -1,8 +1,9 @@
 /**
  * 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: 预览容器按 1/2 缩放、标题实时叠加预览
+ * 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 {
@@ -13,6 +14,8 @@ interface PanelLipsyncPreviewProps {
   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> = {
@@ -26,6 +29,16 @@ function formatTime(seconds: number): string {
   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,
@@ -33,9 +46,14 @@ export function PanelLipsyncPreview({
   onOpenBRollModal,
   onRemoveBRoll,
   titleConfig,
+  onTitlePositionChange,
 }: PanelLipsyncPreviewProps) {
-  const isGenerating =
-    lipsyncJob?.status === "pending" || lipsyncJob?.status === "processing"
+  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"
 
@@ -46,30 +64,121 @@ export function PanelLipsyncPreview({
         ? "排队中…"
         : "对口型生成中…"
 
-  /** 标题叠加样式 */
-  const titleOverlayStyle: React.CSSProperties | null = titleConfig?.title
-    ? {
-        position: "absolute",
-        left: "50%",
-        transform: "translateX(-50%)",
-        color: titleConfig.color || "#ffffff",
-        fontFamily: titleConfig.font || "思源黑体",
-        fontSize: `${(titleConfig.size || 36) * 0.55}px`, // 预览等比缩
-        fontWeight: titleConfig.bold ? 700 : 400,
-        fontStyle: titleConfig.italic ? "italic" : "normal",
-        whiteSpace: "pre-wrap",
-        textAlign: "center",
-        width: "90%",
-        padding: "4px 8px",
-        textShadow: titleConfig.shadow ? "0 2px 4px rgba(0,0,0,0.8)" : undefined,
-        WebkitTextStroke: titleConfig.stroke ? "1.5px #000" : undefined,
-        ...(titleConfig.position === "top"
-          ? { top: 8 }
-          : titleConfig.position === "bottom"
-            ? { bottom: 8 }
-            : { top: "50%", transform: "translateX(-50%) translateY(-50%)" }),
-      }
-    : null
+  // 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">
@@ -78,9 +187,7 @@ export function PanelLipsyncPreview({
         <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>
+            <span className="aa-broll-badge">🎬 {bRollSegments.length} 个画面</span>
           )}
         </div>
         <div className="aa-lipsync-actions">
@@ -104,10 +211,7 @@ export function PanelLipsyncPreview({
                     alt={seg.asset.name}
                   />
                 ) : (
-                  <span
-                    className="aa-broll-item__thumb"
-                    style={{ padding: "6px 4px" }}
-                  >
+                  <span className="aa-broll-item__thumb" style={{ padding: "6px 4px" }}>
                     🎬
                   </span>
                 )}
@@ -140,15 +244,35 @@ export function PanelLipsyncPreview({
         )}
       </div>
 
-      {/* ── 对口型预览（v3.1: 缩放1/2 + 标题叠加） ─ */}
+      {/* ── 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">
+        <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 style={titleOverlayStyle}>{titleConfig!.title}</div>}
+              {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" }}>
@@ -189,11 +313,7 @@ export function PanelLipsyncPreview({
 
         <div className="aa-lipsync-actions">
           {isDone ? (
-            <button
-              type="button"
-              className="aa-btn aa-btn--full"
-              onClick={onGenerateLipsync}
-            >
+            <button type="button" className="aa-btn aa-btn--full" onClick={onGenerateLipsync}>
               🔄 重新生成对口型
             </button>
           ) : isGenerating ? (

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

@@ -55,9 +55,7 @@ export function PanelScript({
         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…"
+          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)}
       />

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

@@ -2,17 +2,16 @@
  * 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"
-import type { AiAvatarTitleConfig } from "../types"
-import { getFontFamily } from "@/pages/generate/constants"
 
 export interface PanelVideoSelectorProps {
   selectedVideo: AssetItem | null
   /** 触发打开素材库弹窗 */
   onSelectVideo: () => void
   onRemoveVideo: () => void
-  titleConfig?: AiAvatarTitleConfig
 }
 
 /** 格式化时长（秒 → mm:ss） */
@@ -27,7 +26,6 @@ export function PanelVideoSelector({
   selectedVideo,
   onSelectVideo,
   onRemoveVideo,
-  titleConfig,
 }: PanelVideoSelectorProps) {
   /* 未选视频：虚线上传区，点击打开素材库弹窗 */
   if (!selectedVideo) {
@@ -57,42 +55,13 @@ export function PanelVideoSelector({
 
   return (
     <div>
-      {/* 竖屏 9:16 视频预览播放器 + 标题实时预览 */}
-      <div className="aa-video-preview" style={{ position: "relative" }}>
+      {/* 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 />
         ) : (
           <div className="aa-video-preview__placeholder">视频暂不可预览</div>
         )}
-        {titleConfig?.title && (
-          <div
-            style={{
-              position: "absolute",
-              left: "50%",
-              transform: "translateX(-50%)",
-              ...(titleConfig.position === "top"
-                ? { top: "10%" }
-                : titleConfig.position === "bottom"
-                  ? { bottom: "10%" }
-                  : { top: "50%", transform: "translate(-50%, -50%)" }),
-              fontSize: Math.max(titleConfig.size, 32),
-              fontFamily: getFontFamily(titleConfig.font),
-              color: titleConfig.color,
-              fontWeight: titleConfig.bold ? 700 : 400,
-              fontStyle: titleConfig.italic ? "italic" : "normal",
-              textShadow: "0 2px 4px rgba(0,0,0,0.5)",
-              WebkitTextStroke: "2px #000",
-              pointerEvents: "none",
-              zIndex: 10,
-              maxWidth: "90%",
-              textAlign: "center",
-              whiteSpace: "pre-wrap",
-              lineHeight: 1.3,
-            }}
-          >
-            {titleConfig.title}
-          </div>
-        )}
       </div>
 
       {/* 视频信息卡片：文件名 / 时长 / 分辨率 */}

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

@@ -139,13 +139,6 @@ export function PanelVoiceSelector({
       }
       const targetId = voice.voice_clone_profile_id || voice.id
       // DEBUG: 打印请求参数，帮助定位 /tts/preview 失败原因
-      console.log("[AI数字人-克隆试听] previewTts 请求:", {
-        voice_id: targetId,
-        voice_name: voice.name,
-        voice_type: voice.type,
-        voice_clone_profile_id: voice.voice_clone_profile_id,
-        voice_id_field: voice.voice_id,
-      })
       setPreviewingId(voice.id)
       try {
         const res = await previewTts({
@@ -154,10 +147,6 @@ export function PanelVoiceSelector({
           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),
-          duration: res.duration,
-        })
         if (!res.audio_url) {
           setPreviewingId(null)
           message.error("合成试听失败：未返回音频")

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"
@@ -77,7 +99,7 @@ export interface AiAvatarTitleConfig {
   shadow: boolean
   color: string
   auto_subtitle: boolean
-  /** 自定义位置坐标（position=custom 时生效，像素） */
+  /** 自定义位置坐标（position=custom 时生效，百分比 0-100） */
   pos_x?: number
   pos_y?: number
 }
@@ -101,6 +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
 }
@@ -110,7 +133,7 @@ export const DEFAULT_TITLE_CONFIG: AiAvatarTitleConfig = {
   title: "",
   position: "bottom",
   font: "思源黑体",
-  size: 28,
+  size: 48,
   bold: true,
   italic: false,
   stroke: false,

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

@@ -28,10 +28,13 @@ export function normalizeEmotion(raw: string | undefined | null): VoiceEmotion {
  * 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)。
+ * 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): Record<string, unknown> {
+export function buildTitleConfigPayload(
+  cfg: AiAvatarTitleConfig,
+  titleImageDataUrl?: string | null,
+): Record<string, unknown> {
   const text = (cfg.title || "").trim()
   if (!text) return {}
   const position = cfg.position || "bottom"
@@ -39,7 +42,7 @@ export function buildTitleConfigPayload(cfg: AiAvatarTitleConfig): Record<string
     text,
     enabled: true,
     font: cfg.font || "思源黑体",
-    font_size: Math.round(cfg.size) || 36,
+    font_size: Math.round(cfg.size) || 48,
     font_color: cfg.color || "#ffffff",
     position,
     bold: !!cfg.bold,
@@ -53,6 +56,10 @@ export function buildTitleConfigPayload(cfg: AiAvatarTitleConfig): Record<string
     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
 }
 
@@ -67,9 +74,14 @@ export function buildCoverConfigPayload(
     // build_cover_extract_command 读取 timestamp（截帧秒数）
     timestamp: cfg.frame_time || 0,
   }
-  if (smartCoverUrl) payload.cover_url = smartCoverUrl
+  // 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/换行（按句断，不在逗号处切；保持与后端一致） */
+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/editing-planner/components/PreviewPlayer.tsx

@@ -5,6 +5,7 @@
 import React from "react"
 import type { ClipData, ClipType } from "../types"
 import type { TitleConfig } from "@/api/template-editor"
+import { getFontFamily } from "@/pages/generate/constants"
 
 interface SubtitleSettings {
   enabled: boolean
@@ -80,7 +81,7 @@ const PreviewPlayer: React.FC<PreviewPlayerProps> = ({
               className="ep-preview-title"
               style={{
                 fontSize: `${Math.min(titleConfig.font_size, 20)}px`,
-                fontFamily: titleConfig.font_preset,
+                fontFamily: getFontFamily(titleConfig.font_preset),
                 fontWeight: "bold",
                 fontStyle: "normal",
                 textShadow: "2px 2px 4px rgba(0,0,0,0.5)",
@@ -106,7 +107,7 @@ const PreviewPlayer: React.FC<PreviewPlayerProps> = ({
               className="ep-preview-subtitle"
               style={{
                 fontSize: `${Math.min(subtitleSettings.size, 14)}px`,
-                fontFamily: subtitleSettings.font,
+                fontFamily: getFontFamily(subtitleSettings.font),
                 top:
                   subtitleSettings.position === "top"
                     ? "8px"

apps/web/src/pages/editing-planner/constants/clipProperties.ts

@@ -9,15 +9,8 @@ export const POSITION_OPTIONS = [
   { value: "bottom", label: "底部" },
 ]
 
-export const FONT_OPTIONS = [
-  "思源黑体",
-  "思源宋体",
-  "苹方",
-  "PingFang",
-  "微软雅黑",
-  "楷体",
-  "华康俪金黑",
-]
+// FONT_OPTIONS 统一从 generate/constants 导入，避免多处维护遗漏
+export { FONT_OPTIONS } from "@/pages/generate/constants"
 
 export const ANIMATION_OPTIONS = [
   { value: "none", label: "无" },

apps/web/src/pages/editing-planner/constants/subtitleStyle.ts

@@ -7,15 +7,8 @@ export const POSITION_OPTIONS = [
   { value: "bottom", label: "底部" },
 ]
 
-export const FONT_OPTIONS = [
-  "思源黑体",
-  "思源宋体",
-  "苹方",
-  "PingFang",
-  "微软雅黑",
-  "楷体",
-  "华康俪金黑",
-]
+// FONT_OPTIONS 统一从 generate/constants 导入
+export { FONT_OPTIONS } from "@/pages/generate/constants"
 
 export const ANIMATION_OPTIONS = [
   { value: "none", label: "无" },

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

@@ -231,9 +231,13 @@ const GeneratePage: React.FC = () => {
   /* ── 批量变体真实片段（#1744）：后端独立选片，预览即成片；失败静默降级本地模拟 ──
      仅批量（N>1）且在第 4 步预览时申请，避免选素材阶段频繁请求；
      变体 0 沿用草稿 plan（与单视频一致），变体 1..N-1 后端 reselect 独立选片 */
+  // P0 fix：批量变体计划请求需携带配音参数，避免后端按"无配音"选片导致 clips 时长与配音错位
+  const batchVoiceLibraryId =
+    voiceMode === "clone" ? selectedClonedVoice || selectedVoice || "" : selectedVoice || ""
   const {
     clipsByVariant: variantClips,
     planIdsByVariant: variantPlanIds,
+    voiceDurationsByVariant: variantVoiceDurations,
     loading: variantClipsLoading,
     error: variantClipsError,
     retry: retryVariantClips,
@@ -243,6 +247,9 @@ const GeneratePage: React.FC = () => {
     templateId: selectedTemplate || "",
     assetIds: previewAssetIds,
     sourcePlanId: storedSourceEditPlanId || sourceEditPlanId || "",
+    voiceLibraryId: batchVoiceLibraryId,
+    voiceLibraryIds: voiceLibraryIds || [],
+    voiceModePerVideo,
   })
 
   /* ── 批量变体配音预览 URL（#1750）：独立模式每变体挂各自配音，共用模式全挂同一条；
@@ -476,6 +483,7 @@ const GeneratePage: React.FC = () => {
                   titles={previewTitles}
                   titleSettings={titleSettings}
                   voiceAudioUrls={variantVoiceAudioUrls}
+                  voiceDurations={variantVoiceDurations}
                   variantClips={variantClips}
                   clipsLoading={variantClipsLoading}
                   clipsError={variantClipsError}

apps/web/src/pages/generate/components/CanvasPreviewGrid.tsx

@@ -31,6 +31,12 @@ interface CanvasPreviewGridProps {
    * 元素为 null 表示该变体暂无音频（AI 音色 TTS 合成中））
    */
   voiceAudioUrls?: (string | null)[]
+  /**
+   * 各变体配音时长（秒）：后端返回 voice_duration 优先；未返回则为 undefined，
+   * 由 FrontendPreviewPlayer 在 audio loadedmetadata 时自测兜底。
+   * 长度=count，undefined 项表示该变体未提供后端时长。
+   */
+  voiceDurations?: (number | undefined)[]
   /**
    * 各变体的后端真实片段（#1744/#1750）：长度=count。
    * 仅 clipsLoading=false 且 clipsError=false 时才会传给播放器。
@@ -56,6 +62,7 @@ const CanvasPreviewGrid: React.FC<CanvasPreviewGridProps> = ({
   titles,
   titleSettings,
   voiceAudioUrls,
+  voiceDurations,
   variantClips,
   clipsLoading = false,
   clipsError = false,
@@ -120,6 +127,7 @@ const CanvasPreviewGrid: React.FC<CanvasPreviewGridProps> = ({
                 serverClips={variantClips[i]}
                 variantTitle={titles[i] || ""}
                 voiceAudioUrl={voiceAudioUrls?.[i] || undefined}
+                voiceDurationHint={voiceDurations?.[i]}
                 activePlayToken={activePlayToken}
                 onPlayTokenChange={setActivePlayToken}
                 compact

apps/web/src/pages/generate/components/FrontendPreviewPlayer.tsx

@@ -1,25 +1,23 @@
 /**
- * 前端预览播放器 — Canvas + WebCodecs 方案
+ * 前端预览播放器 — 原生 Video 元素方案（浏览器硬件解码，独立线程，不阻塞 UI）
  *
  * 架构：
- * - 浏览器支持 WebCodecs → Canvas 渲染（帧级精确控制 + 标题合成）
- * - 浏览器不支持 → fallback 到多 video 元素方案
+ * - 默认走原生 video 元素多片段切换播放（useSegmentScheduler 调度），
+ *   叠加标题 CSS 浮层、配音音轨（usePreviewAudio）、尾段冻结看门狗、批量播放互斥 token。
+ *   UI 拆分为 PreviewControls（控制条/按钮） + PreviewProgressBar（进度条）两个子组件。
+ * - WebCodecs 路径已废弃（原 useWebCodecs 常量恒为 false，相关死代码已移除），
+ *   保留 useCanvasPlayer hook 文件供未来兜底（不影响当前打包体积）。
  *
- * 对外 API 不变：assets, template, videoRatio, ready, voiceAudioUrl
+ * 对外 API 完全不变：assets / videoRatio / ready / voiceAudioUrl / serverClips 等。
  */
 import React, { useMemo, useCallback, useState, useRef, useEffect } from "react"
-import {
-  PlayCircleOutlined,
-  PauseCircleOutlined,
-  SoundOutlined,
-  LoadingOutlined,
-  AudioOutlined,
-  AudioMutedOutlined,
-} from "@ant-design/icons"
+import { PlayCircleOutlined, SoundOutlined } from "@ant-design/icons"
 import type { AssetItem } from "@/api/assets"
 import type { EditPlanClip } from "@/api/template-editor"
 import { useSegmentScheduler, type PlaybackSegment } from "../hooks/useSegmentScheduler"
-import { useCanvasPlayer } from "../hooks/useCanvasPlayer"
+import { usePreviewAudio } from "../hooks/usePreviewAudio"
+import { PreviewControls } from "./PreviewControls"
+import { getFontFamily } from "../constants"
 
 interface FrontendPreviewPlayerProps {
   assets: AssetItem[]
@@ -58,12 +56,11 @@ interface FrontendPreviewPlayerProps {
   activePlayToken?: number | null
   /** 播放权变化回调：本实例请求播放时传自身 playToken，暂停时传 null */
   onPlayTokenChange?: (token: number | null) => void
-}
-
-function formatTime(seconds: number): string {
-  const m = Math.floor(seconds / 60)
-  const s = Math.floor(seconds % 60)
-  return `${m}:${s.toString().padStart(2, "0")}`
+  /**
+   * 后端返回的配音时长（秒）P0 对齐：优先以该值作为音画时长锚点；
+   * 未提供则在 audio loadedmetadata 后自测兜底。
+   */
+  voiceDurationHint?: number
 }
 
 /**
@@ -116,6 +113,7 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
   ready,
   serverClips,
   voiceAudioUrl,
+  voiceDurationHint,
   titleSettings,
   onTitlePositionChange,
   playToken,
@@ -124,26 +122,11 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
   activePlayToken = null,
   onPlayTokenChange,
 }) => {
-  // #1754：测量配音时长，计算缩放因子
-  const [voiceDuration, setVoiceDuration] = useState(0)
-  useEffect(() => {
-    if (!voiceAudioUrl) {
-      setVoiceDuration(0)
-      return
-    }
-    const audio = new Audio()
-    audio.preload = "metadata"
-    const onLoaded = () => {
-      if (audio.duration && isFinite(audio.duration)) {
-        setVoiceDuration(audio.duration)
-      }
-    }
-    audio.addEventListener("loadedmetadata", onLoaded)
-    audio.src = voiceAudioUrl
-    return () => {
-      audio.removeEventListener("loadedmetadata", onLoaded)
-    }
-  }, [voiceAudioUrl])
+  // #1754→P0：配音时长作为音画时长锚点。
+  // 优先使用后端返回的 voiceDurationHint；音频 loadedmetadata 后再以自测值覆盖（更精确）。
+  const [voiceDuration, setVoiceDuration] = useState<number>(() =>
+    voiceDurationHint && voiceDurationHint > 0 ? voiceDurationHint : 0,
+  )
 
   // #1756：clips 原始总时长 + 转场时长（后端等比分配配音时包含转场占位）
   const rawClipsDuration = useMemo(() => {
@@ -189,9 +172,10 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
       ? (titleSettings.posY / playRes.height) * 100
       : null
 
-  // ── 拖拽状态（用 ref 避免在每帧渲染中触发重渲染）──
+  // ── 标题拖拽（用 ref 避免每帧触发 React 重渲染）──
   const draggingTitleRef = useRef(false)
   const titleDragRef = useRef<HTMLDivElement>(null)
+  const playerContainerRef = useRef<HTMLDivElement>(null)
   const handleTitlePointerDown = useCallback(
     (e: React.PointerEvent<HTMLDivElement>) => {
       if (!onTitlePositionChange || !playerContainerRef.current) return
@@ -207,7 +191,6 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
     if (!draggingTitleRef.current || !playerContainerRef.current) return
     e.preventDefault()
     e.stopPropagation()
-    // 拖拽过程中直接修改 DOM，不触发 React 渲染（避免频繁重渲染导致换行）
     if (titleDragRef.current) {
       const rect = playerContainerRef.current.getBoundingClientRect()
       const relX = Math.max(0, Math.min(rect.width, e.clientX - rect.left))
@@ -222,7 +205,6 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
     (e: React.PointerEvent<HTMLDivElement>) => {
       if (!draggingTitleRef.current) return
       draggingTitleRef.current = false
-      // 拖拽结束时才调用 onTitlePositionChange 保存最终位置
       if (onTitlePositionChange && playerContainerRef.current) {
         const rect = playerContainerRef.current.getBoundingClientRect()
         const relX = Math.max(0, Math.min(rect.width, e.clientX - rect.left))
@@ -243,7 +225,6 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
     [onTitlePositionChange, playRes.width, playRes.height],
   )
 
-  const playerContainerRef = useRef<HTMLDivElement>(null)
   const [containerHeight, setContainerHeight] = useState(0)
   useEffect(() => {
     const el = playerContainerRef.current
@@ -268,234 +249,157 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
   const titleSidePct = (TITLE_MARGIN_SIDE / playRes.width) * 100
   const titleTopPct = (TITLE_MARGIN_TOP / playRes.height) * 100
   const titleBottomPct = (TITLE_MARGIN_BOTTOM / playRes.height) * 100
-  // 描边/阴影也要按缩放比例放大
   const titleScale = containerHeight > 0 ? containerHeight / playRes.height : 1
   const titleStrokeWidth = Math.max(1, 2 * titleScale)
   const titleShadowBlur = 4 * titleScale
   const titleShadowOffset = 2 * titleScale
 
-  // 默认走原生 video 播放（浏览器硬件解码，独立线程，不阻塞 UI）
-  // WebCodecs 仅在明确需要时启用（保留代码作为兜底）
-  const useWebCodecs = false
-
-  // ── 两条路径共用同一个 canvas ref（fallback 路径不使用） ──
-  const canvasRef = useRef<HTMLCanvasElement>(null)
-
-  // ── Canvas 播放器（WebCodecs 路径） ──
-  const canvasTitle = titleSettings
-    ? {
-        text: effectiveTitle || "标题预览",
-        fontSize: titleSettings.size,
-        fontFamily: titleSettings.font || "思源黑体",
-        color: titleSettings.color || "#ffffff",
-        position: titleSettings.position || "top",
-        bold: titleSettings.bold,
-        stroke: titleSettings.stroke,
-        shadow: titleSettings.shadow,
-      }
-    : undefined
-
-  const canvasSegments = useMemo(
-    () =>
-      segments.map((s) => ({
-        assetId: s.assetId,
-        videoUrl: s.videoUrl,
-        startTime: s.startTime,
-        endTime: s.endTime,
-      })),
-    [segments],
-  )
-
-  // WebCodecs 解码失败后强制走 video fallback
-  const [forceVideoFallback, setForceVideoFallback] = useState(false)
-
-  const handleCanvasError = useCallback((err: Error) => {
-    console.error("[FrontendPreviewPlayer] Canvas decode Error, switching to video fallback:", err)
-    setForceVideoFallback(true)
-  }, [])
-
-  const { state: canvasState, controls: canvasControls } = useCanvasPlayer(
-    canvasRef,
-    useWebCodecs && !forceVideoFallback ? canvasSegments : [],
-    useWebCodecs && !forceVideoFallback ? canvasTitle : undefined,
-    handleCanvasError,
-    useWebCodecs && !forceVideoFallback,
-  )
-
-  // WebCodecs 报告解码失败时自动切换到 video fallback
-  useEffect(() => {
-    if (canvasState.hasDecodeError && !forceVideoFallback) {
-      console.warn("[FrontendPreviewPlayer] hasDecodeError detected, forcing video fallback")
-      setForceVideoFallback(true)
-    }
-  }, [canvasState.hasDecodeError, forceVideoFallback])
-
-  // ── Video 播放器（fallback 路径） ──
+  // ── Video 播放器（默认路径，浏览器原生硬件解码） ──
   const {
-    isPlaying: videoIsPlaying,
-    currentTime: videoCurrentTime,
-    totalDuration: videoTotalDuration,
-    currentSegmentIndex: videoCurrentSegIdx,
-    canPlay: videoCanPlay,
-    togglePlayPause: videoTogglePlayPause,
-    seekTo: videoSeekTo,
-    pause: videoPause,
+    isPlaying,
+    currentTime,
+    totalDuration,
+    currentSegmentIndex,
+    canPlay,
+    togglePlayPause,
+    seekTo,
+    pause,
     videoRefs,
   } = useSegmentScheduler(segments)
 
-  // 选择哪条路径的状态（WebCodecs 解码失败时强制走 video fallback）
-  const effectiveUseWebCodecs = useWebCodecs && !forceVideoFallback
-  const isPlaying = effectiveUseWebCodecs ? canvasState.isPlaying : videoIsPlaying
-  const currentTime = effectiveUseWebCodecs ? canvasState.currentTime : videoCurrentTime
-  const totalDuration = effectiveUseWebCodecs ? canvasState.duration : videoTotalDuration
-  const canPlay = effectiveUseWebCodecs ? canvasState.isReady : videoCanPlay
-  const isBuffering = effectiveUseWebCodecs ? canvasState.isBuffering : false
+  // P0 fix：以配音时长为音画同步锚点。
+  // 有配音时总时长 = 配音时长（短则末帧冻结，长则硬停）；无配音时沿用视频总时长（素材原声兜底）。
+  const effectiveTotalDuration =
+    !!voiceAudioUrl && voiceDuration > 0 ? voiceDuration : totalDuration
 
-  // ── 配音音频同步 ──
-  const audioRef = useRef<HTMLAudioElement | null>(null)
-  const prevIsPlayingRef = useRef(false)
   // 本卡片静音开关（#1741）：默认有声，用户可点喇叭单独静音某张卡片
   const [muted, setMuted] = useState(false)
   // 有配音时 video 素材保持静音（避免原声与配音混音）；无配音时取消静音，素材原声兜底
   const hasVoice = !!voiceAudioUrl
 
-  useEffect(() => {
-    if (!voiceAudioUrl) {
-      if (audioRef.current) {
-        audioRef.current.pause()
-        audioRef.current.src = ""
-        audioRef.current = null
-      }
-      return
-    }
-    if (!audioRef.current) {
-      audioRef.current = new Audio()
-      audioRef.current.preload = "auto"
-    }
-    if (audioRef.current.src !== voiceAudioUrl) {
-      audioRef.current.src = voiceAudioUrl
-    }
-    audioRef.current.muted = muted
-  }, [voiceAudioUrl, muted])
+  // 音频 ended：兜底触发暂停与释放播放权
+  const handleAudioEnded = useCallback(() => {
+    if (!isPlaying) return
+    pause()
+    if (playToken != null) onPlayTokenChange?.(null)
+  }, [isPlaying, pause, playToken, onPlayTokenChange])
 
-  useEffect(() => {
-    const audio = audioRef.current
-    if (!audio || !audio.src) return
-    if (isPlaying && !prevIsPlayingRef.current) {
-      audio.currentTime = currentTime
-      audio.play().catch(() => {})
-    } else if (!isPlaying && prevIsPlayingRef.current) {
-      audio.pause()
-    }
-    prevIsPlayingRef.current = isPlaying
-  }, [isPlaying, currentTime])
+  const {
+    seekTo: audioSeekTo,
+    ensurePlayingAt: audioEnsurePlayingAt,
+    pause: audioPause,
+  } = usePreviewAudio({
+    voiceAudioUrl,
+    voiceDurationHint,
+    muted,
+    isPlaying,
+    currentTime,
+    onVoiceDurationChange: setVoiceDuration,
+    onEnded: handleAudioEnded,
+  })
 
-  // 片段切换时同步音频（仅 fallback 路径需要）
-  const segmentSyncKey = effectiveUseWebCodecs ? -1 : videoCurrentSegIdx
+  // 片段切换时同步音频时间（video fallback）
   useEffect(() => {
-    const audio = audioRef.current
-    if (!audio || !audio.src || !isPlaying) return
-    audio.currentTime = currentTime
+    if (!isPlaying) return
+    audioSeekTo(currentTime)
     // 注意：不要把 currentTime 放进依赖数组，否则每200ms会重置音频位置导致卡顿
     // eslint-disable-next-line react-hooks/exhaustive-deps
-  }, [segmentSyncKey, isPlaying])
+  }, [currentSegmentIndex, isPlaying])
+
+  // P0 fix：视频比配音短时的「末帧冻结+音频续播」模式。
+  // 视频调度器播完最后一段自动 pause，此时若配音仍在播，用 rAF 虚拟时钟推进 currentTime 直到配音结束。
+  const [tailCurrentTime, setTailCurrentTime] = useState<number | null>(null)
+  const tailStartRef = useRef<number>(0)
+  const tailBaseRef = useRef<number>(0)
+  const tailAudioRef = useRef({ ensurePlayingAt: audioEnsurePlayingAt, pause: audioPause })
+  tailAudioRef.current = { ensurePlayingAt: audioEnsurePlayingAt, pause: audioPause }
+
+  useEffect(() => {
+    const needTail =
+      !!voiceAudioUrl &&
+      voiceDuration > 0 &&
+      !isPlaying &&
+      typeof currentTime === "number" &&
+      currentTime >= totalDuration - 0.1 &&
+      currentTime < voiceDuration - 0.1
+    if (needTail && tailCurrentTime === null) {
+      tailBaseRef.current = currentTime
+      tailStartRef.current = performance.now()
+      setTailCurrentTime(currentTime)
+      tailAudioRef.current.ensurePlayingAt(currentTime)
+      return
+    }
+    if (!needTail && tailCurrentTime !== null) {
+      setTailCurrentTime(null)
+    }
+  }, [isPlaying, currentTime, totalDuration, voiceDuration, voiceAudioUrl, tailCurrentTime])
+
+  useEffect(() => {
+    if (tailCurrentTime === null) return
+    let raf = 0
+    const tick = () => {
+      const elapsed = (performance.now() - tailStartRef.current) / 1000
+      const t = Math.min(tailBaseRef.current + elapsed, voiceDuration || tailBaseRef.current)
+      setTailCurrentTime(t)
+      tailAudioRef.current.ensurePlayingAt(t)
+      if (t >= (voiceDuration || 0) - 0.05) {
+        tailAudioRef.current.pause()
+        if (playToken != null) onPlayTokenChange?.(null)
+        setTailCurrentTime(null)
+        return
+      }
+      raf = requestAnimationFrame(tick)
+    }
+    raf = requestAnimationFrame(tick)
+    return () => cancelAnimationFrame(raf)
+  }, [tailCurrentTime, voiceDuration, playToken, onPlayTokenChange])
+
+  // 呈现给 UI/进度条的「当前时间」：尾段用虚拟时间，否则用视频时间
+  const displayCurrentTime = tailCurrentTime !== null ? tailCurrentTime : currentTime
 
   const handleSeekTo = useCallback(
     (time: number) => {
-      if (effectiveUseWebCodecs) {
-        canvasControls.seek(time)
-      } else {
-        videoSeekTo(time)
-      }
-      const audio = audioRef.current
-      if (audio && audio.src) {
-        audio.currentTime = time
-      }
+      setTailCurrentTime(null)
+      seekTo(time)
+      audioSeekTo(time)
     },
-    [effectiveUseWebCodecs, canvasControls, videoSeekTo],
+    [seekTo, audioSeekTo],
   )
 
-  // ── 批量网格播放互斥（#1741）：播放权属于其他实例时，本实例自动暂停（视频+配音） ──
+  // ── 批量网格播放互斥（#1741）：播放权属于其他实例时，本实例自动暂停 ──
   useEffect(() => {
     if (activePlayToken == null || playToken == null || activePlayToken === playToken) return
-    if (effectiveUseWebCodecs) {
-      if (canvasState.isPlaying) canvasControls.pause()
-    } else if (isPlaying) {
-      videoPause()
+    if (isPlaying) {
+      pause()
     }
-    // isPlaying/canvasState.isPlaying 不放依赖：只在 token 变化时执行一次暂停，
-    // token 等于自身时本实例的播放在 handleTogglePlay 里处理
+    // isPlaying 不放依赖：只在 token 变化时执行一次暂停
     // eslint-disable-next-line react-hooks/exhaustive-deps
-  }, [activePlayToken, playToken, effectiveUseWebCodecs])
+  }, [activePlayToken, playToken])
 
   const handleTogglePlay = useCallback(() => {
-    if (effectiveUseWebCodecs) {
-      if (canvasState.isPlaying) {
-        canvasControls.pause()
-        onPlayTokenChange?.(null)
-      } else {
-        if (playToken != null) onPlayTokenChange?.(playToken)
-        canvasControls.play()
-      }
-    } else {
-      // video fallback：先上报播放权（暂停其他卡片），再切换本卡片播放/暂停
-      if (playToken != null) onPlayTokenChange?.(isPlaying ? null : playToken)
-      videoTogglePlayPause()
-    }
+    if (playToken != null) onPlayTokenChange?.(isPlaying ? null : playToken)
+    togglePlayPause()
+  }, [togglePlayPause, isPlaying, playToken, onPlayTokenChange])
+
+  // P0 fix：音画同步看门狗——有配音时播放时间达到配音时长立即暂停视频+音频（末帧冻结）
+  useEffect(() => {
+    if (!isPlaying) return
+    if (!voiceAudioUrl || voiceDuration <= 0) return
+    if (displayCurrentTime < voiceDuration - 0.08) return
+    pause()
+    audioPause()
+    if (playToken != null) onPlayTokenChange?.(null)
   }, [
-    effectiveUseWebCodecs,
-    canvasState.isPlaying,
-    canvasControls,
-    videoTogglePlayPause,
     isPlaying,
+    displayCurrentTime,
+    voiceAudioUrl,
+    voiceDuration,
+    pause,
+    audioPause,
     playToken,
     onPlayTokenChange,
   ])
 
-  // ── 进度条拖拽 ──
-  const [isDragging, setIsDragging] = useState(false)
-  const progressRef = useRef<HTMLDivElement>(null)
-
-  const handleProgressClick = useCallback(
-    (e: React.MouseEvent<HTMLDivElement>) => {
-      if (!progressRef.current || totalDuration <= 0) return
-      const rect = progressRef.current.getBoundingClientRect()
-      const ratio = Math.max(0, Math.min(1, (e.clientX - rect.left) / rect.width))
-      handleSeekTo(ratio * totalDuration)
-    },
-    [totalDuration, handleSeekTo],
-  )
-
-  const handleMouseDown = useCallback(
-    (e: React.MouseEvent<HTMLDivElement>) => {
-      setIsDragging(true)
-      handleProgressClick(e)
-    },
-    [handleProgressClick],
-  )
-
-  useEffect(() => {
-    if (!isDragging) return
-    const handleMouseMove = (e: MouseEvent) => {
-      if (!progressRef.current || totalDuration <= 0) return
-      const rect = progressRef.current.getBoundingClientRect()
-      const ratio = Math.max(0, Math.min(1, (e.clientX - rect.left) / rect.width))
-      handleSeekTo(ratio * totalDuration)
-    }
-    const handleMouseUp = () => setIsDragging(false)
-    window.addEventListener("mousemove", handleMouseMove)
-    window.addEventListener("mouseup", handleMouseUp)
-    return () => {
-      window.removeEventListener("mousemove", handleMouseMove)
-      window.removeEventListener("mouseup", handleMouseUp)
-    }
-  }, [isDragging, totalDuration, handleSeekTo])
-
-  const progressPercent = totalDuration > 0 ? (currentTime / totalDuration) * 100 : 0
-
-  // ── Canvas 容器 ref（保留声明，WebCodecs 兜底路径仍引用） ──
-  const canvasContainerRef = useRef<HTMLDivElement>(null)
-
   // ── 未就绪 ──
   if (!ready || !assets.length) {
     return (
@@ -529,7 +433,6 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
 
   // ── 无播放片段 ──
   if (!canPlay) {
-    const showDecodeError = forceVideoFallback && canvasState.hasDecodeError
     return (
       <div
         style={{
@@ -549,48 +452,15 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
           padding: 24,
         }}
       >
-        {isBuffering ? (
-          <>
-            <LoadingOutlined style={{ fontSize: 40, color: "#fff", marginBottom: 12 }} spin />
-            <p style={{ color: "rgba(255,255,255,0.8)", fontSize: 14, margin: 0 }}>加载中...</p>
-          </>
-        ) : showDecodeError ? (
-          <>
-            <PlayCircleOutlined style={{ fontSize: 40, color: "#ef4444", marginBottom: 12 }} />
-            <p
-              style={{
-                color: "rgba(255,255,255,0.9)",
-                fontSize: 14,
-                margin: "0 0 4px",
-                fontWeight: 500,
-              }}
-            >
-              视频解码失败
-            </p>
-            <p
-              style={{
-                color: "rgba(255,255,255,0.5)",
-                fontSize: 12,
-                margin: 0,
-                textAlign: "center",
-              }}
-            >
-              {canvasState.errorMessage || "当前浏览器不支持该视频编码格式，请刷新重试"}
-            </p>
-          </>
-        ) : (
-          <>
-            <PlayCircleOutlined
-              style={{ fontSize: 40, color: "rgba(255,255,255,0.3)", marginBottom: 12 }}
-            />
-            <p style={{ color: "rgba(255,255,255,0.6)", fontSize: 14, margin: "0 0 4px" }}>
-              暂无可播放素材
-            </p>
-            <p style={{ color: "rgba(255,255,255,0.35)", fontSize: 12, margin: 0 }}>
-              请先在左侧选择素材
-            </p>
-          </>
-        )}
+        <PlayCircleOutlined
+          style={{ fontSize: 40, color: "rgba(255,255,255,0.3)", marginBottom: 12 }}
+        />
+        <p style={{ color: "rgba(255,255,255,0.6)", fontSize: 14, margin: "0 0 4px" }}>
+          暂无可播放素材
+        </p>
+        <p style={{ color: "rgba(255,255,255,0.35)", fontSize: 12, margin: 0 }}>
+          请先在左侧选择素材
+        </p>
       </div>
     )
   }
@@ -612,53 +482,30 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
           : "0 4px 6px -1px rgba(0,0,0,0.3), 0 20px 50px -12px rgba(0,0,0,0.5), inset 0 0 0 1px rgba(255,255,255,0.06)",
       }}
     >
-      {/* ── Canvas 渲染层（WebCodecs 路径） ── */}
-      {effectiveUseWebCodecs && (
-        <div
-          ref={canvasContainerRef}
+      {/* ── Video 渲染层（默认路径，浏览器原生硬件解码） ── */}
+      {segments.map((seg, i) => (
+        <video
+          key={seg.assetId}
+          ref={(el) => {
+            videoRefs.current[i] = el
+          }}
+          preload="auto"
+          src={seg.videoUrl}
           style={{
             position: "absolute",
             inset: 0,
-            zIndex: 1,
+            width: "100%",
+            height: "100%",
+            objectFit: "cover",
             background: "#000",
+            zIndex: 1,
+            opacity: i === currentSegmentIndex ? 1 : 0,
+            pointerEvents: i === currentSegmentIndex ? "auto" : "none",
           }}
-        >
-          <canvas
-            ref={canvasRef}
-            style={{
-              width: "100%",
-              height: "100%",
-              objectFit: "cover",
-            }}
-          />
-        </div>
-      )}
-
-      {/* ── Video 渲染层（默认路径，浏览器原生硬件解码） ── */}
-      {!effectiveUseWebCodecs &&
-        segments.map((seg, i) => (
-          <video
-            key={seg.assetId}
-            muted={hasVoice || muted}
-            ref={(el) => {
-              videoRefs.current[i] = el
-            }}
-            preload="auto"
-            src={seg.videoUrl}
-            style={{
-              position: "absolute",
-              inset: 0,
-              width: "100%",
-              height: "100%",
-              objectFit: "cover",
-              background: "#000",
-              zIndex: 1,
-              opacity: i === videoCurrentSegIdx ? 1 : 0,
-              pointerEvents: i === videoCurrentSegIdx ? "auto" : "none",
-            }}
-            playsInline
-          />
-        ))}
+          muted={hasVoice || muted}
+          playsInline
+        />
+      ))}
 
       {/* 标题CSS叠加层 — 与后端 ASS 烧录坐标系 1:1 对齐 */}
       {titleSettings?.title && (
@@ -710,7 +557,7 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
             <span
               style={{
                 fontSize: `${titleFontSizePx}px`,
-                fontFamily: titleSettings.font || "思源黑体",
+                fontFamily: getFontFamily(titleSettings.font || "思源黑体"),
                 color: titleSettings.color || "#ffffff",
                 fontWeight: titleSettings.bold ? 700 : 400,
                 fontStyle: titleSettings.italic ? "italic" : "normal",
@@ -735,200 +582,19 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
         </div>
       )}
 
-      {/* 中央播放按钮 */}
-      {!isPlaying && (
-        <button
-          onClick={handleTogglePlay}
-          style={{
-            position: "absolute",
-            top: "50%",
-            left: "50%",
-            transform: "translate(-50%, -50%)",
-            background: "rgba(0,0,0,0.45)",
-            backdropFilter: "blur(12px)",
-            WebkitBackdropFilter: "blur(12px)",
-            border: "1px solid rgba(255,255,255,0.15)",
-            borderRadius: "50%",
-            width: 52,
-            height: 52,
-            cursor: "pointer",
-            display: "flex",
-            alignItems: "center",
-            justifyContent: "center",
-            color: "#fff",
-            fontSize: 26,
-            zIndex: 10,
-            transition: "transform 0.2s ease, background 0.2s ease",
-            boxShadow: "0 4px 20px rgba(0,0,0,0.4)",
-          }}
-          onMouseEnter={(e) => {
-            e.currentTarget.style.transform = "translate(-50%, -50%) scale(1.08)"
-            e.currentTarget.style.background = "rgba(0,0,0,0.6)"
-          }}
-          onMouseLeave={(e) => {
-            e.currentTarget.style.transform = "translate(-50%, -50%) scale(1)"
-            e.currentTarget.style.background = "rgba(0,0,0,0.45)"
-          }}
-        >
-          <PlayCircleOutlined />
-        </button>
-      )}
-
-      {/* 静音/有声切换（#1741）：左上角，默认有声；批量与单视频均可单独静音 */}
-      {segments.length > 0 && (
-        <button
-          type="button"
-          aria-label={muted ? "取消静音" : "静音"}
-          title={muted ? "取消静音" : "静音"}
-          onClick={() => setMuted((m) => !m)}
-          style={{
-            position: "absolute",
-            top: 8,
-            left: 8,
-            width: compact ? 26 : 30,
-            height: compact ? 26 : 30,
-            display: "flex",
-            alignItems: "center",
-            justifyContent: "center",
-            background: "rgba(0,0,0,0.45)",
-            backdropFilter: "blur(8px)",
-            WebkitBackdropFilter: "blur(8px)",
-            border: "1px solid rgba(255,255,255,0.1)",
-            borderRadius: "50%",
-            color: muted ? "rgba(255,255,255,0.45)" : "rgba(255,255,255,0.92)",
-            fontSize: compact ? 13 : 15,
-            cursor: "pointer",
-            zIndex: 10,
-            padding: 0,
-            transition: "background 0.15s, color 0.15s",
-          }}
-          onMouseEnter={(e) => {
-            e.currentTarget.style.background = "rgba(0,0,0,0.65)"
-          }}
-          onMouseLeave={(e) => {
-            e.currentTarget.style.background = "rgba(0,0,0,0.45)"
-          }}
-        >
-          {muted ? <AudioMutedOutlined /> : <AudioOutlined />}
-        </button>
-      )}
-
-      {/* 片段指示器 — 右上角胶囊 */}
-      <div
-        style={{
-          position: "absolute",
-          top: 8,
-          right: 8,
-          background: "rgba(0,0,0,0.45)",
-          backdropFilter: "blur(8px)",
-          WebkitBackdropFilter: "blur(8px)",
-          color: "rgba(255,255,255,0.9)",
-          fontSize: compact ? 9 : 10,
-          fontWeight: 500,
-          padding: compact ? "1px 6px" : "2px 8px",
-          borderRadius: 999,
-          zIndex: 10,
-          border: "1px solid rgba(255,255,255,0.1)",
-          letterSpacing: 0.3,
-        }}
-      >
-        {`${videoCurrentSegIdx + 1} / ${segments.length}`}
-      </div>
-
-      {/* 控制条 — 手机风格毛玻璃 */}
-      <div
-        style={{
-          position: "absolute",
-          bottom: 0,
-          left: 0,
-          right: 0,
-          display: "flex",
-          alignItems: "center",
-          gap: compact ? 6 : 10,
-          padding: compact ? "8px 10px 10px" : "12px 16px 16px",
-          background: "linear-gradient(transparent, rgba(0,0,0,0.7))",
-          backdropFilter: "blur(4px)",
-          WebkitBackdropFilter: "blur(4px)",
-          zIndex: 10,
-        }}
-      >
-        <button
-          onClick={handleTogglePlay}
-          style={{
-            background: "rgba(255,255,255,0.15)",
-            border: "none",
-            color: "#fff",
-            fontSize: compact ? 14 : 16,
-            cursor: "pointer",
-            width: compact ? 26 : 32,
-            height: compact ? 26 : 32,
-            borderRadius: "50%",
-            display: "flex",
-            alignItems: "center",
-            justifyContent: "center",
-            flexShrink: 0,
-            transition: "background 0.15s",
-          }}
-          onMouseEnter={(e) => {
-            e.currentTarget.style.background = "rgba(255,255,255,0.25)"
-          }}
-          onMouseLeave={(e) => {
-            e.currentTarget.style.background = "rgba(255,255,255,0.15)"
-          }}
-        >
-          {isPlaying ? <PauseCircleOutlined /> : <PlayCircleOutlined />}
-        </button>
-
-        <span
-          style={{
-            fontSize: compact ? 10 : 11,
-            color: "rgba(255,255,255,0.85)",
-            minWidth: compact ? 58 : 72,
-            fontVariantNumeric: "tabular-nums",
-            letterSpacing: 0.2,
-          }}
-        >
-          {formatTime(currentTime)} / {formatTime(totalDuration)}
-        </span>
-
-        <div
-          ref={progressRef}
-          onMouseDown={handleMouseDown}
-          style={{
-            flex: 1,
-            height: 3,
-            background: "rgba(255,255,255,0.2)",
-            borderRadius: 2,
-            cursor: "pointer",
-            position: "relative",
-          }}
-        >
-          <div
-            style={{
-              height: "100%",
-              width: `${progressPercent}%`,
-              background: "#fff",
-              borderRadius: 2,
-              transition: isDragging ? "none" : "width 0.1s linear",
-            }}
-          />
-          <div
-            style={{
-              position: "absolute",
-              top: "50%",
-              left: `${progressPercent}%`,
-              transform: "translate(-50%, -50%)",
-              width: 10,
-              height: 10,
-              borderRadius: "50%",
-              background: "#fff",
-              boxShadow: "0 0 6px rgba(255,255,255,0.5)",
-              opacity: isDragging ? 1 : 0,
-              transition: "opacity 0.15s",
-            }}
-          />
-        </div>
-      </div>
+      <PreviewControls
+        isPlaying={isPlaying}
+        onTogglePlay={handleTogglePlay}
+        muted={muted}
+        onToggleMute={() => setMuted((m) => !m)}
+        hasSegments={segments.length > 0}
+        segmentIndex={currentSegmentIndex}
+        segmentCount={segments.length}
+        currentTime={displayCurrentTime}
+        totalDuration={effectiveTotalDuration}
+        onSeek={handleSeekTo}
+        compact={compact}
+      />
     </div>
   )
 }

apps/web/src/pages/generate/components/PreviewControls.tsx

@@ -0,0 +1,195 @@
+import React from "react"
+import {
+  PlayCircleOutlined,
+  PauseCircleOutlined,
+  AudioOutlined,
+  AudioMutedOutlined,
+} from "@ant-design/icons"
+import { PreviewProgressBar } from "./PreviewProgressBar"
+
+interface PreviewControlsProps {
+  isPlaying: boolean
+  onTogglePlay: () => void
+  muted: boolean
+  onToggleMute: () => void
+  hasSegments: boolean
+  segmentIndex: number
+  segmentCount: number
+  currentTime: number
+  totalDuration: number
+  onSeek: (time: number) => void
+  compact?: boolean
+}
+
+/**
+ * 播放控制 UI 组件（静音按钮 / 片段指示器 / 中央播放按钮 / 底部毛玻璃控制条）
+ */
+export const PreviewControls: React.FC<PreviewControlsProps> = ({
+  isPlaying,
+  onTogglePlay,
+  muted,
+  onToggleMute,
+  hasSegments,
+  segmentIndex,
+  segmentCount,
+  currentTime,
+  totalDuration,
+  onSeek,
+  compact = false,
+}) => {
+  return (
+    <>
+      {/* 静音/有声切换（#1741）：左上角 */}
+      {hasSegments && (
+        <button
+          type="button"
+          aria-label={muted ? "取消静音" : "静音"}
+          title={muted ? "取消静音" : "静音"}
+          onClick={onToggleMute}
+          style={{
+            position: "absolute",
+            top: 8,
+            left: 8,
+            width: compact ? 26 : 30,
+            height: compact ? 26 : 30,
+            display: "flex",
+            alignItems: "center",
+            justifyContent: "center",
+            background: "rgba(0,0,0,0.45)",
+            backdropFilter: "blur(8px)",
+            WebkitBackdropFilter: "blur(8px)",
+            border: "1px solid rgba(255,255,255,0.1)",
+            borderRadius: "50%",
+            color: muted ? "rgba(255,255,255,0.45)" : "rgba(255,255,255,0.92)",
+            fontSize: compact ? 13 : 15,
+            cursor: "pointer",
+            zIndex: 10,
+            padding: 0,
+            transition: "background 0.15s, color 0.15s",
+          }}
+          onMouseEnter={(e) => {
+            e.currentTarget.style.background = "rgba(0,0,0,0.65)"
+          }}
+          onMouseLeave={(e) => {
+            e.currentTarget.style.background = "rgba(0,0,0,0.45)"
+          }}
+        >
+          {muted ? <AudioMutedOutlined /> : <AudioOutlined />}
+        </button>
+      )}
+
+      {/* 片段指示器 — 右上角胶囊 */}
+      <div
+        style={{
+          position: "absolute",
+          top: 8,
+          right: 8,
+          background: "rgba(0,0,0,0.45)",
+          backdropFilter: "blur(8px)",
+          WebkitBackdropFilter: "blur(8px)",
+          color: "rgba(255,255,255,0.9)",
+          fontSize: compact ? 9 : 10,
+          fontWeight: 500,
+          padding: compact ? "1px 6px" : "2px 8px",
+          borderRadius: 999,
+          zIndex: 10,
+          border: "1px solid rgba(255,255,255,0.1)",
+          letterSpacing: 0.3,
+        }}
+      >
+        {`${segmentIndex + 1} / ${segmentCount}`}
+      </div>
+
+      {/* 中央播放按钮 */}
+      {!isPlaying && (
+        <button
+          onClick={onTogglePlay}
+          style={{
+            position: "absolute",
+            top: "50%",
+            left: "50%",
+            transform: "translate(-50%, -50%)",
+            background: "rgba(0,0,0,0.45)",
+            backdropFilter: "blur(12px)",
+            WebkitBackdropFilter: "blur(12px)",
+            border: "1px solid rgba(255,255,255,0.15)",
+            borderRadius: "50%",
+            width: 52,
+            height: 52,
+            cursor: "pointer",
+            display: "flex",
+            alignItems: "center",
+            justifyContent: "center",
+            color: "#fff",
+            fontSize: 26,
+            zIndex: 10,
+            transition: "transform 0.2s ease, background 0.2s ease",
+            boxShadow: "0 4px 20px rgba(0,0,0,0.4)",
+          }}
+          onMouseEnter={(e) => {
+            e.currentTarget.style.transform = "translate(-50%, -50%) scale(1.08)"
+            e.currentTarget.style.background = "rgba(0,0,0,0.6)"
+          }}
+          onMouseLeave={(e) => {
+            e.currentTarget.style.transform = "translate(-50%, -50%) scale(1)"
+            e.currentTarget.style.background = "rgba(0,0,0,0.45)"
+          }}
+        >
+          <PlayCircleOutlined />
+        </button>
+      )}
+
+      {/* 控制条 — 手机风格毛玻璃 */}
+      <div
+        style={{
+          position: "absolute",
+          bottom: 0,
+          left: 0,
+          right: 0,
+          display: "flex",
+          alignItems: "center",
+          gap: compact ? 6 : 10,
+          padding: compact ? "8px 10px 10px" : "12px 16px 16px",
+          background: "linear-gradient(transparent, rgba(0,0,0,0.7))",
+          backdropFilter: "blur(4px)",
+          WebkitBackdropFilter: "blur(4px)",
+          zIndex: 10,
+        }}
+      >
+        <button
+          onClick={onTogglePlay}
+          style={{
+            background: "rgba(255,255,255,0.15)",
+            border: "none",
+            color: "#fff",
+            fontSize: compact ? 14 : 16,
+            cursor: "pointer",
+            width: compact ? 26 : 32,
+            height: compact ? 26 : 32,
+            borderRadius: "50%",
+            display: "flex",
+            alignItems: "center",
+            justifyContent: "center",
+            flexShrink: 0,
+            transition: "background 0.15s",
+          }}
+          onMouseEnter={(e) => {
+            e.currentTarget.style.background = "rgba(255,255,255,0.25)"
+          }}
+          onMouseLeave={(e) => {
+            e.currentTarget.style.background = "rgba(255,255,255,0.15)"
+          }}
+        >
+          {isPlaying ? <PauseCircleOutlined /> : <PlayCircleOutlined />}
+        </button>
+
+        <PreviewProgressBar
+          currentTime={currentTime}
+          totalDuration={totalDuration}
+          onSeek={onSeek}
+          compact={compact}
+        />
+      </div>
+    </>
+  )
+}

apps/web/src/pages/generate/components/PreviewProgressBar.tsx

@@ -0,0 +1,108 @@
+import React, { useCallback, useEffect, useRef, useState } from "react"
+import { formatDuration } from "../utils/formatDuration"
+
+interface PreviewProgressBarProps {
+  currentTime: number
+  totalDuration: number
+  onSeek: (time: number) => void
+  compact?: boolean
+}
+
+/**
+ * 进度条组件：点击/拖拽 seek
+ */
+export const PreviewProgressBar: React.FC<PreviewProgressBarProps> = ({
+  currentTime,
+  totalDuration,
+  onSeek,
+  compact = false,
+}) => {
+  const progressRef = useRef<HTMLDivElement>(null)
+  const [isDragging, setIsDragging] = useState(false)
+
+  const seekByClientX = useCallback(
+    (clientX: number) => {
+      if (!progressRef.current || totalDuration <= 0) return
+      const rect = progressRef.current.getBoundingClientRect()
+      const ratio = Math.max(0, Math.min(1, (clientX - rect.left) / rect.width))
+      onSeek(ratio * totalDuration)
+    },
+    [totalDuration, onSeek],
+  )
+
+  const handleMouseDown = useCallback(
+    (e: React.MouseEvent<HTMLDivElement>) => {
+      setIsDragging(true)
+      seekByClientX(e.clientX)
+    },
+    [seekByClientX],
+  )
+
+  useEffect(() => {
+    if (!isDragging) return
+    const handleMouseMove = (e: MouseEvent) => seekByClientX(e.clientX)
+    const handleMouseUp = () => setIsDragging(false)
+    window.addEventListener("mousemove", handleMouseMove)
+    window.addEventListener("mouseup", handleMouseUp)
+    return () => {
+      window.removeEventListener("mousemove", handleMouseMove)
+      window.removeEventListener("mouseup", handleMouseUp)
+    }
+  }, [isDragging, seekByClientX])
+
+  const progressPercent = totalDuration > 0 ? (currentTime / totalDuration) * 100 : 0
+
+  return (
+    <>
+      <span
+        style={{
+          fontSize: compact ? 10 : 11,
+          color: "rgba(255,255,255,0.85)",
+          minWidth: compact ? 58 : 72,
+          fontVariantNumeric: "tabular-nums",
+          letterSpacing: 0.2,
+        }}
+      >
+        {formatDuration(currentTime)} / {formatDuration(totalDuration)}
+      </span>
+
+      <div
+        ref={progressRef}
+        onMouseDown={handleMouseDown}
+        style={{
+          flex: 1,
+          height: 3,
+          background: "rgba(255,255,255,0.2)",
+          borderRadius: 2,
+          cursor: "pointer",
+          position: "relative",
+        }}
+      >
+        <div
+          style={{
+            height: "100%",
+            width: `${progressPercent}%`,
+            background: "#fff",
+            borderRadius: 2,
+            transition: isDragging ? "none" : "width 0.1s linear",
+          }}
+        />
+        <div
+          style={{
+            position: "absolute",
+            top: "50%",
+            left: `${progressPercent}%`,
+            transform: "translate(-50%, -50%)",
+            width: 10,
+            height: 10,
+            borderRadius: "50%",
+            background: "#fff",
+            boxShadow: "0 0 6px rgba(255,255,255,0.5)",
+            opacity: isDragging ? 1 : 0,
+            transition: "opacity 0.15s",
+          }}
+        />
+      </div>
+    </>
+  )
+}

apps/web/src/pages/generate/constants.ts

@@ -50,15 +50,7 @@ export const POSITION_OPTIONS = [
 ]
 
 /* ── 标题字体选项 ── */
-export const FONT_OPTIONS = [
-  "思源黑体",
-  "思源宋体",
-  "苹方",
-  "PingFang",
-  "微软雅黑",
-  "楷体",
-  "华康俪金黑",
-]
+export const FONT_OPTIONS = ["思源黑体", "思源宋体", "苹方", "微软雅黑", "楷体"]
 
 /* ── 标题字体 CSS font-family 映射（中文显示名 → 浏览器可识别的字体栈） ── */
 export const FONT_FAMILY_MAP: Record<string, string> = {
@@ -68,7 +60,6 @@ export const FONT_FAMILY_MAP: Record<string, string> = {
   PingFang: '"PingFang SC", -apple-system, "Helvetica Neue", sans-serif',
   微软雅黑: '"Microsoft YaHei", "PingFang SC", sans-serif',
   楷体: '"KaiTi", "STKaiti", "DFKai-SB", serif',
-  华康俪金黑: '"华康俪金黑", "DFLiJinHei-W8", "Source Han Sans SC", "Microsoft YaHei", sans-serif',
 }
 
 export function getFontFamily(font: string): string {

apps/web/src/pages/generate/hooks/useBatchCovers.ts

@@ -171,7 +171,6 @@ export function useBatchCovers({
     let okCount = 0
     let failCount = 0
     for (const i of pending) {
-      // eslint-disable-next-line no-await-in-loop
       const ok = await generateOne(i)
       if (ok) okCount += 1
       else failCount += 1

apps/web/src/pages/generate/hooks/useBatchVariantPlans.ts

@@ -26,6 +26,8 @@ export interface BatchVariantClipsState {
   clipsByVariant: EditPlanClip[][]
   /** 各变体的 plan_id（正式生成回传，保证预览即成片）；未就绪为空串 */
   planIdsByVariant: string[]
+  /** 各变体的后端返回配音时长（秒）；未就绪/未返回为 undefined */
+  voiceDurationsByVariant: (number | undefined)[]
   /** 是否正在向后端申请变体计划 */
   loading: boolean
   /** 后端真实片段是否全部可用（每个变体都有 ≥1 条片段） */
@@ -44,6 +46,12 @@ interface UseBatchVariantPlansOptions {
   assetIds: string[]
   /** 源剪辑计划 ID（草稿/预览关联），无则空串由后端兜底最新 plan */
   sourcePlanId?: string
+  /** 统一配音 ID（共用配音模式），参考 useGenerateVideo voiceLibraryId 计算 */
+  voiceLibraryId?: string
+  /** 独立配音 ID 列表（每变体一条），voiceModePerVideo=true 时使用 */
+  voiceLibraryIds?: string[]
+  /** 是否启用独立配音模式（每变体各自一条配音） */
+  voiceModePerVideo?: boolean
 }
 
 export function useBatchVariantPlans({
@@ -52,9 +60,13 @@ export function useBatchVariantPlans({
   templateId,
   assetIds,
   sourcePlanId = "",
+  voiceLibraryId = "",
+  voiceLibraryIds = [],
+  voiceModePerVideo = false,
 }: UseBatchVariantPlansOptions): BatchVariantClipsState {
   const [clipsByVariant, setClipsByVariant] = useState<EditPlanClip[][]>([])
   const [planIdsByVariant, setPlanIdsByVariant] = useState<string[]>([])
+  const [voiceDurationsByVariant, setVoiceDurationsByVariant] = useState<(number | undefined)[]>([])
   const [loading, setLoading] = useState(false)
   const [error, setError] = useState(false)
 
@@ -70,17 +82,30 @@ export function useBatchVariantPlans({
     setLoading(true)
     setError(false)
     try {
+      // 配音参数：与 useGenerateVideo 保持一致的传参逻辑
+      // - 独立配音模式 + voiceLibraryIds 非空：传 voice_library_ids
+      // - 统一配音：传 voice_library_id
+      // - 都没选：不传
+      const voiceParam: { voice_library_id?: string; voice_library_ids?: string[] } = {}
+      if (voiceModePerVideo && voiceLibraryIds.length > 0) {
+        voiceParam.voice_library_ids = voiceLibraryIds
+      } else if (voiceLibraryId) {
+        voiceParam.voice_library_id = voiceLibraryId
+      }
+
       const resp = await createBatchVariantPlans({
         template_id: templateId,
         asset_ids: assetIds,
         count,
         ...(sourcePlanId ? { source_edit_plan_id: sourcePlanId } : {}),
+        ...voiceParam,
       })
       if (seq !== requestSeqRef.current) return
 
       const items: VariantPlan[] = Array.isArray(resp.items) ? resp.items : []
       const clips: EditPlanClip[][] = Array.from({ length: count }, () => [])
       const planIds: string[] = Array.from({ length: count }, () => "")
+      const voiceDurs: (number | undefined)[] = Array.from({ length: count }, () => undefined)
       for (const item of items) {
         const idx = item.variant_index
         if (idx < 0 || idx >= count) continue
@@ -88,6 +113,9 @@ export function useBatchVariantPlans({
         clips[idx] = (item.clips || [])
           .filter((c) => c && c.asset_id && c.status === "ready")
           .sort((a, b) => a.order - b.order)
+        if (typeof item.voice_duration === "number" && item.voice_duration > 0) {
+          voiceDurs[idx] = item.voice_duration
+        }
       }
       // 数据完整性校验：每个变体都必须有真实片段，否则视为失败（不允许假数据冒充）
       const incomplete = clips.some((list) => list.length === 0)
@@ -95,10 +123,12 @@ export function useBatchVariantPlans({
         console.warn("[useBatchVariantPlans] 变体计划数据不完整（存在空片段变体），标记加载失败")
         setClipsByVariant([])
         setPlanIdsByVariant([])
+        setVoiceDurationsByVariant([])
         setError(true)
       } else {
         setClipsByVariant(clips)
         setPlanIdsByVariant(planIds)
+        setVoiceDurationsByVariant(voiceDurs)
         setError(false)
       }
     } catch (err) {
@@ -107,11 +137,20 @@ export function useBatchVariantPlans({
       console.warn("[useBatchVariantPlans] 申请变体计划失败，预览加载失败:", err)
       setClipsByVariant([])
       setPlanIdsByVariant([])
+      setVoiceDurationsByVariant([])
       setError(true)
     } finally {
       if (seq === requestSeqRef.current) setLoading(false)
     }
-  }, [templateId, count, sourcePlanId, assetIds])
+  }, [
+    templateId,
+    count,
+    sourcePlanId,
+    assetIds,
+    voiceLibraryId,
+    voiceLibraryIds,
+    voiceModePerVideo,
+  ])
 
   /** 用户点击「重试」：nonce +1 驱动 effect 重新发起请求（effect 内 lastKey 校验保证只发一次） */
   const retry = useCallback(() => {
@@ -125,24 +164,40 @@ export function useBatchVariantPlans({
       // 避免父组件传入内联字面量数组导致 effect 每次 render 触发 → 无限 setState 循环
       setClipsByVariant((prev) => (prev.length === 0 ? prev : []))
       setPlanIdsByVariant((prev) => (prev.length === 0 ? prev : []))
+      setVoiceDurationsByVariant((prev) => (prev.length === 0 ? prev : []))
       setLoading((prev) => (prev === false ? prev : false))
       setError((prev) => (prev === false ? prev : false))
       lastKeyRef.current = ""
       return
     }
+    const voiceKey = voiceModePerVideo
+      ? `per:${[...voiceLibraryIds].sort().join(",")}`
+      : `one:${voiceLibraryId}`
     const key = `${retryNonce}|${templateId}|${count}|${sourcePlanId}|${[...assetIds]
       .sort()
-      .join(",")}`
+      .join(",")}|${voiceKey}`
     if (key === lastKeyRef.current) return
     lastKeyRef.current = key
     load()
-  }, [enabled, templateId, count, sourcePlanId, assetIds, load, retryNonce])
+  }, [
+    enabled,
+    templateId,
+    count,
+    sourcePlanId,
+    assetIds,
+    load,
+    retryNonce,
+    voiceLibraryId,
+    voiceLibraryIds,
+    voiceModePerVideo,
+  ])
 
   const ready = !error && !loading && clipsByVariant.every((list) => list.length > 0)
 
   return {
     clipsByVariant,
     planIdsByVariant,
+    voiceDurationsByVariant,
     loading,
     ready,
     error,

apps/web/src/pages/generate/hooks/useGenerateFormState/useTitleCoverSync.ts

@@ -51,7 +51,6 @@ export function useTitleCoverSync({
         thumbnail_url: tpl.cover_config!.thumbnail_url || prev.thumbnail_url,
       }))
     }
-    // eslint-disable-next-line react-hooks/exhaustive-deps
   }, [selectedTemplate, setTitleSettings, setCoverSettings])
   // ↑ 移除 userTemplates，只在 selectedTemplate 真正变化时触发
 }

apps/web/src/pages/generate/hooks/usePreviewAudio.ts

@@ -0,0 +1,131 @@
+import { useCallback, useEffect, useRef } from "react"
+
+interface UsePreviewAudioOptions {
+  voiceAudioUrl: string | undefined
+  voiceDurationHint: number | undefined
+  muted: boolean
+  isPlaying: boolean
+  currentTime: number
+  onVoiceDurationChange: (d: number) => void
+  onEnded: () => void
+}
+
+interface UsePreviewAudioReturn {
+  seekTo: (time: number) => void
+  ensurePlayingAt: (time: number) => void
+  pause: () => void
+}
+
+/**
+ * 配音音频管理 hook：加载配音、loadedmetadata 自测时长、play/pause 同步、
+ * ended 事件回调、seek 同步、末帧冻结期间续播。
+ */
+export function usePreviewAudio({
+  voiceAudioUrl,
+  voiceDurationHint,
+  muted,
+  isPlaying,
+  currentTime,
+  onVoiceDurationChange,
+  onEnded,
+}: UsePreviewAudioOptions): UsePreviewAudioReturn {
+  const audioRef = useRef<HTMLAudioElement | null>(null)
+  const prevIsPlayingRef = useRef(false)
+
+  // 外部 hint 初始化（自测值前的兜底）
+  useEffect(() => {
+    if (voiceDurationHint && voiceDurationHint > 0) {
+      onVoiceDurationChange(voiceDurationHint)
+    }
+  }, [voiceDurationHint, onVoiceDurationChange])
+
+  // 创建/替换 audio 元素，加载 metadata 时自测时长并监听 ended
+  useEffect(() => {
+    if (!voiceAudioUrl) {
+      if (audioRef.current) {
+        audioRef.current.pause()
+        audioRef.current.src = ""
+        audioRef.current = null
+      }
+      return
+    }
+    if (!audioRef.current) {
+      audioRef.current = new Audio()
+      audioRef.current.preload = "auto"
+    }
+    if (audioRef.current.src !== voiceAudioUrl) {
+      audioRef.current.src = voiceAudioUrl
+    }
+    audioRef.current.muted = muted
+
+    const audio = audioRef.current
+    const onLoaded = () => {
+      if (audio.duration && isFinite(audio.duration) && audio.duration > 0) {
+        onVoiceDurationChange(audio.duration)
+      }
+    }
+    const onEndedHandler = () => onEnded()
+    audio.addEventListener("loadedmetadata", onLoaded)
+    audio.addEventListener("ended", onEndedHandler)
+    return () => {
+      audio.removeEventListener("loadedmetadata", onLoaded)
+      audio.removeEventListener("ended", onEndedHandler)
+    }
+  }, [voiceAudioUrl, muted, onVoiceDurationChange, onEnded])
+
+  // mute 变化即时同步
+  useEffect(() => {
+    if (audioRef.current) audioRef.current.muted = muted
+  }, [muted])
+
+  // 播放/暂停同步（跟随视频 isPlaying）
+  useEffect(() => {
+    const audio = audioRef.current
+    if (!audio || !audio.src) return
+    if (isPlaying && !prevIsPlayingRef.current) {
+      if (Math.abs(audio.currentTime - currentTime) > 0.3) {
+        try {
+          audio.currentTime = currentTime
+        } catch {
+          /* ignore */
+        }
+      }
+      audio.play().catch(() => {})
+    } else if (!isPlaying && prevIsPlayingRef.current) {
+      audio.pause()
+    }
+    prevIsPlayingRef.current = isPlaying
+  }, [isPlaying, currentTime])
+
+  const seekTo = useCallback((time: number) => {
+    const audio = audioRef.current
+    if (audio && audio.src) {
+      try {
+        audio.currentTime = time
+      } catch {
+        /* ignore */
+      }
+    }
+  }, [])
+
+  const ensurePlayingAt = useCallback((time: number) => {
+    const audio = audioRef.current
+    if (!audio || !audio.src) return
+    try {
+      if (Math.abs(audio.currentTime - time) > 0.5) audio.currentTime = time
+      if (audio.paused) audio.play().catch(() => {})
+    } catch {
+      /* ignore */
+    }
+  }, [])
+
+  const pause = useCallback(() => {
+    try {
+      audioRef.current?.pause()
+    } catch {
+      /* ignore */
+    }
+  }, [])
+
+  return { seekTo, ensurePlayingAt, pause }
+}

apps/web/src/pages/generate/hooks/useVariantVoicePreview.ts

@@ -125,7 +125,6 @@ export function useVariantVoicePreview({
           continue
         }
         try {
-          // eslint-disable-next-line no-await-in-loop
           const res = await previewTts({ text: job.title, voice_id: job.voiceId })
           if (cancelled || controller.signal.aborted || seq !== seqRef.current) return
           const audioUrl = res.audio_url || ""

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/pages/voice-clone/VoiceClone.tsx

@@ -11,6 +11,7 @@ import CloneModal from "@/components/voice/CloneModal"
 import { VoiceCloneCard } from "./components/VoiceCloneCard"
 import { VoiceCloneEmpty, VoiceCloneSkeleton, ToastContainer } from "./components/States"
 import { EditNameDialog } from "./components/EditNameDialog"
+import VoiceClonePreviewPanel from "./components/VoiceClonePreviewPanel"
 import { useVoiceCloneList } from "./hooks/useVoiceCloneList"
 import "./voice-clone.css"
 
@@ -47,6 +48,9 @@ const VoiceClone: React.FC = () => {
         }
       />
 
+      {/* 音色试听面板（自定义文本 + 语速/情绪） */}
+      {!isLoading && voices.length > 0 && <VoiceClonePreviewPanel voices={voices} />}
+
       {/* 加载状态 — 骨架屏 */}
       {isLoading && <VoiceCloneSkeleton />}
 

apps/web/src/pages/voice-clone/components/VoiceClonePreviewPanel.tsx

@@ -0,0 +1,133 @@
+/**
+ * 克隆音色试听面板
+ * - 选择就绪音色、输入试听文本、调节语速/情绪，点试听
+ * - 复用配音库 TTS 弹窗的 SpeedControl / EmotionControl 组件
+ */
+import React, { useState, useRef, useCallback } from "react"
+import { message } from "antd"
+import { SoundOutlined, LoadingOutlined } from "@ant-design/icons"
+import { getVoiceClonePreview, type VoiceClone } from "@/api/voice-clone"
+import SpeedControl from "@/pages/voices/components/tts-modal/SpeedControl"
+import EmotionControl from "@/pages/voices/components/tts-modal/EmotionControl"
+import { DEFAULT_TTS_EMOTION, type TtsEmotion } from "@/pages/voices/components/tts-modal/constants"
+import { TTS_CONFIG } from "@/pages/voices/components/tts-modal/types"
+
+interface VoiceClonePreviewPanelProps {
+  voices: VoiceClone[]
+}
+
+const DEFAULT_PREVIEW_TEXT = "你好呀，欢迎使用小虾智剪，这是我的声音效果，希望你喜欢。"
+
+const VoiceClonePreviewPanel: React.FC<VoiceClonePreviewPanelProps> = ({ voices }) => {
+  const readyVoices = voices.filter((v) => v.status === "ready")
+
+  const [selectedId, setSelectedId] = useState<string>(readyVoices[0]?.id ?? "")
+  const [previewText, setPreviewText] = useState(DEFAULT_PREVIEW_TEXT)
+  const [speed, setSpeed] = useState<number>(TTS_CONFIG.DEFAULT_SPEED)
+  const [emotion, setEmotion] = useState<TtsEmotion>(DEFAULT_TTS_EMOTION)
+  const [previewing, setPreviewing] = useState(false)
+  const [audioUrl, setAudioUrl] = useState<string | null>(null)
+  const audioRef = useRef<HTMLAudioElement | null>(null)
+
+  const handlePreview = useCallback(async () => {
+    if (!selectedId) {
+      message.warning("请先选择要试听的音色")
+      return
+    }
+    const text = previewText.trim()
+    if (!text) {
+      message.warning("请输入试听文本")
+      return
+    }
+    if (audioRef.current) {
+      audioRef.current.pause()
+      audioRef.current = null
+    }
+    setPreviewing(true)
+    setAudioUrl(null)
+    try {
+      const res = await getVoiceClonePreview(selectedId, text, { speed, emotion })
+      setAudioUrl(res.audio_url)
+      const audio = new Audio(res.audio_url)
+      audioRef.current = audio
+      audio.play().catch(() => {
+        message.error("播放失败，请重试")
+      })
+    } catch (err) {
+      const msg = err instanceof Error ? err.message : "试听失败"
+      message.error(msg)
+    } finally {
+      setPreviewing(false)
+    }
+  }, [selectedId, previewText, speed, emotion])
+
+  if (readyVoices.length === 0) return null
+
+  return (
+    <div className="vc-preview-panel">
+      <div className="vc-preview-header">
+        <SoundOutlined style={{ color: "var(--primary-color)" }} />
+        <span className="vc-preview-title">音色试听</span>
+      </div>
+
+      <div className="vc-preview-body">
+        <div className="vc-preview-field">
+          <label className="vc-preview-label">选择音色</label>
+          <select
+            className="vc-preview-select"
+            value={selectedId}
+            onChange={(e) => setSelectedId(e.target.value)}
+          >
+            {readyVoices.map((v) => (
+              <option key={v.id} value={v.id}>
+                {v.name}
+              </option>
+            ))}
+          </select>
+        </div>
+
+        <div className="vc-preview-field">
+          <label className="vc-preview-label">试听文本</label>
+          <textarea
+            className="vc-preview-textarea"
+            value={previewText}
+            onChange={(e) => setPreviewText(e.target.value)}
+            rows={2}
+            maxLength={200}
+            placeholder="输入试听文本（最多200字）"
+          />
+        </div>
+
+        <div className="vc-preview-grid">
+          <div className="vc-preview-field">
+            <SpeedControl speed={speed} onChange={setSpeed} />
+          </div>
+          <div className="vc-preview-field">
+            <EmotionControl emotion={emotion} onChange={setEmotion} />
+          </div>
+        </div>
+
+        <button
+          type="button"
+          className="vc-preview-btn"
+          onClick={handlePreview}
+          disabled={previewing || !selectedId || !previewText.trim()}
+        >
+          {previewing ? (
+            <>
+              <LoadingOutlined /> 合成中...
+            </>
+          ) : (
+            <>▶ 开始试听</>
+          )}
+        </button>
+
+        {audioUrl && !previewing && (
+          <audio controls src={audioUrl} style={{ width: "100%", marginTop: 4 }} />
+        )}
+      </div>
+    </div>
+  )
+}
+
+export default VoiceClonePreviewPanel

apps/web/src/pages/voice-clone/voice-clone.css

@@ -464,3 +464,93 @@
     padding: var(--space-sm);
   }
 }
+
+/* ── 试听面板 ─────────────────────────────────────────── */
+.vc-preview-panel {
+  background: var(--bg-card);
+  border: 1px solid var(--line);
+  border-radius: var(--radius-lg);
+  padding: var(--space-lg);
+}
+
+.vc-preview-header {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  margin-bottom: 12px;
+}
+
+.vc-preview-title {
+  font-size: 15px;
+  font-weight: 600;
+  color: var(--text-primary);
+}
+
+.vc-preview-body {
+  display: flex;
+  flex-direction: column;
+  gap: 12px;
+}
+
+.vc-preview-field {
+  display: flex;
+  flex-direction: column;
+  gap: 6px;
+}
+
+.vc-preview-label {
+  font-size: 13px;
+  color: var(--text-secondary);
+}
+
+.vc-preview-select,
+.vc-preview-textarea {
+  width: 100%;
+  padding: 8px 10px;
+  border-radius: 6px;
+  border: 1px solid var(--line);
+  background: var(--bg-primary);
+  color: var(--text-primary);
+  font-size: 13px;
+  outline: none;
+  font-family: inherit;
+  box-sizing: border-box;
+}
+
+.vc-preview-select:focus,
+.vc-preview-textarea:focus {
+  border-color: var(--primary-color);
+}
+
+.vc-preview-textarea {
+  resize: vertical;
+  min-height: 52px;
+}
+
+.vc-preview-grid {
+  display: grid;
+  grid-template-columns: 1fr 1fr;
+  gap: 12px;
+}
+
+.vc-preview-btn {
+  width: 100%;
+  padding: 10px 0;
+  border-radius: 8px;
+  border: none;
+  background: var(--primary-color);
+  color: #fff;
+  font-size: 14px;
+  font-weight: 600;
+  cursor: pointer;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  gap: 8px;
+  transition: opacity 0.15s;
+}
+
+.vc-preview-btn:disabled {
+  opacity: 0.5;
+  cursor: not-allowed;
+}

apps/web/src/pages/voices/VoiceLibrary.tsx

@@ -132,12 +132,16 @@ const VoiceLibrary: React.FC = () => {
     ttsText,
     ttsVoiceId,
     ttsSpeed,
+    ttsEmotion,
+    ttsLanguage,
     ttsStatus,
     ttsAudioUrl,
     ttsError,
     setTtsText,
     setTtsVoiceId,
     setTtsSpeed,
+    setTtsEmotion,
+    setTtsLanguage,
     setTtsOpen,
     handleTtsSynthesize,
     handleTtsSave,
@@ -363,6 +367,8 @@ const VoiceLibrary: React.FC = () => {
         ttsText={ttsText}
         ttsVoiceId={ttsVoiceId}
         ttsSpeed={ttsSpeed}
+        ttsEmotion={ttsEmotion}
+        ttsLanguage={ttsLanguage}
         ttsStatus={ttsStatus}
         ttsAudioUrl={ttsAudioUrl}
         ttsError={ttsError}
@@ -374,6 +380,8 @@ const VoiceLibrary: React.FC = () => {
         onTtsTextChange={setTtsText}
         onTtsVoiceChange={setTtsVoiceId}
         onTtsSpeedChange={setTtsSpeed}
+        onTtsEmotionChange={setTtsEmotion}
+        onTtsLanguageChange={setTtsLanguage}
         onTtsSynthesize={handleTtsSynthesize}
         onTtsSave={handleTtsSave}
       />

apps/web/src/pages/voices/components/TtsModal.tsx

@@ -4,6 +4,8 @@ import { type TtsModalProps, type TtsStatus } from "./tts-modal/types"
 import TextInputSection from "./tts-modal/TextInputSection"
 import VoiceSelector from "./tts-modal/VoiceSelector"
 import SpeedControl from "./tts-modal/SpeedControl"
+import EmotionControl from "./tts-modal/EmotionControl"
+import LanguageControl from "./tts-modal/LanguageControl"
 import SynthesizeButton from "./tts-modal/SynthesizeButton"
 import ErrorAlert from "./tts-modal/ErrorAlert"
 import ResultPanel from "./tts-modal/ResultPanel"
@@ -14,6 +16,8 @@ const TtsModal: React.FC<TtsModalProps> = ({
   ttsText,
   ttsVoiceId,
   ttsSpeed,
+  ttsEmotion,
+  ttsLanguage,
   ttsStatus,
   ttsAudioUrl,
   ttsError,
@@ -23,6 +27,8 @@ const TtsModal: React.FC<TtsModalProps> = ({
   onTextChange,
   onVoiceChange,
   onSpeedChange,
+  onEmotionChange,
+  onLanguageChange,
   onSynthesize,
   onSave,
 }) => {
@@ -43,6 +49,16 @@ const TtsModal: React.FC<TtsModalProps> = ({
           presetVoices={presetVoices}
           clonedVoices={clonedVoices}
         />
+        <div
+          style={{
+            display: "grid",
+            gridTemplateColumns: "1fr 1fr",
+            gap: 12,
+          }}
+        >
+          <EmotionControl emotion={ttsEmotion} onChange={onEmotionChange} />
+          <LanguageControl language={ttsLanguage} onChange={onLanguageChange} />
+        </div>
         <SpeedControl speed={ttsSpeed} onChange={onSpeedChange} />
         <SynthesizeButton status={ttsStatus} text={ttsText} onClick={onSynthesize} />
         {ttsError && <ErrorAlert error={ttsError} />}

apps/web/src/pages/voices/components/VoiceModals.tsx

@@ -6,6 +6,7 @@ import type { ClonedVoiceDisplay, PresetVoiceDisplay } from "../types"
 import type { VoiceClone } from "@/api/voice-clone"
 import type { TtsStatus } from "./TtsModal"
 import type { TtsClonedVoiceOption } from "./tts-modal/VoiceSelector"
+import type { TtsEmotion, TtsLanguage } from "./tts-modal/constants"
 import CloneModal from "@/components/voice/CloneModal"
 import CloneDetailModal from "./CloneDetailModal"
 import UploadVoiceModal from "./UploadVoiceModal"
@@ -42,6 +43,8 @@ export interface VoiceModalsProps {
   ttsText: string
   ttsVoiceId: string
   ttsSpeed: number
+  ttsEmotion: TtsEmotion
+  ttsLanguage: TtsLanguage
   ttsStatus: TtsStatus
   ttsAudioUrl: string | null
   ttsError: string | null
@@ -52,6 +55,8 @@ export interface VoiceModalsProps {
   onTtsTextChange: (text: string) => void
   onTtsVoiceChange: (id: string) => void
   onTtsSpeedChange: (speed: number) => void
+  onTtsEmotionChange: (emotion: TtsEmotion) => void
+  onTtsLanguageChange: (language: TtsLanguage) => void
   onTtsSynthesize: () => void
   onTtsSave: () => void
 }
@@ -80,6 +85,8 @@ export const VoiceModals: React.FC<VoiceModalsProps> = ({
   ttsText,
   ttsVoiceId,
   ttsSpeed,
+  ttsEmotion,
+  ttsLanguage,
   ttsStatus,
   ttsAudioUrl,
   ttsError,
@@ -89,6 +96,8 @@ export const VoiceModals: React.FC<VoiceModalsProps> = ({
   onTtsTextChange,
   onTtsVoiceChange,
   onTtsSpeedChange,
+  onTtsEmotionChange,
+  onTtsLanguageChange,
   onTtsSynthesize,
   onTtsSave,
 }) => {
@@ -129,6 +138,8 @@ export const VoiceModals: React.FC<VoiceModalsProps> = ({
         ttsText={ttsText}
         ttsVoiceId={ttsVoiceId}
         ttsSpeed={ttsSpeed}
+        ttsEmotion={ttsEmotion}
+        ttsLanguage={ttsLanguage}
         ttsStatus={ttsStatus}
         ttsAudioUrl={ttsAudioUrl}
         ttsError={ttsError}
@@ -138,6 +149,8 @@ export const VoiceModals: React.FC<VoiceModalsProps> = ({
         onTextChange={onTtsTextChange}
         onVoiceChange={onTtsVoiceChange}
         onSpeedChange={onTtsSpeedChange}
+        onEmotionChange={onTtsEmotionChange}
+        onLanguageChange={onTtsLanguageChange}
         onSynthesize={onTtsSynthesize}
         onSave={onTtsSave}
       />

apps/web/src/pages/voices/components/tts-modal/EmotionControl.tsx

@@ -0,0 +1,47 @@
+import React from "react"
+import { TTS_EMOTION_OPTIONS, type TtsEmotion } from "./constants"
+
+interface EmotionControlProps {
+  emotion: TtsEmotion
+  onChange: (emotion: TtsEmotion) => void
+}
+
+/** 情绪选择下拉 */
+const EmotionControl: React.FC<EmotionControlProps> = ({ emotion, onChange }) => {
+  return (
+    <div>
+      <div
+        style={{
+          fontSize: 13,
+          color: "var(--text-secondary)",
+          marginBottom: 6,
+        }}
+      >
+        情绪
+      </div>
+      <select
+        value={emotion}
+        onChange={(e) => onChange(e.target.value as TtsEmotion)}
+        style={{
+          width: "100%",
+          padding: "7px 10px",
+          borderRadius: 6,
+          border: "1px solid var(--border-color, #e5e7eb)",
+          background: "var(--bg-primary, #fff)",
+          color: "var(--text-primary)",
+          fontSize: 13,
+          outline: "none",
+          cursor: "pointer",
+        }}
+      >
+        {TTS_EMOTION_OPTIONS.map((opt) => (
+          <option key={opt.value} value={opt.value}>
+            {opt.label}
+          </option>
+        ))}
+      </select>
+    </div>
+  )
+}
+
+export default EmotionControl

apps/web/src/pages/voices/components/tts-modal/LanguageControl.tsx

@@ -0,0 +1,47 @@
+import React from "react"
+import { TTS_LANGUAGE_OPTIONS, type TtsLanguage } from "./constants"
+
+interface LanguageControlProps {
+  language: TtsLanguage
+  onChange: (language: TtsLanguage) => void
+}
+
+/** 语言选择下拉 */
+const LanguageControl: React.FC<LanguageControlProps> = ({ language, onChange }) => {
+  return (
+    <div>
+      <div
+        style={{
+          fontSize: 13,
+          color: "var(--text-secondary)",
+          marginBottom: 6,
+        }}
+      >
+        语言
+      </div>
+      <select
+        value={language}
+        onChange={(e) => onChange(e.target.value as TtsLanguage)}
+        style={{
+          width: "100%",
+          padding: "7px 10px",
+          borderRadius: 6,
+          border: "1px solid var(--border-color, #e5e7eb)",
+          background: "var(--bg-primary, #fff)",
+          color: "var(--text-primary)",
+          fontSize: 13,
+          outline: "none",
+          cursor: "pointer",
+        }}
+      >
+        {TTS_LANGUAGE_OPTIONS.map((opt) => (
+          <option key={opt.value} value={opt.value}>
+            {opt.label}
+          </option>
+        ))}
+      </select>
+    </div>
+  )
+}
+
+export default LanguageControl

apps/web/src/pages/voices/components/tts-modal/constants.ts

@@ -0,0 +1,22 @@
+/** TTS 情绪选项（对齐后端 CosyVoice 支持：natural/excited/calm/friendly） */
+export const TTS_EMOTION_OPTIONS = [
+  { value: "natural", label: "自然" },
+  { value: "excited", label: "兴奋" },
+  { value: "calm", label: "沉稳" },
+  { value: "friendly", label: "亲切" },
+] as const
+
+export type TtsEmotion = (typeof TTS_EMOTION_OPTIONS)[number]["value"]
+
+/** TTS 语言选项 */
+export const TTS_LANGUAGE_OPTIONS = [
+  { value: "zh-CN", label: "中文" },
+  { value: "en", label: "英文" },
+  { value: "ja", label: "日文" },
+  { value: "ko", label: "韩文" },
+] as const
+
+export type TtsLanguage = (typeof TTS_LANGUAGE_OPTIONS)[number]["value"]
+
+export const DEFAULT_TTS_EMOTION: TtsEmotion = "natural"
+export const DEFAULT_TTS_LANGUAGE: TtsLanguage = "zh-CN"

apps/web/src/pages/voices/components/tts-modal/types.ts

@@ -1,5 +1,6 @@
 import { type PresetVoiceDisplay } from "@/pages/voices/types"
 import type { TtsClonedVoiceOption } from "./VoiceSelector"
+import type { TtsEmotion, TtsLanguage } from "./constants"
 
 export type TtsStatus = "idle" | "synthesizing" | "done" | "error"
 
@@ -8,6 +9,8 @@ export interface TtsModalProps {
   ttsText: string
   ttsVoiceId: string
   ttsSpeed: number
+  ttsEmotion: TtsEmotion
+  ttsLanguage: TtsLanguage
   ttsStatus: TtsStatus
   ttsAudioUrl: string | null
   ttsError: string | null
@@ -18,6 +21,8 @@ export interface TtsModalProps {
   onTextChange: (text: string) => void
   onVoiceChange: (voiceId: string) => void
   onSpeedChange: (speed: number) => void
+  onEmotionChange: (emotion: TtsEmotion) => void
+  onLanguageChange: (language: TtsLanguage) => void
   onSynthesize: () => void
   onSave: () => void
 }

apps/web/src/pages/voices/hooks/useTtsSynthesize.ts

@@ -4,6 +4,12 @@ import { message } from "antd"
 import { synthesizeSpeech, getTTSJobStatus, saveTtsToLibrary } from "@/api/tts"
 import { type PresetVoiceDisplay } from "../types"
 import type { TtsClonedVoiceOption } from "../components/tts-modal/VoiceSelector"
+import {
+  DEFAULT_TTS_EMOTION,
+  DEFAULT_TTS_LANGUAGE,
+  type TtsEmotion,
+  type TtsLanguage,
+} from "../components/tts-modal/constants"
 
 export type TtsStatus = "idle" | "synthesizing" | "done" | "error"
 
@@ -29,6 +35,8 @@ export function useTtsSynthesize({
   const [ttsText, setTtsText] = useState("")
   const [ttsVoiceId, setTtsVoiceId] = useState<string>("")
   const [ttsSpeed, setTtsSpeed] = useState(1.0)
+  const [ttsEmotion, setTtsEmotion] = useState<TtsEmotion>(DEFAULT_TTS_EMOTION)
+  const [ttsLanguage, setTtsLanguage] = useState<TtsLanguage>(DEFAULT_TTS_LANGUAGE)
   const [ttsJobId, setTtsJobId] = useState<string | null>(null)
   const [ttsStatus, setTtsStatus] = useState<TtsStatus>("idle")
   const [ttsAudioUrl, setTtsAudioUrl] = useState<string | null>(null)
@@ -51,6 +59,8 @@ export function useTtsSynthesize({
         text: ttsText.trim(),
         voice_id: ttsVoiceId || undefined,
         speed: ttsSpeed,
+        emotion: ttsEmotion,
+        language: ttsLanguage,
       })
       setTtsJobId(resp.job_id)
 
@@ -81,7 +91,7 @@ export function useTtsSynthesize({
       setTtsStatus("error")
       setTtsError(msg)
     }
-  }, [ttsText, ttsVoiceId, ttsSpeed])
+  }, [ttsText, ttsVoiceId, ttsSpeed, ttsEmotion, ttsLanguage])
 
   /** 保存 TTS 结果到素材库 */
   const handleTtsSave = useCallback(async () => {
@@ -103,6 +113,8 @@ export function useTtsSynthesize({
     setTtsText("")
     setTtsVoiceId("")
     setTtsSpeed(1.0)
+    setTtsEmotion(DEFAULT_TTS_EMOTION)
+    setTtsLanguage(DEFAULT_TTS_LANGUAGE)
     setTtsStatus("idle")
     setTtsAudioUrl(null)
     setTtsError(null)
@@ -132,6 +144,8 @@ export function useTtsSynthesize({
     ttsText,
     ttsVoiceId,
     ttsSpeed,
+    ttsEmotion,
+    ttsLanguage,
     ttsJobId,
     ttsStatus,
     ttsAudioUrl,
@@ -143,6 +157,8 @@ export function useTtsSynthesize({
     setTtsText,
     setTtsVoiceId,
     setTtsSpeed,
+    setTtsEmotion,
+    setTtsLanguage,
     setTtsOpen,
     // Actions
     handleTtsSynthesize,

apps/web/src/test/pages/generate/FrontendPreviewPlayer.audio.test.tsx

@@ -1,7 +1,7 @@
 /**
  * FrontendPreviewPlayer 音频行为单测（Issue #1741 / #1750）
  *
- * useSegmentScheduler/useCanvasPlayer 用 mock 控制播放态，专注验证本组件的音频逻辑：
+ * useSegmentScheduler 用 mock 控制播放态，专注验证本组件的音频逻辑：
  * - 有配音时 video 保持 muted（素材原声不与配音混音）
  * - 无配音时 video 不 muted（素材原声兜底，保证任何情况下播放有声）
  * - 静音按钮：默认有声；点击后切 muted，aria-label 与图标切换

apps/worker/video_processing/render_adapter.py

@@ -221,17 +221,18 @@ class RenderAdapter:
 
         except subprocess.CalledProcessError as exc:
             stderr_text = (exc.stderr or "").strip()
+            stderr_tail = stderr_text[-5000:] if len(stderr_text) > 5000 else stderr_text
             logger.error(
                 "[render-adapter] ffmpeg渲染失败: plan_id=%s job_id=%s exit_code=%d\nstderr:\n%s",
                 plan_id,
                 job_id,
                 exc.returncode,
-                stderr_text[-2000:] if len(stderr_text) > 2000 else stderr_text,
+                stderr_tail,
             )
             return RenderAdapterResult(
                 success=False,
-                error_message=f"FFmpeg渲染失败(exit={exc.returncode}): {stderr_text[:200]}",
-                error_detail=stderr_text[-2000:] if len(stderr_text) > 2000 else stderr_text,
+                error_message=f"FFmpeg渲染失败(exit={exc.returncode}): {stderr_text[-500:]}",
+                error_detail=stderr_tail,
             )
         except Exception as exc:
             logger.exception(
@@ -721,16 +722,17 @@ class RenderAdapter:
 
         except subprocess.CalledProcessError as exc:
             stderr_text = (exc.stderr or "").strip()
+            stderr_tail = stderr_text[-5000:] if len(stderr_text) > 5000 else stderr_text
             logger.error(
                 "[render-adapter] 内存模式渲染失败: plan_id=%s exit_code=%d\nstderr:\n%s",
                 actual_plan_id,
                 exc.returncode,
-                stderr_text[-2000:] if len(stderr_text) > 2000 else stderr_text,
+                stderr_tail,
             )
             return RenderAdapterResult(
                 success=False,
-                error_message=f"FFmpeg渲染失败(exit={exc.returncode}): {stderr_text[:200]}",
-                error_detail=stderr_text[-2000:] if len(stderr_text) > 2000 else stderr_text,
+                error_message=f"FFmpeg渲染失败(exit={exc.returncode}): {stderr_text[-500:]}",
+                error_detail=stderr_tail,
             )
         except Exception as exc:
             logger.exception(

apps/worker/video_processing/subtitle_render_engine.py

@@ -321,8 +321,14 @@ def build_subtitles_from_plan(
 
     has_any = False
 
-    # 1. 标题
-    title_cfg = plan_config.get("title_config") or {}
+    # 1. 标题（#1901 统一读 "title"，兼容老数据 "title_config"）
+    title_cfg = plan_config.get("title") or {}
+    if not isinstance(title_cfg, dict) or not (title_cfg.get("text") or "").strip():
+        _alt = plan_config.get("title_config") or {}
+        if isinstance(_alt, dict):
+            title_cfg = _alt
+    if not isinstance(title_cfg, dict):
+        title_cfg = {}
     if isinstance(title_cfg, dict):
         title_text = str(title_cfg.get("text", ""))
         title_enabled = title_cfg.get("enabled", True)

apps/worker/video_processing/unified_render_service.py

@@ -572,7 +572,12 @@ class UnifiedRenderService:
             ASS 文件路径，没有字幕时返回 None
         """
         config = self.plan.config or {}
+        # #1901 统一读 "title"，兼容老数据 "title_config"
         title_cfg = config.get("title", {}) or {}
+        if not isinstance(title_cfg, dict) or not (title_cfg.get("text") or "").strip():
+            _alt = config.get("title_config") or {}
+            if isinstance(_alt, dict):
+                title_cfg = _alt
         if not isinstance(title_cfg, dict):
             title_cfg = {}
         subtitle_cfg = config.get("subtitle", {}) or {}
@@ -2301,13 +2306,13 @@ class UnifiedRenderService:
                     filters.append(f"eq=contrast={contrast:.3f}")
 
             elif filt == "color_balance":
-                # RGB 通道偏移：color_balance=rs=...:gs=...:bs=...
+                # RGB 通道偏移：colorbalance=rs=...:gs=...:bs=...
                 r = pixel_pert.get("color_r", 0)
                 g = pixel_pert.get("color_g", 0)
                 b = pixel_pert.get("color_b", 0)
                 if r != 0 or g != 0 or b != 0:
                     # color_balance 参数范围 -1.0 ~ 1.0，这里用 /100 转换
-                    filters.append(f"color_balance=rs={r/100:.3f}:gs={g/100:.3f}:bs={b/100:.3f}")
+                    filters.append(f"colorbalance=rs={r/100:.3f}:gs={g/100:.3f}:bs={b/100:.3f}")
 
     @staticmethod
     def _clip_volume(clip: ResolvedClip) -> float:

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()

apps/worker/worker_app/tasks/generation.py

@@ -705,7 +705,8 @@ def _render_from_edit_plan(
         )
 
         if not result.success:
-            raise RuntimeError(f"渲染失败: {result.error_message}")
+            detail_suffix = f"\n[detail] {result.error_detail}" if result.error_detail else ""
+            raise RuntimeError(f"渲染失败: {result.error_message}{detail_suffix}")
 
         render_elapsed = time.monotonic() - render_start
         logger.info(

docs/membership-points-design.md

@@ -0,0 +1,907 @@
+# 会员制 + 积分方案设计文档
+
+> **Issue**: [#1895](https://git.xiaoxiajianji.com/xiaoxia/xiaoxia-saas/issues/1895)
+> **版本**: v1.0
+> **状态**: 方案设计（待确认）
+> **创建时间**: 2026-09-14
+
+---
+
+## 一、设计背景与目标
+
+### 1.1 背景
+
+当前系统采用**纯订阅制**计费模式（4 档：free/standard/pro/enterprise），存在以下问题：
+
+1. **灵活性不足**：用户付费后只能按套餐配额使用，无法按需消费高频 AI 能力
+2. **前后端不一致**：前端 Plans.tsx 写死 3 档，后端硬编码 4 档
+3. **积分能力缺失**：`quota.py` 已预留 `AI_VOICE_CREDITS` 枚举，`module_registry.py` 已注册消耗规则，但无实际账户、流水、扣费和充值链路
+4. **收入天花板**：纯订阅制难以覆盖 AI 调用成本波动，高频用户和低频用户无法差异化变现
+
+### 1.2 目标
+
+- **会员制**保留基础权益（存储、项目数、并发数等），提供稳定收入
+- **积分制**覆盖 AI 消耗型功能（配音、数字人、视频生成等），按需付费、用多少扣多少
+- 两者结合，既降低轻度用户入门门槛，又提升重度用户 ARPU
+- 在现有 `QuotaChecker` + `ModuleRegistry` 架构上扩展，不推倒重来
+
+---
+
+## 二、会员等级设计
+
+### 2.1 等级定义
+
+保留 4 档，与现有后端 `PLAN_QUOTAS` 对齐，重新定义权益：
+
+| 等级 | 月付价格 | 年付价格 | 定位 |
+|------|----------|----------|------|
+| **免费版（free）** | ¥0 | ¥0 | 体验用户，验证产品价值 |
+| **标准版（standard）** | ¥99/月 | ¥999/年（约 ¥83/月） | 个人创作者 |
+| **专业版（pro）** | ¥299/月 | ¥2,999/年（约 ¥250/月） | 专业团队 |
+| **企业版（enterprise）** | ¥999/月 | ¥9,999/年（约 ¥833/月） | 企业 / 工作室 |
+
+### 2.2 会员权益对照表
+
+| 权益维度 | 免费版 | 标准版 | 专业版 | 企业版 |
+|----------|--------|--------|--------|--------|
+| 项目数量 | 3 个 | 10 个 | 不限 | 不限 |
+| 存储空间 | 10 GB | 50 GB | 100 GB | 1 TB |
+| 并发任务数 | 3 | 10 | 20 | 50 |
+| 导出分辨率 | 720p | 1080p | 4K | 4K |
+| 模板数量 | 3 个 | 15 个 | 不限 | 不限 |
+| 标题库 | 50 条 | 500 条 | 500 条 | 不限 |
+| 配音库 | 10 条 | 100 条 | 100 条 | 不限 |
+| AI 配音 | ✗ | ✓ | ✓ | ✓ |
+| 批量导出 | ✗ | ✓ | ✓ | ✓ |
+| 多平台发布 | ✗ | ✗ | ✓ | ✓ |
+| 去重检测报告 | ✗ | ✗ | ✓ | ✓ |
+| **每月赠送积分** | 0 | 200 | 800 | 3,000 |
+| 技术支持 | 社区 | 邮件 | 优先响应 | 专属客服 |
+
+> 注：`AI_VOICE_ENABLED`、`BATCH_EXPORT_ENABLED`、`MULTI_PLATFORM_ENABLED`、`DEDUP_REPORT_ENABLED` 等开关类权益由现有 `QuotaRegistry` 直接控制，无需积分参与。
+
+### 2.3 与现有代码的映射
+
+- 后端 `PLAN_QUOTAS` 字典扩展，新增 `monthly_credits` 字段
+- 前端 `Plans.tsx` 对齐后端 4 档，与后端保持一致
+- 现有 `QuotaTier.limits` 中扩展 `monthly_credits` 维度，由 `QuotaRegistry` 统一管理
+
+---
+
+## 三、积分体系设计
+
+### 3.1 积分获取方式
+
+| 获取方式 | 说明 | 频率 |
+|----------|------|------|
+| **会员每月赠送** | 各等级每月自动到账（见上表） | 每月 1 日 00:00 自动发放 |
+| **单独充值** | 用户按需购买积分包 | 随时可买 |
+| **任务奖励** | 完成指定任务赠送（新手引导、邀请好友、反馈 Bug 等） | 一次性 |
+
+### 3.2 积分包定价
+
+| 积分包 | 积分数量 | 价格 | 单价 | 备注 |
+|--------|----------|------|------|------|
+| 体验包 | 50 积分 | ¥9.9 | ¥0.198/积分 | 首次购买限购 1 次 |
+| 基础包 | 200 积分 | ¥36 | ¥0.18/积分 | |
+| 标准包 | 500 积分 | ¥80 | ¥0.16/积分 | |
+| 专业包 | 1,500 积分 | ¥210 | ¥0.14/积分 | 热门 |
+| 企业包 | 5,000 积分 | ¥600 | ¥0.12/积分 | |
+
+> 积分永久有效，不随会员过期清零。会员过期后停止每月赠送，但已有积分不受影响。
+
+### 3.3 任务奖励规则
+
+| 任务 | 奖励积分 | 次数限制 |
+|------|----------|----------|
+| 新用户注册 | 50 | 1 次 |
+| 完善个人信息 | 20 | 1 次 |
+| 邀请新用户注册 | 30/人 | 每月限 10 人 |
+| 首次完成视频生成 | 20 | 1 次 |
+| 提交有效 Bug 反馈 | 50 | 不限（审核通过后发放） |
+
+---
+
+## 四、积分消耗场景
+
+### 4.1 消耗清单
+
+| 功能模块 | 消耗场景 | 每次消耗积分 | 说明 |
+|----------|----------|-------------|------|
+| **AI 配音** | 生成一条配音 | 1 | 已存在于 `module_registry.py` |
+| **AI 数字人** | 生成一段数字人视频片段 | 5 | 新场景 |
+| **视频生成** | 生成一段 AI 视频（≤10s） | 10 | 新场景 |
+| **视频生成（长时长）** | 每增加 10s | +5 | 累进计费 |
+| **抖音文案提取** | 提取一次 | 1 | 新场景 |
+| **AI 改写** | 改写一段文案 | 1 | 新场景 |
+| **AI 标题生成** | 批量生成一次（≤10 条） | 1 | 新场景 |
+| **AI 封面生成** | 生成一张封面 | 2 | 新场景 |
+| **AI 字幕翻译** | 翻译一条字幕（≤50 字） | 1 | 新场景 |
+
+### 4.2 消耗规则扩展机制
+
+复用现有 `ModuleRegistry` + `QuotaRule` 模式：
+
+```python
+# 新增模块注册示例
+module_registry.register(Module(
+    name="ai_digital_human",
+    version="1.0.0",
+    description="AI 数字人生成模块",
+    capabilities=[
+        ModuleCapability(
+            name="generate_digital_human",
+            description="生成数字人视频片段",
+            quota_rules=[QuotaRule("points", 5.0, "每段数字人视频消耗 5 积分")],
+        ),
+    ],
+))
+```
+
+新增 `QuotaDimension.POINTS = "points"` 作为通用积分维度，所有消耗型功能统一通过积分维度扣费。
+
+### 4.3 消费折扣（预留）
+
+未来可扩展按会员等级设置折扣：
+
+| 等级 | 积分消耗折扣 |
+|------|-------------|
+| 免费版 | 无折扣 |
+| 标准版 | 9.5 折 |
+| 专业版 | 9 折 |
+| 企业版 | 8 折 |
+
+> 折扣仅在会员有效期内生效，过期后恢复原价。实现时通过 `discount_rate` 配置字段支持。
+
+---
+
+## 五、数据库表设计
+
+### 5.1 新增表结构
+
+#### 5.1.1 积分账户表 `points_accounts`
+
+每个用户一个积分账户，记录余额和累计值。
+
+```sql
+CREATE TABLE IF NOT EXISTS points_accounts (
+    id VARCHAR(36) PRIMARY KEY,
+    user_id VARCHAR(36) NOT NULL UNIQUE REFERENCES users(id) ON DELETE CASCADE,
+    balance INTEGER NOT NULL DEFAULT 0,              -- 当前可用积分
+    total_earned INTEGER NOT NULL DEFAULT 0,         -- 累计获得积分
+    total_spent INTEGER NOT NULL DEFAULT 0,          -- 累计消耗积分
+    total_recharged INTEGER NOT NULL DEFAULT 0,      -- 累计充值积分
+    total_gifted INTEGER NOT NULL DEFAULT 0,         -- 累计赠送积分（会员赠送 + 任务奖励）
+    created_at TIMESTAMP NOT NULL DEFAULT NOW(),
+    updated_at TIMESTAMP NOT NULL DEFAULT NOW()
+);
+
+CREATE INDEX idx_points_accounts_user ON points_accounts(user_id);
+```
+
+**设计说明**：
+- `user_id` 设为 UNIQUE，每个用户只有一个积分账户
+- 余额通过 `total_earned - total_spent` 可交叉校验 `balance`，保证数据一致性
+- 不使用悲观锁，而是通过事务 + 乐观锁（`updated_at`）保证并发安全
+
+#### 5.1.2 积分流水表 `points_transactions`
+
+每笔积分变动都记录一条流水，支持对账和审计。
+
+```sql
+CREATE TABLE IF NOT EXISTS points_transactions (
+    id VARCHAR(36) PRIMARY KEY,
+    user_id VARCHAR(36) NOT NULL REFERENCES users(id) ON DELETE CASCADE,
+    account_id VARCHAR(36) NOT NULL REFERENCES points_accounts(id) ON DELETE CASCADE,
+    type VARCHAR(20) NOT NULL,           -- earn(获得) / spend(消耗) / refund(退还) / expire(过期)
+    source VARCHAR(50) NOT NULL,         -- recharge(充值) / membership_gift(会员赠送) / task_reward(任务奖励) / ai_voice / ai_digital_human / ai_video / ...
+    amount INTEGER NOT NULL,             -- 变动数量（正数）
+    balance_after INTEGER NOT NULL,      -- 变动后余额
+    description VARCHAR(255) DEFAULT '', -- 描述
+    ref_id VARCHAR(100) DEFAULT '',      -- 关联业务 ID（订单号、任务 ID 等）
+    created_at TIMESTAMP NOT NULL DEFAULT NOW()
+);
+
+CREATE INDEX idx_points_tx_user ON points_transactions(user_id);
+CREATE INDEX idx_points_tx_type ON points_transactions(type);
+CREATE INDEX idx_points_tx_source ON points_transactions(source);
+CREATE INDEX idx_points_tx_created ON points_transactions(created_at);
+```
+
+**设计说明**：
+- `source` 字段标识具体来源/场景，新增消耗场景时只需新增 source 值，不需要改表结构
+- `ref_id` 关联具体业务，方便追溯（如充值关联订单号，消费关联生成任务 ID）
+- 流水只追加不修改，保证审计完整性
+
+#### 5.1.3 积分订单表 `points_orders`
+
+记录用户充值积分的支付订单。
+
+```sql
+CREATE TABLE IF NOT EXISTS points_orders (
+    id VARCHAR(36) PRIMARY KEY,
+    user_id VARCHAR(36) NOT NULL REFERENCES users(id) ON DELETE CASCADE,
+    package_name VARCHAR(50) NOT NULL,   -- 积分包名称
+    points_amount INTEGER NOT NULL,      -- 积分数量
+    price_cents INTEGER NOT NULL,        -- 支付金额（分），避免浮点精度问题
+    currency VARCHAR(10) NOT NULL DEFAULT 'CNY',
+    status VARCHAR(20) NOT NULL DEFAULT 'pending',  -- pending / paid / failed / refunded
+    payment_method VARCHAR(50),          -- alipay / wechat_pay / ...
+    payment_id VARCHAR(100),             -- 第三方支付流水号
+    paid_at TIMESTAMP,
+    created_at TIMESTAMP NOT NULL DEFAULT NOW(),
+    expire_at TIMESTAMP                  -- 订单过期时间（未支付自动关闭）
+);
+
+CREATE INDEX idx_points_orders_user ON points_orders(user_id);
+CREATE INDEX idx_points_orders_status ON points_orders(status);
+```
+
+#### 5.1.4 积分消耗配置表 `points_consumption_config`
+
+集中管理各场景的积分消耗规则，支持动态调整。
+
+```sql
+CREATE TABLE IF NOT EXISTS points_consumption_config (
+    id VARCHAR(36) PRIMARY KEY,
+    scene_key VARCHAR(50) NOT NULL UNIQUE,   -- 场景标识，如 ai_voice, ai_video
+    scene_name VARCHAR(100) NOT NULL,         -- 场景显示名称
+    points_per_use INTEGER NOT NULL DEFAULT 1, -- 每次消耗积分数
+    is_active BOOLEAN NOT NULL DEFAULT TRUE,  -- 是否启用
+    description VARCHAR(255) DEFAULT '',
+    created_at TIMESTAMP NOT NULL DEFAULT NOW(),
+    updated_at TIMESTAMP NOT NULL DEFAULT NOW()
+);
+```
+
+### 5.2 现有表变更
+
+#### 5.2.1 users 表新增字段
+
+```sql
+-- 会员等级字段保留现有 subscription_plan，无需改动
+-- 无需在 users 表加 credits 字段，积分独立在 points_accounts 表管理
+```
+
+### 5.3 表关系
+
+```
+users (1) ──── (1) points_accounts
+                    │
+                    └── (1:N) points_transactions
+                              │
+                              └── ref_id ──> generation_tasks / points_orders / ...
+
+users (1) ──── (1:N) points_orders
+```
+
+### 5.4 与现有模型的映射
+
+| 现有模型 | 改动 |
+|----------|------|
+| `UserModel` | 不新增积分字段，积分由独立表管理 |
+| `BillingRecordModel` | 保留，用于订阅支付记录；积分充值走新的 `points_orders` |
+| `QuotaDimension` | 新增 `POINTS = "points"` 维度 |
+| `QuotaTier` | 各等级新增 `monthly_credits` 配额 |
+| `QuotaRule` | `dimension` 支持 `"points"` 值 |
+
+---
+
+## 六、API 设计
+
+### 6.1 积分账户 API
+
+#### `GET /api/v1/points/balance`
+
+获取当前积分余额。
+
+```json
+// Response
+{
+  "balance": 580,
+  "total_earned": 1200,
+  "total_spent": 620,
+  "membership_monthly_gift": 200,
+  "membership_expires_at": "2027-01-15T00:00:00Z"
+}
+```
+
+#### `GET /api/v1/points/transactions`
+
+查询积分流水，支持分页和筛选。
+
+```
+?page=1&page_size=20&type=spend&source=ai_voice&start_date=2026-09-01&end_date=2026-09-30
+```
+
+```json
+// Response
+{
+  "total": 156,
+  "page": 1,
+  "page_size": 20,
+  "items": [
+    {
+      "id": "tx_xxx",
+      "type": "spend",
+      "source": "ai_voice",
+      "amount": 1,
+      "balance_after": 579,
+      "description": "AI 配音 - 温柔女声",
+      "ref_id": "task_xxx",
+      "created_at": "2026-09-14T10:30:00Z"
+    }
+  ]
+}
+```
+
+### 6.2 积分充值 API
+
+#### `POST /api/v1/points/recharge`
+
+创建积分充值订单。
+
+```json
+// Request
+{
+  "package_id": "standard_pack"  // 或自定义 amount
+}
+
+// Response
+{
+  "order_id": "po_xxx",
+  "package_name": "标准包",
+  "points_amount": 500,
+  "price_cents": 8000,
+  "payment_url": "https://pay.alipay.com/...",
+  "expire_at": "2026-09-14T16:30:00Z"
+}
+```
+
+#### `POST /api/v1/points/payment-callback`
+
+支付回调（内部接口 + 第三方支付通知）。
+
+```json
+// Request (来自支付平台)
+{
+  "order_id": "po_xxx",
+  "payment_method": "alipay",
+  "payment_id": "2026xxx",
+  "status": "paid"
+}
+```
+
+回调处理流程：
+1. 验证支付签名
+2. 更新 `points_orders.status = "paid"`
+3. 增加 `points_accounts.balance += points_amount`
+4. 写入 `points_transactions` 流水（type=earn, source=recharge）
+
+### 6.3 积分消费 API
+
+#### 内部扣费接口（供各功能模块调用）
+
+```python
+# packages/domain/points_service.py
+class PointsService:
+    def deduct(self, user_id: str, scene_key: str, amount: int, ref_id: str = "") -> DeductResult:
+        """
+        扣减积分
+        1. 检查积分余额是否充足
+        2. 在事务中扣减余额、写入流水
+        3. 返回扣减结果
+        """
+        pass
+    
+    def check_balance(self, user_id: str, scene_key: str) -> CheckResult:
+        """
+        检查余额是否充足某场景消耗
+        """
+        pass
+```
+
+#### `GET /api/v1/points/consumption-rules`
+
+查询当前所有积分消耗规则（前端展示用）。
+
+```json
+// Response
+{
+  "rules": [
+    { "scene_key": "ai_voice", "scene_name": "AI 配音", "points_per_use": 1 },
+    { "scene_key": "ai_digital_human", "scene_name": "AI 数字人", "points_per_use": 5 },
+    { "scene_key": "ai_video", "scene_name": "视频生成", "points_per_use": 10 },
+    ...
+  ]
+}
+```
+
+### 6.4 会员订阅 API（改造）
+
+保留现有 `subscription.py` 路由结构，新增以下逻辑：
+
+- `POST /subscription/change-plan`：变更套餐时，自动创建/更新积分账户，发放当月赠送积分
+- `GET /subscription/current`：返回中增加 `monthly_credits` 和 `points_balance` 字段
+- 新增 `POST /subscription/claim-monthly-credits`：手动领取每月赠送积分（兜底入口）
+
+### 6.5 API 路由汇总
+
+| 方法 | 路径 | 说明 | 类型 |
+|------|------|------|------|
+| GET | `/api/v1/points/balance` | 查询积分余额 | 用户 |
+| GET | `/api/v1/points/transactions` | 积分流水查询 | 用户 |
+| POST | `/api/v1/points/recharge` | 创建充值订单 | 用户 |
+| POST | `/api/v1/points/payment-callback` | 支付回调 | 内部 |
+| GET | `/api/v1/points/consumption-rules` | 消耗规则查询 | 用户 |
+| POST | `/api/v1/points/check` | 消费前余额检查 | 内部 |
+| POST | `/api/v1/points/deduct` | 消费扣减 | 内部 |
+| POST | `/api/v1/points/refund` | 消费退还 | 内部 |
+
+---
+
+## 七、前端页面设计
+
+### 7.1 会员购买页（改造 Plans.tsx）
+
+**路由**: `/app/subscription`（保持不变）
+
+**改动要点**：
+1. 将 3 档对齐为 4 档，与后端一致
+2. 每档卡片增加"每月赠送 XXX 积分"标识
+3. 按钮文案按当前状态动态显示（"当前方案"/"升级"/"降级"/"联系我们"）
+4. 底部增加积分包购买入口
+
+```
+┌─────────────────────────────────────────────────────────────┐
+│                   选择适合您的方案                            │
+├──────────┬──────────┬──────────┬──────────┐                 │
+│  免费版  │  标准版  │ 专业版 ★ │  企业版  │                 │
+│  ¥0/月   │ ¥99/月   │ ¥299/月  │ ¥999/月  │                 │
+│          │          │          │          │                 │
+│  3 项目  │ 10 项目  │ 无限项目 │ 无限项目 │                 │
+│  10GB    │  50GB    │  100GB   │  1TB     │                 │
+│  0 积分  │ 200积分  │ 800积分  │ 3000积分 │  ← 新增        │
+│          │          │          │          │                 │
+│ [当前]   │ [升级]   │ [升级]   │ [联系]   │                 │
+├──────────┴──────────┴──────────┴──────────┘                 │
+│                                                             │
+│  💰 积分充值                              [查看全部积分包]   │
+│  ┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐            │
+│  │50积分 │ │200积分│ │500积分│ │1500积分│ │5000积分│           │
+│  │ ¥9.9 │ │ ¥36  │ │ ¥80  │ │ ¥210 │ │ ¥600 │            │
+│  └──────┘ └──────┘ └──────┘ └──────┘ └──────┘            │
+└─────────────────────────────────────────────────────────────┘
+```
+
+### 7.2 积分余额展示位置
+
+#### 顶部导航栏（Header 组件）
+
+在用户头像旁边增加积分余额徽章：
+
+```
+┌─────────────────────────────────────────────────┐
+│  Logo    工作台  模板  素材      🔔  💎 580  👤  │
+└─────────────────────────────────────────────────┘
+```
+
+- 点击积分徽章展开快捷面板，显示余额、本月已用、充值入口
+- 积分不足时徽章变为警告色（橙色）
+
+#### 功能入口处的消耗提示
+
+在每个 AI 功能的操作按钮旁，显示本次操作将消耗的积分：
+
+```
+[生成配音] 💎 -1 积分
+[生成视频] 💎 -10 积分
+```
+
+### 7.3 积分中心页面
+
+**路由**: `/app/points`（新增）
+
+**页面结构**：
+
+```
+┌─────────────────────────────────────────────────┐
+│  积分中心                                        │
+├─────────────────────────────────────────────────┤
+│                                                 │
+│  当前余额           本月获得        本月消耗      │
+│   💎 580            +200           -120          │
+│                                                 │
+├─────────────────────────────────────────────────┤
+│                                                 │
+│  [积分明细]  [充值记录]  [消耗规则]              │
+│  ─────────                                     │
+│                                                 │
+│  时间         类型    场景        数量   余额     │
+│  09-14 10:30  消耗    AI配音      -1    579     │
+│  09-14 09:15  消耗    AI视频     -10    580     │
+│  09-01 00:00  获得    会员赠送   +200   590     │
+│  08-28 14:20  获得    充值+500   +500   390     │
+│  ...                                            │
+│                                                 │
+└─────────────────────────────────────────────────┘
+```
+
+### 7.4 前端组件清单
+
+| 组件 | 路径 | 说明 |
+|------|------|------|
+| `PointsBadge` | `components/common/PointsBadge/` | 顶部积分余额徽章 |
+| `PointsPanel` | `components/common/PointsPanel/` | 点击徽章展开的快捷面板 |
+| `PointsCost` | `components/common/PointsCost/` | 功能入口的消耗提示标签 |
+| `PointsCenter` | `pages/points/Center.tsx` | 积分中心主页面 |
+| `PointsTransactions` | `pages/points/Transactions.tsx` | 积分明细子页 |
+| `PointsPackages` | `pages/points/Packages.tsx` | 充值积分包子页 |
+| `Plans` (改造) | `pages/subscription/Plans.tsx` | 对齐 4 档 + 积分展示 |
+
+---
+
+## 八、与现有订阅制的迁移方案
+
+### 8.1 迁移原则
+
+1. **向前兼容**：迁移期间老用户权益不降低
+2. **平滑过渡**：不需要用户手动操作，自动完成
+3. **灰度发布**：按用户批次逐步迁移，降低风险
+
+### 8.2 迁移步骤
+
+#### Step 1：数据准备（后端 + DBA）
+
+```sql
+-- 1. 创建新表（points_accounts, points_transactions, points_orders, points_consumption_config）
+-- 2. 为所有现有用户创建积分账户
+INSERT INTO points_accounts (id, user_id, balance, total_earned, total_gifted)
+SELECT 
+    uuid(), id, 
+    CASE subscription_plan
+        WHEN 'standard' THEN 200
+        WHEN 'pro' THEN 800
+        WHEN 'enterprise' THEN 3000
+        ELSE 0
+    END,
+    CASE subscription_plan
+        WHEN 'standard' THEN 200
+        WHEN 'pro' THEN 800
+        WHEN 'enterprise' THEN 3000
+        ELSE 0
+    END,
+    CASE subscription_plan
+        WHEN 'standard' THEN 200
+        WHEN 'pro' THEN 800
+        WHEN 'enterprise' THEN 3000
+        ELSE 0
+    END
+FROM users WHERE subscription_status = 'active';
+```
+
+#### Step 2：代码兼容层
+
+```python
+# packages/domain/quota.py 扩展
+class QuotaDimension(str, Enum):
+    # ... 现有维度保持不变
+    POINTS = "points"  # 新增：通用积分维度
+
+# QUOTA_TIERS 扩展
+"standard": QuotaTier(
+    name="standard",
+    limits={
+        # ... 现有配额保持不变
+        QuotaDimension.MONTHLY_CREDITS: 200,  # 新增
+    },
+),
+```
+
+#### Step 3：双轨运行期（1 个月）
+
+- 订阅制功能不变，老用户正常续费
+- 积分系统上线后，所有 AI 消耗型功能改为积分扣费
+- 会员权益中的"基础配额"（存储、项目数等）继续由订阅制控制
+- 在用户首次登录后，弹出迁移通知弹窗，说明变更内容
+
+#### Step 4：完全切换
+
+- 停止订阅制的 AI 配额逻辑（`AI_VOICE_CREDITS` 等旧维度废弃）
+- 所有 AI 功能统一使用积分扣费
+- 订阅制仅控制基础权益（存储、项目数、并发数、导出分辨率等）
+
+### 8.3 老用户过渡策略
+
+| 用户类型 | 过渡方案 |
+|----------|----------|
+| 当前 free 用户 | 不变，积分余额为 0，可充值 |
+| 当前 standard 用户 | 赠送 200 积分作为过渡礼包，当前周期内权益不变 |
+| 当前 pro 用户 | 赠送 800 积分 + 延长 1 个月有效期 |
+| 当前 enterprise 用户 | 赠送 3000 积分 + 延长 1 个月有效期 + 专属客户经理通知 |
+| 年付用户 | 按剩余月数比例折算赠送积分 |
+
+### 8.4 前端迁移
+
+- `Plans.tsx` 从 3 档改为 4 档，增加积分信息展示
+- 新增 `/app/points` 积分中心页面
+- 顶部 Header 增加积分余额徽章
+- 各 AI 功能页增加积分消耗提示
+
+---
+
+## 九、配额检查中间件设计
+
+### 9.1 整体架构
+
+```
+用户请求 → API Route → 积分检查中间件 → 业务逻辑 → 返回结果
+                          │
+                          ├─ 检查会员权益（QuotaChecker）
+                          ├─ 检查积分余额（PointsChecker）
+                          └─ 扣减积分（PointsService.deduct）
+```
+
+### 9.2 中间件设计
+
+#### `apps/api/app/middleware/points_check.py`
+
+```python
+"""积分扣费中间件 - 用于 AI 功能入口的统一检查与扣费"""
+
+from functools import wraps
+from fastapi import HTTPException
+
+def require_points(scene_key: str, amount: int = None):
+    """
+    装饰器：在 AI 功能入口检查积分余额并扣费
+    
+    Args:
+        scene_key: 消耗场景标识，如 "ai_voice", "ai_video"
+        amount: 指定消耗数量，为 None 时从 points_consumption_config 读取
+    
+    使用方式：
+        @router.post("/generate-voice")
+        @require_points(scene_key="ai_voice")
+        async def generate_voice(request: VoiceRequest, current_user = Depends(get_current_user)):
+            # 到这里积分已扣减成功
+            ...
+    """
+    def decorator(func):
+        @wraps(func)
+        async def wrapper(*args, **kwargs):
+            # 1. 从 kwargs 或 args 中提取 current_user
+            current_user = kwargs.get("current_user") or next(
+                (a for a in args if isinstance(a, AuthenticatedUser)), None
+            )
+            if not current_user:
+                raise HTTPException(status_code=401, detail="未登录")
+            
+            # 2. 获取消耗数量
+            consume_amount = amount or get_consumption_config(scene_key)
+            
+            # 3. 检查会员权益（原有 QuotaChecker 逻辑）
+            user = current_user.user
+            tier = quota_registry.get_tier(user.subscription_plan or "free")
+            # ... 检查存储、并发等基础权益
+            
+            # 4. 检查并扣减积分
+            points_service = get_points_service()
+            result = points_service.check_and_deduct(
+                user_id=user.id,
+                scene_key=scene_key,
+                amount=consume_amount,
+            )
+            
+            if not result.success:
+                raise HTTPException(
+                    status_code=402,  # Payment Required
+                    detail={
+                        "code": "INSUFFICIENT_POINTS",
+                        "message": f"积分不足，需要 {consume_amount} 积分，当前余额 {result.balance}",
+                        "recharge_url": "/app/points/recharge"
+                    }
+                )
+            
+            # 5. 将扣减信息注入请求上下文，供业务层使用
+            kwargs["points_deduct_id"] = result.transaction_id
+            
+            try:
+                # 6. 执行业务逻辑
+                return await func(*args, **kwargs)
+            except Exception as e:
+                # 7. 业务失败时退还积分
+                points_service.refund(
+                    user_id=user.id,
+                    transaction_id=result.transaction_id,
+                    reason=f"业务执行失败: {scene_key}"
+                )
+                raise
+        
+        return wrapper
+    return decorator
+```
+
+### 9.3 积分服务层
+
+```python
+# packages/domain/points_service.py
+
+class PointsService:
+    """积分服务 - 核心扣费逻辑"""
+    
+    def __init__(self, account_repo, transaction_repo, config_repo):
+        self.account_repo = account_repo
+        self.transaction_repo = transaction_repo
+        self.config_repo = config_repo
+    
+    def check_and_deduct(self, user_id: str, scene_key: str, amount: int, ref_id: str = "") -> DeductResult:
+        """
+        检查余额并扣减积分（事务操作）
+        
+        流程：
+        1. 查询积分账户
+        2. 检查余额是否 >= amount
+        3. 在事务中：扣减余额 + 写入流水
+        4. 返回扣减结果
+        """
+        pass
+    
+    def refund(self, user_id: str, transaction_id: str, reason: str = "") -> bool:
+        """退还积分（业务失败时调用）"""
+        pass
+    
+    def gift(self, user_id: str, amount: int, source: str, ref_id: str = ""):
+        """赠送积分（会员赠送 / 任务奖励）"""
+        pass
+    
+    def get_balance(self, user_id: str) -> int:
+        """查询余额"""
+        pass
+    
+    def get_transactions(self, user_id: str, page: int = 1, page_size: int = 20, 
+                         type: str = None, source: str = None) -> list:
+        """查询流水"""
+        pass
+```
+
+### 9.4 与现有 QuotaChecker 的集成
+
+```python
+# 改造后的检查流程
+async def check_all_quotas(user, scene_key: str, consume_amount: int):
+    """统一配额检查入口"""
+    
+    # 1. 基础配额检查（存储空间、项目数、并发数等）
+    #    复用现有 QuotaChecker
+    plan = user.subscription_plan or "free"
+    quota_results = quota_checker.check_multiple(plan, {
+        QuotaDimension.STORAGE_GB.value: get_used_storage(user.id),
+        QuotaDimension.VIDEOS_PER_MONTH.value: get_monthly_video_count(user.id),
+        QuotaDimension.MAX_CONCURRENT.value: get_concurrent_count(user.id),
+    })
+    for result in quota_results:
+        if not result.allowed:
+            raise QuotaExceededError(result)
+    
+    # 2. 积分检查
+    #    新增 PointsChecker
+    balance = points_service.get_balance(user.id)
+    if balance < consume_amount:
+        raise InsufficientPointsError(
+            required=consume_amount,
+            balance=balance
+        )
+    
+    return True
+```
+
+### 9.5 各功能模块接入方式
+
+```python
+# apps/api/app/api/routes/ai_voice.py（示例）
+@router.post("/generate")
+@require_points(scene_key="ai_voice")
+async def generate_voice(
+    request: VoiceGenerateRequest,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    points_deduct_id: str = None,  # 由中间件注入
+):
+    # 积分已扣减，直接执行业务逻辑
+    result = await voice_service.generate(request, current_user.user.id)
+    return VoiceGenerateResponse(
+        voice_id=result.id,
+        points_consumed=1,
+    )
+
+# apps/api/app/api/routes/ai_video.py（示例）
+@router.post("/generate")
+@require_points(scene_key="ai_video")
+async def generate_video(
+    request: VideoGenerateRequest,
+    current_user: AuthenticatedUser = Depends(get_current_user),
+    points_deduct_id: str = None,
+):
+    # 视频按秒计费，需要在中间件外动态计算
+    ...
+```
+
+> **注意**：对于视频生成等按量计费场景（时长不确定），中间件支持动态计算消耗量：
+> ```python
+> @require_points(scene_key="ai_video", dynamic=True)
+> # dynamic=True 时，中间件只检查余额 > 0，实际扣费由业务层调用 points_service.deduct()
+> ```
+
+---
+
+## 十、实施计划
+
+### 10.1 开发阶段拆分
+
+| 阶段 | 内容 | 预估工时 |
+|------|------|----------|
+| **P1 - 基础框架** | 数据库迁移脚本、积分账户/流水/订单 Model、Repository | 3 天 |
+| **P2 - 核心服务** | PointsService 核心逻辑（扣费/退还/赠送/查询） | 3 天 |
+| **P3 - API 层** | 积分 API 路由、中间件、与现有订阅 API 集成 | 3 天 |
+| **P4 - 前端页面** | Plans.tsx 改造、积分中心页面、Header 积分徽章、消耗提示 | 5 天 |
+| **P5 - 功能接入** | 各 AI 功能模块接入 `@require_points` 中间件 | 3 天 |
+| **P6 - 迁移与测试** | 数据迁移脚本、灰度方案、集成测试、端到端测试 | 3 天 |
+
+### 10.2 文件改动清单（预览）
+
+| 类型 | 文件路径 | 改动说明 |
+|------|----------|----------|
+| 新增 | `migrations/007_membership_points.sql` | 新建积分相关 4 张表 |
+| 新增 | `packages/domain/points_service.py` | 积分核心服务 |
+| 新增 | `packages/domain/points_models.py` 或追加到 `models.py` | 积分相关 ORM Model |
+| 新增 | `packages/adapters/sqlalchemy_impl/points_repository.py` | 积分 Repository |
+| 新增 | `apps/api/app/middleware/points_check.py` | 积分扣费中间件 |
+| 新增 | `apps/api/app/api/routes/points.py` | 积分 API 路由 |
+| 新增 | `apps/api/app/schemas/points.py` | 积分 Schema |
+| 新增 | `apps/web/src/pages/points/Center.tsx` | 积分中心页面 |
+| 新增 | `apps/web/src/pages/points/Transactions.tsx` | 积分明细页面 |
+| 新增 | `apps/web/src/pages/points/Packages.tsx` | 积分包充值页面 |
+| 新增 | `apps/web/src/components/common/PointsBadge/` | 积分徽章组件 |
+| 新增 | `apps/web/src/components/common/PointsCost/` | 消耗提示组件 |
+| 修改 | `packages/domain/quota.py` | 新增 POINTS 维度和 monthly_credits 配额 |
+| 修改 | `packages/adapters/sqlalchemy_impl/models.py` | 新增积分相关 Model |
+| 修改 | `apps/api/app/api/routes/subscription.py` | 集成积分逻辑 |
+| 修改 | `apps/api/app/schemas/subscription.py` | 返回中增加积分信息 |
+| 修改 | `apps/web/src/pages/subscription/Plans.tsx` | 对齐 4 档 + 积分展示 |
+| 修改 | 各 AI 功能路由文件 | 添加 `@require_points` 装饰器 |
+
+### 10.3 测试计划
+
+| 测试类型 | 覆盖范围 |
+|----------|----------|
+| 单元测试 | PointsService 扣费/退还/赠送逻辑、PointsChecker 余额检查 |
+| 集成测试 | API 端到端：充值→到账→消费→扣减→流水查询 |
+| 并发测试 | 同一用户多请求并发扣费的余额一致性 |
+| 前端测试 | Plans 页面渲染、积分中心交互、Header 徽章实时更新 |
+| 迁移测试 | 老用户数据迁移正确性验证 |
+
+---
+
+## 十一、风险与注意事项
+
+| 风险 | 应对方案 |
+|------|----------|
+| 并发扣费导致余额不一致 | 数据库事务 + 行锁，`points_accounts` 使用 `SELECT ... FOR UPDATE` |
+| 支付回调延迟导致积分未到账 | 订单创建后 30 分钟未支付自动关闭；回调支持幂等重试 |
+| 积分消耗规则变更影响用户 | 变更前 7 天公告通知；已购买的服务按旧价格执行 |
+| 前后端积分展示不一致 | 统一从 `GET /api/v1/points/balance` 获取，前端不本地缓存余额 |
+| 老用户迁移产生不满 | 过渡期权益不降低 + 额外赠送积分礼包 |
+
+---
+
+## 十二、开放问题（待确认）
+
+1. **积分有效期**：当前设计为永久有效。是否需要设置有效期（如 1 年）？
+2. **退款策略**：积分充值后是否支持退款？已消费的积分如何计算？
+3. **企业版定制**：企业版是否需要支持自定义积分消耗规则？
+4. **支付渠道**：第一期接入支付宝 + 微信支付，是否需要支持其他渠道？
+5. **发票需求**：积分充值是否需要单独开发票？与订阅发票合并还是分开？
+
+---
+
+*本文档为方案设计阶段产物，待确认后将按「实施计划」分阶段开发。*

infra/docker/api-base.Dockerfile

@@ -12,8 +12,9 @@ RUN sed -i 's|deb.debian.org|mirrors.aliyun.com|g' /etc/apt/sources.list.d/debia
     sed -i 's|deb.debian.org|mirrors.aliyun.com|g' /etc/apt/sources.list 2>/dev/null || true
 
 # 预装系统依赖（gcc 编译 psycopg/pg 扩展，libpq-dev 编译期，libpq5 运行期，ffmpeg 封面取帧）
-# 字体修复：fonts-noto-cjk 包的 .ttc 文件混入了 Mono 变体，导致 Bold 匹配到等宽字体
-# 解决方案：删除有问题的 .ttc，使用仓库内预下载的 Noto Sans SC Variable Font（不含 Mono）
+# 字体修复：fonts-noto-cjk 包的 Sans .ttc 文件混入了 Mono 变体，导致 Bold 匹配到等宽字体
+# 解决方案：删除 Sans .ttc，保留 Serif .ttc；仓库内预下载 VF 可变字体（不含 Mono）
+# #1896 补充开源字体：NotoSerifCJKsc-VF.otf（思源宋体）、LXGWWenKai-Regular.ttf（霞鹜文楷开源楷体）
 RUN apt-get update && apt-get install -y --no-install-recommends \
     gcc \
     libpq-dev \
@@ -21,12 +22,15 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
     ffmpeg \
     fonts-noto-cjk \
     && rm -rf /var/lib/apt/lists/* \
-    # 删除有问题的 .ttc 文件（包含 Mono 变体）
+    # 删除有问题的 Sans .ttc（含 Mono 变体），保留 Serif .ttc 作为宋体 fallback
     && rm -f /usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc \
-    && rm -f /usr/share/fonts/opentype/noto/NotoSansCJK-Bold.ttc
+    && rm -f /usr/share/fonts/opentype/noto/NotoSansCJK-Bold.ttc \
+    && mkdir -p /usr/share/fonts/truetype/lxgw
 
-# 复制预下载的 Noto Sans SC Variable Font（包含所有字重，不含 Mono 变体）
+# 复制开源中文字体：思源黑体 VF + 思源宋体 VF（#1896）+ 霞鹜文楷（#1896，开源楷体）
 COPY infra/fonts/NotoSansSC-VF.ttf /usr/share/fonts/opentype/noto/NotoSansSC-VF.ttf
+COPY infra/fonts/NotoSerifCJKsc-VF.otf /usr/share/fonts/opentype/noto/NotoSerifCJKsc-VF.otf
+COPY infra/fonts/LXGWWenKai-Regular.ttf /usr/share/fonts/truetype/lxgw/LXGWWenKai-Regular.ttf
 RUN fc-cache -fv
 
 # 创建虚拟环境

infra/docker/compose.yml

@@ -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"
@@ -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

@@ -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-base.Dockerfile

@@ -12,8 +12,11 @@ RUN sed -i 's|deb.debian.org|mirrors.aliyun.com|g' /etc/apt/sources.list.d/debia
     sed -i 's|deb.debian.org|mirrors.aliyun.com|g' /etc/apt/sources.list 2>/dev/null || true
 
 # 预装系统依赖（编译工具 + 运行时 + CJK 字体用于 ASS 字幕渲染）
-# 字体修复：fonts-noto-cjk 包的 .ttc 文件混入了 Mono 变体，导致 Bold 匹配到等宽字体
-# 解决方案：删除有问题的 .ttc，使用仓库内预下载的 Noto Sans SC Variable Font（不含 Mono）
+# 字体修复：fonts-noto-cjk 包的 Sans .ttc 文件混入了 Mono 变体，导致 Bold 匹配到等宽字体
+# 解决方案：删除 Sans .ttc，保留 Serif .ttc（宋体 fallback）；仓库内预下载 VF 可变字体（不含 Mono）
+# #1896 补充开源字体：NotoSerifCJKsc-VF.otf（思源宋体衬线）、LXGWWenKai-Regular.ttf（霞鹜文楷开源楷体，SIL OFL）
+# - 苹方/微软雅黑为 macOS/Windows 系统字体，服务器无对应文件，映射到 Noto Sans SC fallback
+# - 华康俪金黑为商业字体有版权风险，前端已移除，后端映射到 Noto Sans SC 兼容老数据
 RUN apt-get update && apt-get install -y --no-install-recommends \
     gcc \
     g++ \
@@ -23,12 +26,15 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
     libglib2.0-0 \
     fonts-noto-cjk \
     && rm -rf /var/lib/apt/lists/* \
-    # 删除有问题的 .ttc 文件（包含 Mono 变体）
+    # 删除有问题的 Sans .ttc 文件（包含 Mono 变体会导致粗体匹配错误），保留 Serif .ttc 作为宋体 fallback
     && rm -f /usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc \
-    && rm -f /usr/share/fonts/opentype/noto/NotoSansCJK-Bold.ttc
+    && rm -f /usr/share/fonts/opentype/noto/NotoSansCJK-Bold.ttc \
+    && mkdir -p /usr/share/fonts/truetype/lxgw
 
-# 复制预下载的 Noto Sans SC Variable Font（包含所有字重，不含 Mono 变体）
+# 复制开源中文字体：思源黑体 VF + 思源宋体 VF（#1896）+ 霞鹜文楷（#1896，开源楷体）
 COPY infra/fonts/NotoSansSC-VF.ttf /usr/share/fonts/opentype/noto/NotoSansSC-VF.ttf
+COPY infra/fonts/NotoSerifCJKsc-VF.otf /usr/share/fonts/opentype/noto/NotoSerifCJKsc-VF.otf
+COPY infra/fonts/LXGWWenKai-Regular.ttf /usr/share/fonts/truetype/lxgw/LXGWWenKai-Regular.ttf
 RUN fc-cache -fv
 
 # 创建虚拟环境

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

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

@@ -703,6 +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)

packages/domain/ass_subtitle_builder.py

@@ -25,13 +25,20 @@ TITLE_MARGIN_BOTTOM = 100
 TITLE_MARGIN_SIDE = 40
 
 # 字体名称映射：前端中文字体名 → 服务器实际注册名（ffmpeg/ASS 通过注册名匹配字体）
+# #1896 字体映射修复：每个字体映射到独立的注册名，而非全部回退到 Noto Sans SC
+# - 思源宋体 → Noto Serif CJK SC（fonts-noto-cjk 包预装 + VF.otf）
+# - 楷体 → LXGW WenKai（霞鹜文楷，#1896 新增 SIL OFL 开源楷体）
+# - 苹方/PingFang/微软雅黑：服务器 Linux 无对应字体，fallback 思源黑体
+# - 华康俪金黑：商业字体有版权风险，前端已移除，后端保留映射 fallback 思源黑体（兼容老数据）
 FONT_NAME_MAP: dict[str, str] = {
     "思源黑体": "Noto Sans SC",
     "思源宋体": "Noto Serif CJK SC",
     "苹方": "Noto Sans SC",
     "PingFang": "Noto Sans SC",
     "微软雅黑": "Noto Sans SC",
-    "楷体": "Noto Serif CJK SC",
+    "Microsoft YaHei": "Noto Sans SC",
+    "楷体": "LXGW WenKai",
+    "霞鹜文楷": "LXGW WenKai",
     "华康俪金黑": "Noto Sans SC",
 }
 

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 +1,206 @@
+"""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（中英文句号/问号/感叹号/分号/换行，不含逗号，与前端 sentences.ts 保持一致）
+- `_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/换行分句（与前端 sentences.ts 的 SENTENCE_SPLIT_RE 一致）.
+
+    仅在句末标点（。！？!?；;）和换行处分句，**不再用逗号（,，）切分**。
+    按逗号切分会把连贯句子拆得过碎，导致 B-roll/口播画面按句插入时句数过多、
+    时长过短，效果不符合预期（Issue #1892）。
+    """
+    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