feat(generate): 合并步骤4(选择标题)和步骤5(生成预览)为一个步骤

- STEPS 从7步缩减为6步，合并后步骤名称保持"选择标题"
- Step4TitleSettings 合并原 Step5GeneratePreview 的 TitleStylePanel
  - 标题文字输入 + AI生成标题（原Step4）
  - 位置/字号/字体/颜色/粗斜体/描边阴影/样式预设（原Step5）
- 右侧实时预览播放器从 step 4-6 调整为 step 4-5
- 下一步校验增加预览就绪检查（previewReady）
- 步骤6/7重新编号为5/6（封面/确认生成）
- 删除 Step5GeneratePreview.tsx（功能已并入 Step4）
- 更新 smoke test 移除对已删除组件的引用

alembic/versions/057_add_title_config_to_generation_tasks.py

@@ -0,0 +1,26 @@
+"""Add title_config to generation_tasks
+
+Revision ID: 057_title_config
+Revises: 056_fix_cover_templates_config
+Create Date: 2026-08-23
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "057_title_config"
+down_revision = "056_fix_cover_templates_config"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "generation_tasks",
+        sa.Column("title_config", sa.JSON(), nullable=False, server_default="{}"),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("generation_tasks", "title_config")

apps/api/app/api/routes/assets.py

@@ -46,10 +46,16 @@ def _to_asset_response(item, storage_service=None) -> AssetResponse:
             logger.warning("生成签名URL失败: storage_key=%s", item.storage_key, exc_info=True)
             file_url = None
 
-    # 缩略图：优先用已有 thumbnail_url，否则对视频素材复用文件签名 URL
-    thumbnail_url = item.thumbnail_url
-    if not thumbnail_url and item.mime_type and item.mime_type.startswith("video") and file_url:
-        thumbnail_url = file_url
+    # 缩略图：存储的是 storage_key，需要生成签名 URL 供前端使用
+    # 不再降级使用视频文件 URL（浏览器 <img> 无法渲染 .mp4，会显示黑屏）
+    thumbnail_url = None
+    if item.thumbnail_url:
+        try:
+            svc = storage_service or get_storage_service()
+            thumbnail_url = svc.get_download_url(item.thumbnail_url)
+        except Exception:
+            logger.warning("生成缩略图签名URL失败: key=%s", item.thumbnail_url, exc_info=True)
+            thumbnail_url = None
 
     return AssetResponse(
         id=item.id,

apps/api/app/api/routes/generation_cover.py

@@ -31,8 +31,6 @@ logger = logging.getLogger(__name__)
 router = APIRouter(tags=["Generation"])
 
 
-
-
 # ── Schemas ──────────────────────────────────────────────────────────────
 
 
@@ -65,6 +63,83 @@ class GenerateCoverResponse(BaseModel):
 # ── Route ────────────────────────────────────────────────────────────────
 
 
+def _persist_cover_frame(
+    frame_url: str,
+    plan_id: str,
+    title_text: str = "",
+    *,
+    title_color: str = "#ffffff",
+    title_position: str = "bottom",
+    title_font_size: int | None = None,
+) -> str:
+    """下载 MediaKit 返回的临时帧图，可选叠加标题后转存到 OSS covers/ 路径。
+
+    Args:
+        frame_url: MediaKit 返回的临时帧图 URL
+        plan_id: 剪辑计划 ID（生成 OSS key）
+        title_text: 非空时用 Pillow 在帧上叠加标题（用于 E2 从源素材抽帧，
+            因为源素材本身没有烧录标题）
+        title_color: 标题字体颜色（#RRGGBB）
+        title_position: 标题位置 top/center/bottom
+        title_font_size: 标题字号，None 时自动计算
+    """
+    import tempfile
+    import uuid
+    from pathlib import Path
+
+    tmp_path: str | None = None
+    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 tempfile.NamedTemporaryFile(suffix=".jpg", delete=False) as tmp:
+            tmp.write(resp.content)
+            tmp_path = tmp.name
+
+        # E2 从源素材抽帧时，源素材无标题，叠加标题文字
+        if title_text and title_text.strip():
+            try:
+                from packages.shared.title_overlay import apply_title_to_image
+
+                applied = apply_title_to_image(
+                    tmp_path,
+                    title_text,
+                    color=title_color,
+                    position=title_position,
+                    font_size=title_font_size,
+                )
+                if applied:
+                    logger.info("[封面生成] E2 帧图已叠加标题: plan_id=%s", plan_id)
+            except Exception:
+                logger.warning(
+                    "[封面生成] E2 标题叠加失败（返回无标题帧）: plan_id=%s",
+                    plan_id,
+                    exc_info=True,
+                )
+
+        from packages.shared.storage import get_shared_storage_service
+
+        storage = get_shared_storage_service()
+        cover_key = f"covers/{plan_id}/cover_{uuid.uuid4().hex[:8]}.jpg"
+        storage.upload_file(
+            file_or_path=tmp_path,
+            storage_key=cover_key,
+            content_type="image/jpeg",
+        )
+        public_url = storage.get_url(cover_key)
+        return public_url or frame_url
+    except Exception:
+        logger.warning("封面帧转存失败，返回原始 URL: plan_id=%s", plan_id, exc_info=True)
+        return frame_url
+    finally:
+        if tmp_path:
+            Path(tmp_path).unlink(missing_ok=True)
+
+
 @router.post("/generate-cover", response_model=GenerateCoverResponse)
 def generate_cover(
     body: GenerateCoverRequest,
@@ -198,44 +273,31 @@ def generate_cover(
                 exc_info=True,
             )
 
-    # 仍然找不到才报 400
-    if not rendered_storage_key:
-        logger.error("[封面生成] ❌ 找不到预览视频: plan_id=%s", plan_id)
-        raise HTTPException(
-            status_code=400,
-            detail="请先生成预览视频，再生成封面",
-        )
-
-    # 回写到 plan.config
-    plan_svc.update_plan_config(plan_id, {"rendered_storage_key": rendered_storage_key})
-
-    # 使用裸 URL（rendered/* 已配置公开读）
+    # 使用裸 URL（rendered/* 已配置公开读）；找不到渲染视频时不立即报错，
+    # 因为步骤 E 可以直接从源素材抽帧（历史数据或 Worker 抽帧失败时的兜底）
     primary_video_url = None
-    try:
-        if rendered_storage_key.startswith("http"):
-            primary_video_url = rendered_storage_key
-        else:
-            from packages.shared.storage import get_shared_storage_service
+    if rendered_storage_key:
+        plan_svc.update_plan_config(plan_id, {"rendered_storage_key": rendered_storage_key})
+        try:
+            if rendered_storage_key.startswith("http"):
+                primary_video_url = rendered_storage_key
+            else:
+                from packages.shared.storage import get_shared_storage_service
 
-            storage_svc = get_shared_storage_service()
-            primary_video_url = storage_svc.get_url(rendered_storage_key)
-        # 防御性规范化：合并路径中的双斜杠（// -> /），但保留协议头的 ://
-        # 历史数据中 project_id 为空时会产生 projects//tasks/ 路径，
-        # MediaKit 的 HTTP 客户端会规范化 URL 导致 404
-        if primary_video_url:
-            import re as _re
+                storage_svc = get_shared_storage_service()
+                primary_video_url = storage_svc.get_url(rendered_storage_key)
+            if primary_video_url:
+                import re as _re
 
-            primary_video_url = _re.sub(r"(?<!:)//", "/", primary_video_url)
-        logger.info(
-            "获取预览视频URL用于封面生成: plan_id=%s url=%s",
-            plan_id,
-            primary_video_url[:80] if primary_video_url else "",
-        )
-    except Exception as e:
-        raise HTTPException(
-            status_code=500,
-            detail=f"获取预览视频URL失败: {e}",
-        ) from e
+                primary_video_url = _re.sub(r"(?<!:)//", "/", primary_video_url)
+            logger.info(
+                "获取预览视频URL用于封面生成: plan_id=%s url=%s",
+                plan_id,
+                primary_video_url[:80] if primary_video_url else "",
+            )
+        except Exception as e:
+            logger.warning("获取预览视频URL失败: plan_id=%s err=%s", plan_id, e)
+            primary_video_url = None
 
     # 统一封面管道：优先从 GenerationTask.cover_url 读取渲染后视频抽帧的封面
     # 多步查找 cover_url，和查找视频 URL 一样的 fallback 逻辑
@@ -310,6 +372,129 @@ def generate_cover(
                     exc_info=True,
                 )
 
+        # 步骤 D：从 plan.config.cover_candidates 读取（Worker 渲染时写入）
+        if not cover_url_from_task:
+            _candidates = (plan.config or {}).get("cover_candidates") or []
+            if isinstance(_candidates, list) and _candidates:
+                _first = _candidates[0]
+                if isinstance(_first, dict):
+                    cover_url_from_task = _first.get("image_url") or _first.get("url") or ""
+                    if cover_url_from_task:
+                        logger.info(
+                            "[封面生成] 统一管道封面(步骤D-cover_candidates): plan_id=%s url=%s",
+                            plan_id,
+                            cover_url_from_task[:80],
+                        )
+
+        # 步骤 E1：如果有已渲染的预览视频 URL 但 cover_url 未持久化（历史数据），
+        # 直接从渲染视频抽帧
+        if not cover_url_from_task and primary_video_url:
+            try:
+                from packages.shared.mediakit_client import get_mediakit_client
+
+                mk_client = get_mediakit_client()
+                if mk_client.is_available:
+                    logger.info(
+                        "[封面生成] 步骤E1-从渲染视频抽帧: plan_id=%s url=%s",
+                        plan_id,
+                        primary_video_url[:80],
+                    )
+                    snapshots = mk_client.extract_frames(
+                        video_url=primary_video_url,
+                        strategy="SpecifiedFrames",
+                        max_frames=1,
+                        poll_interval=2.0,
+                        max_poll_attempts=5,
+                        max_retries=0,
+                    )
+                    if snapshots:
+                        raw = snapshots[0].get("image_url") or snapshots[0].get("url") or ""
+                        if raw:
+                            cover_url_from_task = _persist_cover_frame(raw, plan_id)
+                            logger.info(
+                                "[封面生成] 统一管道封面(步骤E1-rendered-video): plan_id=%s url=%s",
+                                plan_id,
+                                cover_url_from_task[:80],
+                            )
+            except Exception:
+                logger.warning(
+                    "[封面生成] 步骤E1从渲染视频抽帧失败: plan_id=%s",
+                    plan_id,
+                    exc_info=True,
+                )
+
+        # 步骤 E2：当 A/B/C/D/E1 均未命中（如历史预览任务无 cover_url）时，
+        # 直接从用户选择的第一个视频素材中抽取封面帧作为兜底。API 请求内短超时，不阻塞。
+        if not cover_url_from_task and body.asset_ids:
+            from packages.adapters.sqlalchemy_impl.asset_repository import (
+                SQLAlchemyAssetRepository,
+            )
+            from packages.shared.mediakit_client import get_mediakit_client
+            from packages.shared.storage import get_shared_storage_service
+
+            asset_repo = SQLAlchemyAssetRepository(db)
+            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）
+            _e2_title_cfg = (plan.config or {}).get("title", {}) or {}
+            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 ""
+            # 读取标题样式：前端可能传 color 或 font_color，都兼容
+            _e2_title_color = _e2_title_cfg.get("color") or _e2_title_cfg.get("font_color") or "#ffffff"
+            _e2_title_position = _e2_title_cfg.get("position", "bottom") or "bottom"
+            _e2_title_font_size = _e2_title_cfg.get("font_size") or _e2_title_cfg.get("size")
+            if mk_client.is_available:
+                for aid in body.asset_ids:
+                    try:
+                        asset = asset_repo.get(aid)
+                        if not asset or asset.file_type != "video":
+                            continue
+                        sk = asset.storage_key or ""
+                        if not sk:
+                            continue
+                        src_url = sk if sk.startswith("http") else storage_svc.get_url(sk)
+                        if not src_url:
+                            continue
+                        logger.info(
+                            "[封面生成] 步骤E-从素材抽帧: plan_id=%s asset_id=%s url=%s",
+                            plan_id,
+                            aid,
+                            src_url[:80],
+                        )
+                        snapshots = mk_client.extract_frames(
+                            video_url=src_url,
+                            strategy="SpecifiedFrames",
+                            max_frames=1,
+                            poll_interval=2.0,
+                            max_poll_attempts=5,
+                            max_retries=0,
+                        )
+                        if snapshots:
+                            raw = snapshots[0].get("image_url") or snapshots[0].get("url") or ""
+                            if raw:
+                                cover_url_from_task = _persist_cover_frame(
+                                    raw,
+                                    plan_id,
+                                    title_text=_e2_title_text,
+                                    title_color=_e2_title_color,
+                                    title_position=_e2_title_position,
+                                    title_font_size=_e2_title_font_size,
+                                )
+                                logger.info(
+                                    "[封面生成] 统一管道封面(步骤E-source-asset): plan_id=%s url=%s",
+                                    plan_id,
+                                    cover_url_from_task[:80],
+                                )
+                                break
+                    except Exception:
+                        logger.warning(
+                            "[封面生成] 步骤E从素材抽帧失败: plan_id=%s asset_id=%s",
+                            plan_id,
+                            aid,
+                            exc_info=True,
+                        )
+
         if cover_url_from_task:
             # 标题已在预览视频渲染时烧录（ASS字幕），封面帧自然包含标题
             cover_data = {
@@ -325,13 +510,13 @@ def generate_cover(
             return GenerateCoverResponse(plan_id=plan_id, cover=cover_data)
 
         logger.warning(
-            "[封面生成] 统一管道未找到 cover_url: plan_id=%s",
+            "[封面生成] 统一管道未找到 cover_url (A/B/C/D均未命中): plan_id=%s",
             plan_id,
         )
         # ai_frame/ai_regenerate 类型必须从渲染管道获取，不再回退到 AI 服务
         raise HTTPException(
             status_code=400,
-            detail="封面尚未生成，请先重新生成预览视频以触发封面自动提取",
+            detail="封面生成失败：未找到可抽帧的视频素材，请确认已上传视频素材后重试",
         )
 
     from packages.shared.ai_service import run_generate_cover

apps/api/app/api/routes/generation_preview.py

@@ -5,7 +5,6 @@
 
 from __future__ import annotations
 
-import json
 import logging
 
 from app.auth import AuthenticatedUser, get_current_user
@@ -265,22 +264,38 @@ def create_preview_generation_task(
     if not video_ratio and request.template_id:
         video_ratio = _infer_video_ratio_from_template(request.template_id, db, user_id)
 
+    # 根据 video_ratio 计算输出分辨率（默认竖屏 1080x1920）
+    output_width, output_height = 1080, 1920
+    if video_ratio:
+        parts = video_ratio.split(":")
+        if len(parts) == 2:
+            try:
+                w, h = int(parts[0]), int(parts[1])
+                base = 1920
+                if w < h:
+                    # 竖屏
+                    output_width = round(base * w / h)
+                    output_height = base
+                else:
+                    # 横屏
+                    output_width = base
+                    output_height = round(base * h / w)
+                # 对齐到偶数
+                output_width = output_width - output_width % 2
+                output_height = output_height - output_height % 2
+            except (ValueError, ZeroDivisionError):
+                output_width, output_height = 1080, 1920
+    resolution = f"{output_width}x{output_height}"
+
+    logger.info(
+        "[预览生成] 分辨率: video_ratio=%s → %s (%dx%d)",
+        video_ratio, resolution, output_width, output_height,
+    )
+
     # 从模板读取 editing_mode / mode 作为 strategy_id（渲染 pipeline 的 mode 参数）
     strategy_id = _resolve_strategy_id_from_template(request.template_id, db, user_id)
 
-    # 处理标题配置：如果有标题文本，序列化到 custom_title 字段传递给 worker
     title_config = request.title_config or {}
-    title_text = (title_config.get("text") or "").strip()
-    custom_title_value = ""
-    if title_text:
-        # 将标题文本和样式配置序列化为 JSON 存入 custom_title
-        # Worker 端会解析 JSON 获取完整标题配置
-        custom_title_value = json.dumps(title_config, ensure_ascii=False)
-        logger.info(
-            "[预览生成] 标题配置: text=%s, config_keys=%s",
-            title_text[:30],
-            list(title_config.keys()),
-        )
 
     use_case = CreateGenerationTaskUseCase(generation_task_repository)
 
@@ -300,12 +315,14 @@ def create_preview_generation_task(
                 asset_select_mode="",
                 batch_id="",
                 video_title=request.video_title,
-                resolution="",
+                resolution=resolution,
                 bgm_config=request.bgm_config or {},
                 auto_retry_enabled=False,
                 auto_retry_max=0,
                 is_preview=True,
-                custom_title=custom_title_value,
+                title_config=title_config,
+                output_width=output_width,
+                output_height=output_height,
             )
         )
     except ValueError as e:

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

@@ -16,6 +16,7 @@ from app.core.task_enqueue import (
 from app.dependencies import (
     get_asset_library_repository,
     get_asset_repository,
+    get_db_session,
     get_generated_video_repository,
     get_generation_task_repository,
     get_project_repository,
@@ -32,6 +33,7 @@ from app.schemas.generation_task import (
     ListGenerationTasksResponse,
 )
 from fastapi import APIRouter, Depends, HTTPException
+from sqlalchemy.orm import Session
 
 from packages.application import (
     CreateGenerationTaskCommand,
@@ -68,7 +70,7 @@ def _to_generation_task_response(task) -> GenerationTaskResponse:
         output_width=getattr(task, "output_width", 1280),
         output_height=getattr(task, "output_height", 720),
         cover_url=getattr(task, "cover_url", ""),
-        custom_title=getattr(task, "custom_title", ""),
+        title_config=getattr(task, "title_config", {}) or {},
         logs=getattr(task, "logs", "[]"),
         status=task.status,
         progress=task.progress,
@@ -142,6 +144,68 @@ def _select_assets_from_library(
     return [a.id for a in ready_video_assets]
 
 
+
+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("[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
+
+
 def _resolve_project_and_library(
     request: CreateGenerationTaskRequest,
     project_repository: Any,
@@ -187,6 +251,7 @@ def create_generation_task(
     project_repository: Any = Depends(get_project_repository),
     asset_library_repository: Any = Depends(get_asset_library_repository),
     asset_repository: Any = Depends(get_asset_repository),
+    db: Session = Depends(get_db_session),
 ) -> BatchGenerationTaskResponse:
     logger.info(
         "[生成任务] 接收请求: user_id=%s, template_id=%s, asset_count=%d, mode=%s, count=%d",
@@ -242,6 +307,89 @@ def create_generation_task(
                     detail="当前项目没有符合条件的视频素材，请先上传并等待导入完成后再生成。",
                 )
 
+    # ── 兜底复用预览产物 ──
+    # 前端刷新后 previewTaskId 丢失，降级调 create 接口时，
+    # 如果同一 edit_plan 有已完成的预览任务，直接复用（秒出）。
+    if request.source_edit_plan_id and not request.is_preview:
+        try:
+            from packages.adapters.sqlalchemy_impl.models import (
+                GenerationTaskModel,
+            )
+
+            _preview_model = (
+                db.query(GenerationTaskModel)
+                .filter(
+                    GenerationTaskModel.source_edit_plan_id == request.source_edit_plan_id,
+                    GenerationTaskModel.is_preview.is_(True),
+                    GenerationTaskModel.status == "completed",
+                    GenerationTaskModel.created_by_user_id == authenticated_user.user.id,
+                )
+                .order_by(GenerationTaskModel.created_at.desc())
+                .first()
+            )
+            if _preview_model is not None:
+                # 校验分辨率一致性（与 confirm 端点逻辑相同）
+                req_w = request.output_width or 0
+                req_h = request.output_height or 0
+                src_w = getattr(_preview_model, "output_width", 0) or 0
+                src_h = getattr(_preview_model, "output_height", 0) or 0
+                resolution_match = (req_w == 0 or req_w == src_w) and (req_h == 0 or req_h == src_h)
+
+                if resolution_match:
+                    from packages.adapters.sqlalchemy_impl.generation_task_repository import (
+                        _to_domain,
+                    )
+
+                    preview_task = _to_domain(_preview_model)
+
+                    # 如果传了标题，更新 title_config
+                    fallback_title_config = None
+                    if request.title_config and request.title_config.get("text", "").strip():
+                        fallback_title_config = dict(preview_task.title_config or {})
+                        fallback_title_config.update(request.title_config)
+
+                    preview_task.mark_confirmed(
+                        cover_url=request.cover_url or preview_task.cover_url,
+                        output_width=request.output_width or preview_task.output_width,
+                        output_height=request.output_height or preview_task.output_height,
+                        title_config=fallback_title_config,
+                    )
+                    generation_task_repository.update(preview_task)
+
+                    # 同步标题到 EditPlan.config
+                    if fallback_title_config:
+                        _writeback_edit_plan_config(
+                            plan_id=request.source_edit_plan_id,
+                            task_id=preview_task.id,
+                            title_config=fallback_title_config,
+                            db=db,
+                        )
+
+                    logger.info(
+                        "[生成任务] 兜底复用预览产物: preview_task_id=%s, plan_id=%s",
+                        preview_task.id,
+                        request.source_edit_plan_id,
+                    )
+                    return BatchGenerationTaskResponse(
+                        items=[_to_generation_task_response(preview_task)],
+                        total=1,
+                    )
+                else:
+                    logger.info(
+                        "[生成任务] 兜底复用跳过（分辨率不一致）: plan_id=%s, src=%sx%s, req=%sx%s",
+                        request.source_edit_plan_id,
+                        src_w,
+                        src_h,
+                        req_w,
+                        req_h,
+                    )
+        except Exception:
+            logger.warning(
+                "[生成任务] 兜底复用预览产物异常（不影响主流程）: plan_id=%s",
+                request.source_edit_plan_id,
+                exc_info=True,
+            )
+
     use_case = CreateGenerationTaskUseCase(generation_task_repository)
     count = request.count
     created_tasks = []
@@ -303,10 +451,53 @@ def create_generation_task(
                     output_width=request.output_width,
                     output_height=request.output_height,
                     cover_url=request.cover_url,
-                    custom_title=request.custom_title,
+                    title_config=request.title_config or {},
                 )
             )
             try:
+                # 兜底关联编辑计划：前端未传 source_edit_plan_id 时，
+                # 通过 template_id + user_id 在 DB 层直接查找最新的 plan。
+                # 必须在 enqueue 之前执行，避免 worker 读取时 source_edit_plan_id 为空（竞态条件）
+                if not task.source_edit_plan_id and request.template_id:
+                    try:
+                        from packages.adapters.sqlalchemy_impl.models import EditPlanModel
+
+                        _plan_model = (
+                            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 _plan_model:
+                            task.source_edit_plan_id = _plan_model.id
+                            generation_task_repository.update(task)
+                            logger.info(
+                                "[生成任务] 自动关联编辑计划: task_id=%s plan_id=%s",
+                                task.id,
+                                _plan_model.id,
+                            )
+                    except Exception:
+                        logger.warning(
+                            "[生成任务] 查找关联编辑计划失败（不影响主流程）: task_id=%s",
+                            task.id,
+                            exc_info=True,
+                        )
+
+                # 回写 plan.config：必须在 enqueue 之前执行，
+                # 确保 worker 读取 plan 时 config 中已包含 generation_task_id。
+                # 只在首个任务时回写一次，避免批量生成时循环覆盖。
+                _effective_plan_id = task.source_edit_plan_id
+                if _effective_plan_id and len(created_tasks) == 0:
+                    _writeback_edit_plan_config(
+                        plan_id=_effective_plan_id,
+                        task_id=task.id,
+                        title_config=request.title_config,
+                        db=db,
+                    )
+
                 if safe_enqueue_generation_task(
                     task,
                     generation_task_repository,
@@ -351,6 +542,7 @@ def confirm_generation(
     authenticated_user: AuthenticatedUser = Depends(get_current_user),
     generation_task_repository: Any = Depends(get_generation_task_repository),
     project_repository: Any = Depends(get_project_repository),
+    db: Session = Depends(get_db_session),
 ) -> BatchGenerationTaskResponse:
     """确认生成 -- 复用预览渲染产物（预览与正式品质一致）。
 
@@ -379,13 +571,29 @@ def confirm_generation(
         resolution_match = (req_w == 0 or req_w == src_w) and (req_h == 0 or req_h == src_h)
 
         if resolution_match:
+            # 如果用户传了 custom_title，同步更新 title_config
+            confirmed_title_config = None
+            if request.custom_title and request.custom_title.strip():
+                confirmed_title_config = dict(getattr(source_task, "title_config", {}) or {})
+                confirmed_title_config["text"] = request.custom_title.strip()
+
             source_task.mark_confirmed(
                 cover_url=request.cover_url,
-                custom_title=request.custom_title,
                 output_width=request.output_width,
                 output_height=request.output_height,
+                title_config=confirmed_title_config,
             )
             generation_task_repository.update(source_task)
+
+            # 同步标题到 EditPlan.config
+            if confirmed_title_config and source_task.source_edit_plan_id:
+                _writeback_edit_plan_config(
+                    plan_id=source_task.source_edit_plan_id,
+                    task_id=source_task.id,
+                    title_config=confirmed_title_config,
+                    db=db,
+                )
+
             logger.info(
                 "[确认生成] 复用预览产物: task_id=%s, user_id=%s",
                 task_id,
@@ -427,7 +635,6 @@ def confirm_generation(
             output_width=request.output_width,
             output_height=request.output_height,
             cover_url=request.cover_url,
-            custom_title=request.custom_title,
         )
     )
 
@@ -560,7 +767,6 @@ def retry_generation_task(
             output_width=getattr(task, "output_width", 1280),
             output_height=getattr(task, "output_height", 720),
             cover_url=getattr(task, "cover_url", ""),
-            custom_title=getattr(task, "custom_title", ""),
         )
     )
     try:

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

@@ -1,10 +1,9 @@
 """模板编辑器 API 路由包.
 
-将原来 2560 行的 templates_editor.py 巨无霸拆分为 12 个模块：
+模块拆分：
 - schemas.py:        所有 Pydantic model
 - dependencies.py:   依赖注入
 - _utils.py:         工具函数
-- _fallback.py:      自动兜底逻辑
 - draft.py:          草稿管理（详情/更新/发布/版本/回滚）
 - clips.py:          片段管理（CRUD/分割/合并/重排/批量删除/从素材创建）
 - adjustments.py:    片段调整（速度/音量/裁剪/批量调速）
@@ -13,7 +12,6 @@
 - export.py:         导出配置
 - subtitles.py:      字幕管理
 - ai_features.py:    AI 推荐
-- generation.py:     生成（触发/进度/记录）
 - timeline.py:       时间线
 
 挂载路径: /api/v1/templates/{template_id}/editor/
@@ -34,7 +32,6 @@ from .dependencies import get_draft_plan_id, get_editor_services  # noqa: F401
 from .draft import router as draft_router
 from .effects import router as effects_router
 from .export import router as export_router
-from .generation import router as generation_router
 from .subtitles import router as subtitles_router
 from .timeline import router as timeline_router
 
@@ -51,7 +48,6 @@ _sub_routers = [
     export_router,
     subtitles_router,
     ai_features_router,
-    generation_router,
     timeline_router,
 ]
 

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

@@ -1,227 +0,0 @@
-"""模板编辑器自动兜底逻辑.
-
-generate_editor_draft 触发生成前的自动修复流程：
-1. draft → editing 状态迁移
-2. 无片段时从模板复制片段配置
-3. 为无素材片段分配指定素材
-4. 项目有素材库时自动选素材
-"""
-
-from __future__ import annotations
-
-import logging
-import random
-from typing import Any
-
-from app.services.edit_plan_service import EditPlanService
-from sqlalchemy.orm import Session
-
-from packages.adapters.sqlalchemy_impl.template_clip_config_repository import (
-    SQLAlchemyTemplateClipConfigRepository,
-)
-from packages.adapters.sqlalchemy_impl.template_repository import (
-    SQLAlchemyTemplateRepository,
-)
-from packages.domain.edit_plan import EditPlanStatus
-
-logger = logging.getLogger(__name__)
-
-
-def _auto_fallback_draft_to_editing(svc: EditPlanService, plan_id: str, plan_check) -> None:
-    """自动兜底 1: draft → editing"""
-    if plan_check.status == EditPlanStatus.DRAFT:
-        logger.info("模板编辑器自动兜底: plan=%s draft→editing", plan_id)
-        svc.transition_status(plan_id, EditPlanStatus.EDITING)
-
-
-def _auto_fallback_copy_template_clips(svc: EditPlanService, plan_id: str, plan_check, db: Session) -> None:
-    """自动兜底 2: 无片段 + 有 template_id → 从模板复制片段配置"""
-    existing_clips = svc.count_clips(plan_id)
-    if existing_clips == 0 and plan_check.template_id:
-        logger.info(
-            "模板编辑器自动兜底: plan=%s 无片段，从模板 %s 复制片段配置",
-            plan_id,
-            plan_check.template_id,
-        )
-        clip_config_repo = SQLAlchemyTemplateClipConfigRepository(db)
-        configs = clip_config_repo.list_by_template(plan_check.template_id)
-        if configs:
-            for cfg in configs:
-                svc.create_clip(
-                    plan_id=plan_id,
-                    clip_type=cfg.clip_type.value if hasattr(cfg.clip_type, "value") else cfg.clip_type,
-                    order=cfg.order,
-                    template_clip_config_id=cfg.id,
-                    duration=cfg.default_duration,
-                    transition_effect=(
-                        cfg.transition_effect.value
-                        if hasattr(cfg.transition_effect, "value")
-                        else cfg.transition_effect
-                    ),
-                )
-            logger.info(
-                "模板编辑器自动兜底: plan=%s 从 template_clip_configs 复制了 %d 个片段",
-                plan_id,
-                len(configs),
-            )
-        else:
-            tpl_repo = SQLAlchemyTemplateRepository(db)
-            segments = tpl_repo.list_segments(plan_check.template_id)
-            for seg in segments:
-                avg_duration = (seg.duration_min + seg.duration_max) / 2
-                svc.create_clip(
-                    plan_id=plan_id,
-                    clip_type="main",
-                    order=seg.segment_order,
-                    duration=avg_duration,
-                    config={
-                        "material_type": seg.material_type or "",
-                        "template_segment_id": seg.id,
-                    },
-                )
-            logger.info(
-                "模板编辑器自动兜底: plan=%s 从旧模板 segments 复制了 %d 个片段",
-                plan_id,
-                len(segments),
-            )
-
-
-def _auto_fallback_assign_assets(svc: EditPlanService, plan_id: str, plan_check) -> list:
-    """自动兜底 3: 为没有素材的片段分配素材。返回剩余无素材片段列表。"""
-    all_clips = svc.list_clips(plan_id)
-    clips_without_asset = [c for c in all_clips if not c.asset_id]
-    config_asset_ids = (plan_check.config or {}).get("asset_ids", [])
-
-    logger.info(
-        "模板编辑器自动兜底3 诊断: plan=%s total_clips=%d " "clips_without_asset=%d config_asset_ids=%r",
-        plan_id,
-        len(all_clips),
-        len(clips_without_asset),
-        config_asset_ids[:5] if config_asset_ids else [],
-    )
-
-    if clips_without_asset and config_asset_ids:
-        logger.info(
-            "模板编辑器自动兜底3: plan=%s 为 %d 个无素材片段分配 %d 个指定素材",
-            plan_id,
-            len(clips_without_asset),
-            len(config_asset_ids),
-        )
-        assigned = 0
-        for i, clip in enumerate(clips_without_asset):
-            asset_idx = i % len(config_asset_ids)
-            try:
-                svc.assign_asset(clip.id, config_asset_ids[asset_idx])
-                assigned += 1
-            except Exception as exc:
-                logger.error(
-                    "模板编辑器自动兜底3: plan=%s clip=%s 分配素材 %s 失败: %s",
-                    plan_id,
-                    clip.id,
-                    config_asset_ids[asset_idx],
-                    exc,
-                )
-        logger.info(
-            "模板编辑器自动兜底3: plan=%s 素材分配完成 assigned=%d/%d",
-            plan_id,
-            assigned,
-            len(clips_without_asset),
-        )
-        # 重新检查剩余无素材片段
-        all_clips_after = svc.list_clips(plan_id)
-        clips_without_asset = [c for c in all_clips_after if not c.asset_id]
-        if clips_without_asset:
-            logger.warning(
-                "模板编辑器自动兜底3: plan=%s 仍有 %d 个片段无素材",
-                plan_id,
-                len(clips_without_asset),
-            )
-    elif not clips_without_asset:
-        logger.info("模板编辑器自动兜底3: plan=%s 所有片段已有素材，跳过", plan_id)
-    elif not config_asset_ids:
-        logger.info(
-            "模板编辑器自动兜底3: plan=%s config.asset_ids 为空，跳过分配",
-            plan_id,
-        )
-
-    return clips_without_asset
-
-
-def _auto_fallback_auto_material_mode(
-    svc: EditPlanService,
-    plan_id: str,
-    plan_check,
-    clips_without_asset: list,
-    asset_library_repo: Any,
-    asset_repo: Any,
-    user_id: str = "",
-) -> None:
-    """自动兜底 4: 自动选素材分配给无素材片段
-
-    查找策略（按优先级）：
-    1. plan 有 project_id → 从项目素材库查找
-    2. plan 无 project_id 但有 user_id → 从用户上传的素材中查找
-    """
-    if not clips_without_asset:
-        return
-
-    ready_videos: list = []
-    source_desc = ""
-
-    # 策略 1: 通过 project_id 查找项目素材库
-    if plan_check.project_id:
-        libs = asset_library_repo.find_by_project(plan_check.project_id)
-        video_lib = None
-        for lib in libs:
-            lib_kind = lib.kind.value if hasattr(lib.kind, "value") else lib.kind
-            if lib_kind == "video":
-                video_lib = lib
-                break
-        if video_lib:
-            assets = asset_repo.find_by_library(video_lib.id)
-            ready_videos = [
-                a
-                for a in assets
-                if (a.status.value if hasattr(a.status, "value") else a.status) == "ready"
-                and a.mime_type
-                and a.mime_type.startswith("video")
-            ]
-            source_desc = f"素材库 {video_lib.name}"
-
-    # 策略 2: 通过 user_id 查找用户上传的素材
-    if not ready_videos and user_id and hasattr(asset_repo, "find_ready_videos_by_user"):
-        logger.info(
-            "模板编辑器自动兜底4: plan=%s project_id 为空，尝试通过 user_id=%s 查找素材",
-            plan_id,
-            user_id,
-        )
-        ready_videos = asset_repo.find_ready_videos_by_user(user_id)
-        source_desc = f"用户上传 (user_id={user_id[:8]}...)"
-
-    if not ready_videos:
-        logger.warning(
-            "模板编辑器自动兜底4: plan=%s 未找到可用素材 (project_id=%s, user_id=%s)",
-            plan_id,
-            plan_check.project_id or "(empty)",
-            user_id[:8] + "..." if user_id else "(empty)",
-        )
-        return
-
-    logger.info(
-        "模板编辑器自动兜底4: plan=%s 自动选素材分配给 %d 个无素材片段 (来源: %s, 共 %d 个)",
-        plan_id,
-        len(clips_without_asset),
-        source_desc,
-        len(ready_videos),
-    )
-    random.shuffle(ready_videos)
-    for i, clip in enumerate(clips_without_asset):
-        asset = ready_videos[i % len(ready_videos)]
-        svc.assign_asset(clip.id, asset.id)
-    logger.info(
-        "模板编辑器自动兜底4: plan=%s 从 %s 分配了 %d 个素材给 %d 个片段",
-        plan_id,
-        source_desc,
-        len(ready_videos),
-        len(clips_without_asset),
-    )

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

@@ -15,16 +15,25 @@
 
 from __future__ import annotations
 
+import json
 import logging
+import random
+import re
 
 from app.auth import AuthenticatedUser, get_current_user
 from app.core.storage import get_storage_service
-from app.dependencies import get_asset_repository
+from app.dependencies import get_asset_repository, get_db_session
 from app.services.edit_plan_service import EditPlanService
 from app.services.edit_template_service import EditTemplateService
-from fastapi import APIRouter, Depends, HTTPException, Query, status
+from fastapi import APIRouter, BackgroundTasks, Depends, HTTPException, Query, status
+from sqlalchemy.orm import Session
 
 from packages.adapters.sqlalchemy_impl.asset_repository import SQLAlchemyAssetRepository
+from packages.adapters.sqlalchemy_impl.template_repository import (
+    SQLAlchemyTemplateRepository,
+)
+from packages.domain.plan_generator_utils import _calc_random_start_time
+from packages.shared.mediakit_client import get_mediakit_client
 
 from .dependencies import get_draft_plan_id, get_editor_services
 from .schemas import (
@@ -45,6 +54,9 @@ from .schemas import (
 logger = logging.getLogger(__name__)
 router = APIRouter(tags=["Template Editor"])
 
+# 编辑器默认片段时长（秒）
+_DEFAULT_EDITOR_CLIP_DURATION = 5.0
+
 
 def _clip_to_response(clip, asset_url: str | None = None) -> EditorClipResponse:
     """统一构造片段响应 — 与 edit_plan_clips 表字段完全对齐"""
@@ -155,10 +167,7 @@ def list_draft_clips(
     url_map = _build_asset_url_map(asset_ids, asset_repo)
 
     return EditorClipListResponse(
-        items=[
-            _clip_to_response(c, asset_url=url_map.get(getattr(c, "asset_id", "") or ""))
-            for c in clips
-        ],
+        items=[_clip_to_response(c, asset_url=url_map.get(getattr(c, "asset_id", "") or "")) for c in clips],
         total=total,
     )
 
@@ -272,9 +281,7 @@ def split_draft_clip(
     try:
         result = plan_svc.split_clip(clip_id, body.split_time)
     except ValueError as exc:
-        raise HTTPException(
-            status_code=status.HTTP_400_BAD_REQUEST, detail=str(exc)
-        ) from exc
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=str(exc)) from exc
     left = result["left_clip"]
     right = result["right_clip"]
     asset_ids = [getattr(left, "asset_id", "") or "", getattr(right, "asset_id", "") or ""]
@@ -304,9 +311,7 @@ def merge_draft_clips(
     try:
         merged = plan_svc.merge_clips(body.clip_ids)
     except ValueError as exc:
-        raise HTTPException(
-            status_code=status.HTTP_400_BAD_REQUEST, detail=str(exc)
-        ) from exc
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=str(exc)) from exc
     asset_id = getattr(merged, "asset_id", "") or ""
     url_map = _build_asset_url_map([asset_id], asset_repo) if asset_id else {}
     return {
@@ -354,40 +359,484 @@ def batch_delete_editor_clips(
     return ClipBatchDeleteResponse(deleted_count=deleted, plan_id=plan_id)
 
 
+
+def _safe_segment_duration(value, default: float) -> float:
+    """安全地将数据库中的时长值转换为正浮点数.
+
+    处理 None、无效类型、负数、NaN 等异常情况。
+    """
+    if value is None:
+        return default
+    try:
+        result = float(value)
+    except (ValueError, TypeError):
+        return default
+    if result != result or result <= 0:  # NaN check or non-positive
+        return default
+    return result
+
+
+def _get_template_segments(
+    template_id: str,
+    tpl_svc: EditTemplateService,
+    db: Session,
+) -> list[tuple[int, float, float]]:
+    """获取模板的片段配置（顺序、最短时长、最长时长）.
+
+    优先从新模板系统（template_clip_configs）查询，
+    若不存在则回退到旧模板系统（template_segments）。
+
+    Returns:
+        [(segment_order, duration_min, duration_max), ...] 按 order 排序
+    """
+    # 优先查新模板系统
+    try:
+        clip_configs = tpl_svc.list_clip_configs(template_id)
+        if clip_configs:
+            result = []
+            for cc in clip_configs:
+                dur_min = _safe_segment_duration(
+                    cc.min_duration, _DEFAULT_EDITOR_CLIP_DURATION
+                )
+                dur_max = _safe_segment_duration(
+                    cc.max_duration or cc.min_duration,
+                    _DEFAULT_EDITOR_CLIP_DURATION,
+                )
+                dur_min, dur_max = min(dur_min, dur_max), max(dur_min, dur_max)
+                result.append((cc.order, dur_min, dur_max))
+            return sorted(result, key=lambda x: x[0])
+    except Exception:
+        logger.warning("新模板系统查询clip_configs失败，回退到旧系统", exc_info=True)
+
+    # 回退到旧模板系统（template_segments表）
+    try:
+        old_repo = SQLAlchemyTemplateRepository(db)
+        segments = old_repo.list_segments(template_id)
+        if segments:
+            result = []
+            for s in segments:
+                dur_min = _safe_segment_duration(s.duration_min, _DEFAULT_EDITOR_CLIP_DURATION)
+                dur_max = _safe_segment_duration(s.duration_max, _DEFAULT_EDITOR_CLIP_DURATION)
+                dur_min, dur_max = min(dur_min, dur_max), max(dur_min, dur_max)
+                result.append((s.segment_order, dur_min, dur_max))
+            return sorted(result, key=lambda x: x[0])
+    except Exception:
+        logger.warning("旧模板系统查询segments失败", exc_info=True)
+
+    return []
+
+
+def _recommended_time_conflicts(
+    start: float,
+    duration: float,
+    used: list[tuple[float, float]],
+) -> bool:
+    """检查推荐起始时间是否与已使用时间段冲突."""
+    end = start + duration
+    for used_start, used_end in used:
+        if start < used_end and end > used_start:
+            return True
+    return False
+
+
+def _get_mediakit_recommendations(
+    asset_ids: list[str],
+    asset_repo,
+) -> dict[str, float]:
+    """调用 MediaKit 视频理解，获取智能选片推荐起始时间.
+
+    尝试让 MediaKit 分析视频内容，返回每个素材的推荐起始时间。
+    任何异常都优雅降级，返回空字典（调用方降级到随机选择）。
+    """
+    try:
+        client = get_mediakit_client()
+        if not client.is_available:
+            logger.info("MediaKit 未配置，使用随机起始时间")
+            return {}
+
+        storage = get_storage_service()
+
+        video_urls: list[str] = []
+        valid_asset_ids: list[str] = []
+        for asset_id in asset_ids[:10]:
+            asset = asset_repo.get(asset_id)
+            if not asset or not getattr(asset, "storage_key", None):
+                continue
+            mime = getattr(asset, "mime_type", "")
+            if not mime.startswith("video/"):
+                continue
+            try:
+                url = storage.get_download_url(asset.storage_key)
+                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)
+
+        if not video_urls:
+            return {}
+
+        prompt = (
+            "请分析每段视频，找出最精彩的5秒片段应该从哪个时间点开始。"
+            "考虑因素：画面清晰度、主体是否明确、是否有明显的动作或场景变化。"
+            '请严格以JSON数组格式返回，不要包含其他文字：'
+            '[{"asset_id": "素材ID", "recommended_start_time": 12.5, "reason": "原因"}]'
+        )
+
+        contents = client.analyze_videos(
+            video_urls=video_urls,
+            prompt=prompt,
+            level="Economy",
+            poll_interval=2.0,
+            max_poll_attempts=15,
+        )
+
+        if not contents:
+            logger.info("MediaKit 分析无结果，降级为随机选择")
+            return {}
+
+        # 按索引映射结果：contents[i] 对应 valid_asset_ids[i]
+        recommendations: dict[str, float] = {}
+        for idx, content_text in enumerate(contents):
+            if idx >= len(valid_asset_ids):
+                break
+            asset_id = valid_asset_ids[idx]
+            if not content_text:
+                continue
+
+            # 尝试从文本中提取 JSON
+            parsed = False
+            # 尝试直接解析
+            try:
+                data = json.loads(content_text.strip())
+                if isinstance(data, list) and data:
+                    for item in data:
+                        if isinstance(item, dict) and "recommended_start_time" in item:
+                            recommendations[asset_id] = float(item["recommended_start_time"])
+                            parsed = True
+                            break
+            except (json.JSONDecodeError, ValueError, TypeError):
+                pass
+
+            # 尝试从 markdown 代码块中提取 JSON
+            if not parsed:
+                json_match = re.search(r"\[\s*(\{.*?\})\s*\]", content_text, re.DOTALL)
+                if json_match:
+                    try:
+                        item = json.loads(json_match.group(1))
+                        if isinstance(item, dict) and "recommended_start_time" in item:
+                            recommendations[asset_id] = float(item["recommended_start_time"])
+                            parsed = True
+                    except (json.JSONDecodeError, ValueError, TypeError):
+                        pass
+
+            # 尝试正则提取
+            if not parsed:
+                time_match = re.search(
+                    r'recommended_start_time["\s:]+([\d.]+)', content_text
+                )
+                if time_match:
+                    try:
+                        recommendations[asset_id] = float(time_match.group(1))
+                    except (ValueError, TypeError):
+                        pass
+
+        if recommendations:
+            logger.info("MediaKit 智能选片推荐: %s", recommendations)
+        else:
+            logger.info("MediaKit 结果解析失败，降级为随机选择")
+
+        return recommendations
+
+    except Exception as e:
+        logger.warning("MediaKit 智能选片异常，降级为随机选择: %s", e)
+        return {}
+
+
 @router.post("/clips/from-assets", response_model=ClipsFromAssetsResponse)
 def create_clips_from_assets_editor(
     template_id: str,
     body: ClipsFromAssetsRequest,
+    background_tasks: BackgroundTasks,
     plan_id: str = Depends(get_draft_plan_id),
     services: tuple[EditTemplateService, EditPlanService] = Depends(get_editor_services),
+    asset_repo: SQLAlchemyAssetRepository = Depends(get_asset_repository),
+    db: Session = Depends(get_db_session),
     current_user: AuthenticatedUser = Depends(get_current_user),
 ) -> ClipsFromAssetsResponse:
-    """从素材批量创建片段"""
-    _, plan_svc = services
-    clips = []
-    for i, asset_id in enumerate(body.asset_ids):
-        try:
-            clip = plan_svc.create_clip(
-                plan_id,
-                clip_type="main",
-                order=body.start_order + i if hasattr(body, "start_order") else i,
-                duration=5.0,
-                asset_id=asset_id,
+    """从素材批量创建片段（按模板segment配置创建，MediaKit异步更新）.
+
+    逻辑：
+    1. 从模板读取 segments，片段数量 = segment 数量（忽略前端传的 required_clips_count）
+    2. 每个片段时长在 segment 的 duration_min ~ duration_max 之间随机取值（保留一位小数）
+    3. 素材按片段顺序轮询分配，素材不够时同一素材切多个片段
+    4. 使用 replace_all_clips_transactional 原子性地清空旧片段并创建新的（随机起始时间）
+    5. 立即返回响应（目标 <1秒）
+    6. 后台异步任务：调用 MediaKit 智能选片并更新片段的 start_time
+    7. 素材时长为 0 或缺失时报 400，不创建无效片段
+    """
+    tpl_svc, plan_svc = services
+
+    # 1. 查询模板 segments
+    segments = _get_template_segments(template_id, tpl_svc, db)
+    if not segments:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail="模板没有片段配置，无法创建片段",
+        )
+
+    if not body.asset_ids:
+        raise HTTPException(
+            status_code=status.HTTP_400_BAD_REQUEST,
+            detail="素材列表为空，无法创建片段",
+        )
+
+    # 2. 获取素材实际时长（去重查询）
+    unique_asset_ids = list(dict.fromkeys(body.asset_ids))
+    asset_durations: dict[str, float] = {}
+    for asset_id in unique_asset_ids:
+        asset = asset_repo.get(asset_id)
+        if asset and hasattr(asset, "duration"):
+            asset_durations[asset_id] = float(asset.duration or 0.0)
+
+    # 3. 在内存中计算所有片段数据（使用随机起始时间，不调用MediaKit）
+    used_segments: dict[str, list[tuple[float, float]]] = {}
+    clips_data: list[dict] = []
+
+    for i, (_seg_order, dur_min, dur_max) in enumerate(segments):
+        # 轮询分配素材
+        asset_id = body.asset_ids[i % len(body.asset_ids)]
+        asset_total = asset_durations.get(asset_id, 0.0)
+
+        # 素材时长为 0 或缺失时无法创建有效片段
+        if asset_total <= 0:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail=f"素材 {asset_id} 时长信息缺失或为0，无法创建片段",
             )
-            clips.append(clip)
-        except ValueError:
-            pass
+
+        # 在 segment 的 duration_min ~ duration_max 之间随机取值（保留一位小数）
+        raw_duration = random.uniform(dur_min, dur_max)
+        clip_duration = round(raw_duration, 1)
+
+        # 素材时长不足时缩短 clip duration
+        clip_duration = min(clip_duration, asset_total)
+
+        if clip_duration <= 0:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail=f"素材 {asset_id} 时长不足，无法创建有效片段",
+            )
+
+        # 使用随机起始时间（不调用MediaKit，保证接口快速返回）
+        start_time = _calc_random_start_time(
+            asset_id, clip_duration, asset_durations, used_segments
+        )
+
+        if start_time is None:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail=f"素材 {asset_id} 时长信息缺失，无法计算起始时间",
+            )
+
+        # 记录已使用时间段
+        used_segments.setdefault(asset_id, []).append(
+            (start_time, start_time + clip_duration)
+        )
+
+        clips_data.append(
+            {
+                "order": i,
+                "asset_id": asset_id,
+                "start_time": start_time,
+                "duration": clip_duration,
+                "clip_type": body.clip_type or "main",
+            }
+        )
+
+    # 4. 事务性替换：清空旧片段 → 创建新片段 → 标记ready（单事务，失败自动回滚）
+    created_count = plan_svc.replace_all_clips_transactional(plan_id, clips_data)
 
     logger.info(
-        "模板编辑器从素材创建片段: template_id=%s plan_id=%s count=%d by user=%s",
+        "from-assets按模板创建片段(异步): template_id=%s plan_id=%s segments=%d created=%d by user=%s",
         template_id,
         plan_id,
-        len(clips),
+        len(segments),
+        created_count,
         current_user.user.id,
     )
 
-    return ClipsFromAssetsResponse(
-        created_count=len(clips),
-        plan_id=plan_id,
-        clip_ids=[c.id for c in clips],
+    # 5. 触发后台任务：异步调用 MediaKit 并更新片段起始时间
+    background_tasks.add_task(
+        _update_mediakit_recommendations_async,
+        plan_id,
+        unique_asset_ids,
     )
+
+    # 6. 立即返回响应
+    return ClipsFromAssetsResponse(
+        created_count=created_count,
+        plan_id=plan_id,
+        clip_ids=[],
+    )
+
+
+def _update_mediakit_recommendations_async(  # pragma: no cover
+    plan_id: str,
+    asset_ids: list[str],
+) -> None:
+    """后台任务：调用 MediaKit 智能选片并更新片段的起始时间.
+
+    此函数在后台异步执行，不影响接口响应时间。
+    失败时静默处理，不影响已创建的片段。
+    """
+    from collections import defaultdict
+
+    from packages.adapters.sqlalchemy_impl.asset_repository import SQLAlchemyAssetRepository
+    from packages.adapters.sqlalchemy_impl.session import SessionLocal
+
+    db = None
+    try:
+        # 复用应用全局 Session（避免每次创建新连接池导致资源泄漏）
+        if SessionLocal is None:
+            logger.warning("后台任务: SessionLocal 未初始化，跳过 MediaKit 更新")
+            return
+        db = SessionLocal()
+
+        # 初始化服务
+        asset_repo = SQLAlchemyAssetRepository(db)
+        plan_svc = EditPlanService(db)
+
+        # 调用 MediaKit 获取推荐时间
+        recommendations = _get_mediakit_recommendations(asset_ids, asset_repo)
+        if not recommendations:
+            logger.info("后台任务: MediaKit 无推荐结果，跳过更新")
+            return
+
+        # 查询该 plan 的所有片段（分批获取，避免硬编码 limit 截断）
+        batch_size = 500
+        all_clips = []
+        offset = 0
+        while True:
+            batch = plan_svc.list_clips(plan_id, skip=offset, limit=batch_size)
+            if not batch:
+                break
+            all_clips.extend(batch)
+            if len(batch) < batch_size:
+                break
+            offset += batch_size
+        clips = all_clips
+
+        if not clips:
+            logger.info("后台任务: plan_id=%s 无片段，跳过更新", plan_id)
+            return
+
+        # 批量预加载所有涉及的素材（消除 N+1 查询）
+        unique_asset_ids = list({getattr(c, "asset_id", "") or "" for c in clips} - {""})
+        assets_map: dict[str, object] = {
+            a.id: a for a in asset_repo.find_by_ids(unique_asset_ids)
+        }
+
+        # 按 asset_id 预分组片段时间段（消除 O(N^2) 嵌套循环）
+        clips_by_asset: dict[str, list[tuple[str, float, float]]] = defaultdict(list)
+        for clip in clips:
+            aid = getattr(clip, "asset_id", "") or ""
+            if aid and clip.start_time is not None:
+                clips_by_asset[aid].append(
+                    (clip.id, clip.start_time, clip.start_time + clip.duration)
+                )
+
+        # 已更新的片段ID（用于排除已移动的旧时间段）
+        updated_clip_ids: set[str] = set()
+        # 已更新的时间段
+        updated_segments: dict[str, list[tuple[float, float]]] = {}
+        updated_count = 0
+
+        # 遍历片段，按 asset_id 匹配推荐时间
+        for clip in clips:
+            asset_id = getattr(clip, "asset_id", "") or ""
+            if not asset_id or asset_id not in recommendations:
+                continue
+
+            recommended_start = recommendations[asset_id]
+            clip_duration = clip.duration
+
+            # 从预加载字典获取素材（O(1) 查找）
+            asset = assets_map.get(asset_id)
+            if not asset:
+                continue
+            asset_total = float(getattr(asset, "duration", 0.0) or 0.0)
+            if asset_total <= 0:
+                continue
+
+            # 推荐时间 + 片段时长不能超过素材总时长
+            if recommended_start + clip_duration > asset_total:
+                logger.info(
+                    "后台任务: 推荐时间越界，跳过: asset_id=%s recommended=%.2f duration=%.1f total=%.1f",
+                    asset_id,
+                    recommended_start,
+                    clip_duration,
+                    asset_total,
+                )
+                continue
+
+            # 构建排除当前片段及已更新片段后的占用列表（O(M)，M=同素材片段数）
+            other_segments: list[tuple[float, float]] = [
+                (cs, ce)
+                for cid, cs, ce in clips_by_asset.get(asset_id, [])
+                if cid != clip.id and cid not in updated_clip_ids
+            ]
+            other_segments.extend(updated_segments.get(asset_id, []))
+
+            # 检查是否与同素材其他片段时间段冲突
+            if _recommended_time_conflicts(recommended_start, clip_duration, other_segments):
+                logger.info(
+                    "后台任务: 推荐时间冲突，跳过: asset_id=%s recommended=%.2f",
+                    asset_id,
+                    recommended_start,
+                )
+                continue
+
+            # 逐个更新并捕获异常（单点失败不影响其他片段）
+            try:
+                plan_svc.update_clip(clip.id, start_time=recommended_start)
+                db.commit()
+                updated_count += 1
+                updated_clip_ids.add(clip.id)
+            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
+                )
+                try:
+                    db.rollback()
+                except Exception:
+                    pass
+                continue
+
+            updated_segments.setdefault(asset_id, []).append(
+                (recommended_start, recommended_start + clip_duration)
+            )
+            logger.info(
+                "后台任务: 更新片段起始时间: clip_id=%s asset_id=%s start_time=%.2f",
+                clip.id,
+                asset_id,
+                recommended_start,
+            )
+
+        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:
+        # 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)
+        if db:
+            try:
+                db.rollback()
+            except Exception:
+                pass
+    finally:
+        if db:
+            try:
+                db.close()
+            except Exception:
+                pass
+

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

@@ -3,7 +3,6 @@
 核心依赖：
 - get_editor_services: 获取模板+计划服务
 - get_draft_plan_id: 根据 template_id 获取或创建草稿，返回 plan_id
-- _check_queue_limits: 生成队列限流检查
 """
 
 from __future__ import annotations
@@ -11,7 +10,6 @@ from __future__ import annotations
 import logging
 
 from app.auth import AuthenticatedUser, get_current_user
-from app.core.task_enqueue import GLOBAL_PENDING_LIMIT, USER_PENDING_LIMIT
 from app.dependencies import get_db_session
 from app.services.edit_plan_service import EditPlanService
 from app.services.edit_template_service import EditTemplateService
@@ -113,29 +111,3 @@ def get_draft_plan_id(
         user_id,
     )
     return plan.id
-
-
-def _check_queue_limits(gen_task_repo, user_id: str) -> None:
-    """队列限流预检查"""
-    try:
-        has_count = (
-            hasattr(gen_task_repo, "count_pending_by_user")
-            and hasattr(gen_task_repo, "count_pending_total")
-        )
-        if has_count:
-            user_pending = gen_task_repo.count_pending_by_user(user_id)
-            global_pending = gen_task_repo.count_pending_total()
-            if user_pending >= USER_PENDING_LIMIT:
-                raise HTTPException(
-                    status_code=429,
-                    detail=f"您的待处理任务过多（当前 {user_pending}/{USER_PENDING_LIMIT}），请等待完成后再提交",
-                )
-            if global_pending >= GLOBAL_PENDING_LIMIT:
-                raise HTTPException(
-                    status_code=503,
-                    detail="系统繁忙，请稍后再试",
-                )
-    except HTTPException:
-        raise
-    except Exception as e:
-        logger.warning("[模板编辑器队列限流] 检查失败，跳过: %s", e)

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

@@ -17,6 +17,8 @@ from fastapi import APIRouter, Depends, HTTPException, Query, status
 
 from .dependencies import get_draft_plan_id, get_editor_services
 from .schemas import (
+    EditorClipBatchUpdateRequest,
+    EditorClipBatchUpdateResponse,
     EditorDraftResponse,
     EditorPublishResponse,
     EditorRollbackRequest,
@@ -126,11 +128,7 @@ def list_template_versions(
             clip_count=len(v.clip_configs),
             change_note=v.change_note,
             published_by=v.published_by,
-            created_at=(
-                v.created_at.isoformat()
-                if hasattr(v.created_at, "isoformat")
-                else str(v.created_at)
-            ),
+            created_at=(v.created_at.isoformat() if hasattr(v.created_at, "isoformat") else str(v.created_at)),
         )
         for v in versions
     ]
@@ -162,3 +160,35 @@ def rollback_template(
         new_version=tpl.version,
         clip_count=len(clip_configs),
     )
+
+
+@router.put("/clips", response_model=EditorClipBatchUpdateResponse)
+def batch_update_clips(
+    template_id: str,
+    req: EditorClipBatchUpdateRequest,
+    plan_id: str = Depends(get_draft_plan_id),
+    services: tuple[EditTemplateService, EditPlanService] = Depends(get_editor_services),
+    _: AuthenticatedUser = Depends(get_current_user),
+):
+    """批量替换草稿clips（全量覆盖，用于前端选择素材后同步片段）
+
+    事务保证：清空→创建→标记ready 在同一数据库事务内完成，
+    任何步骤失败时自动回滚，避免数据不一致。
+    """
+    _, plan_svc = services
+    plan_svc.get_plan_or_raise(plan_id)
+
+    clips_data = []
+    for clip_item in req.clips:
+        item = {
+            "asset_id": clip_item.asset_id,
+            "start_time": clip_item.start_time,
+            "duration": clip_item.duration,
+        }
+        if clip_item.order is not None:
+            item["order"] = clip_item.order
+        clips_data.append(item)
+
+    plan_svc.replace_all_clips_transactional(plan_id, clips_data)
+
+    return EditorClipBatchUpdateResponse(plan_id=plan_id, clip_count=len(req.clips))

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

@@ -1,383 +0,0 @@
-"""草稿生成路由.
-
-端点：
-- POST /generate              触发生成
-- GET  /generation-status     生成进度
-- GET  /generations           生成记录列表
-"""
-
-from __future__ import annotations
-
-import json
-import logging
-from typing import Any, Optional
-
-from app.auth import AuthenticatedUser, get_current_user
-from app.core.celery_app import celery_app
-from app.core.storage import OSSStorageService, get_storage_service
-from app.dependencies import (
-    get_asset_library_repository,
-    get_asset_repository,
-    get_db_session,
-    get_generated_video_repository,
-)
-from app.schemas.generation_task import GenerationTaskResponse
-from app.services.edit_plan_service import EditPlanService
-from app.services.edit_template_service import EditTemplateService
-from fastapi import APIRouter, Depends, HTTPException, status
-from sqlalchemy.orm import Session
-
-from packages.adapters.sqlalchemy_impl.generation_task_repository import (
-    SQLAlchemyGenerationTaskRepository,
-)
-from packages.application.generated_videos import ListGeneratedVideosByTaskUseCase
-from packages.application.generation_tasks import (
-    CreateGenerationTaskCommand,
-    CreateGenerationTaskUseCase,
-)
-from packages.domain.edit_plan import EditPlanStatus
-
-from ._fallback import (
-    _auto_fallback_assign_assets,
-    _auto_fallback_auto_material_mode,
-    _auto_fallback_copy_template_clips,
-    _auto_fallback_draft_to_editing,
-)
-from .dependencies import _check_queue_limits, get_draft_plan_id, get_editor_services
-from .schemas import (
-    ClipStatusItem,
-    EditPlanGenerateRequest,
-    EditPlanGenerateResponse,
-    EditPlanGenerationsResponse,
-    EditPlanGenerationStatusResponse,
-)
-
-logger = logging.getLogger(__name__)
-router = APIRouter(tags=["Template Editor"])
-
-
-@router.post("/generate", response_model=EditPlanGenerateResponse)
-def generate_editor_draft(
-    template_id: str,
-    request: Optional[EditPlanGenerateRequest] = None,
-    plan_id: str = Depends(get_draft_plan_id),
-    services: tuple[EditTemplateService, EditPlanService] = Depends(get_editor_services),
-    db: Session = Depends(get_db_session),
-    current_user: AuthenticatedUser = Depends(get_current_user),
-    asset_library_repo: Any = Depends(get_asset_library_repository),
-    asset_repo: Any = Depends(get_asset_repository),
-) -> EditPlanGenerateResponse:
-    """触发模板草稿渲染生成"""
-    req = request or EditPlanGenerateRequest()
-    _, plan_svc = services
-    plan_check = plan_svc.get_plan_or_raise(plan_id)
-
-    # 自动兜底流程
-    _auto_fallback_draft_to_editing(plan_svc, plan_id, plan_check)
-    _auto_fallback_copy_template_clips(plan_svc, plan_id, plan_check, db)
-    clips_without_asset = _auto_fallback_assign_assets(plan_svc, plan_id, plan_check)
-    _auto_fallback_auto_material_mode(
-        plan_svc,
-        plan_id,
-        plan_check,
-        clips_without_asset,
-        asset_library_repo,
-        asset_repo,
-        user_id=str(current_user.user.id),
-    )
-
-    # 检查是否可复用已完成的预览产物（预览品质已与正式一致）
-    gen_task_repo = SQLAlchemyGenerationTaskRepository(db)
-    reusable_task = _find_reusable_preview_task(gen_task_repo, plan_id, plan_check)
-    if reusable_task:
-        # 复用预览产物：标记为正式产出，跳过渲染
-        # 如果前端传了 title_config，需要创建新任务（因为预览任务的 custom_title 可能不同）
-        title_config_reuse = req.title_config or {}
-        title_text_reuse = (title_config_reuse.get("text") or "").strip()
-        existing_custom_title = getattr(reusable_task, "custom_title", "") or ""
-        if title_text_reuse and existing_custom_title:
-            # 如果新标题和已有标题不同，不能复用，走新建任务流程
-            new_title_json = json.dumps(title_config_reuse, ensure_ascii=False)
-            if new_title_json != existing_custom_title:
-                logger.info(
-                    "[模板生成] 标题已变更，跳过复用: task_id=%s",
-                    reusable_task.id,
-                )
-                reusable_task = None
-        elif title_text_reuse and not existing_custom_title:
-            # 原来没标题，现在有标题，不能复用
-            logger.info(
-                "[模板生成] 新增标题，跳过复用: task_id=%s",
-                reusable_task.id,
-            )
-            reusable_task = None
-        elif not title_text_reuse and existing_custom_title:
-            # 原来有标题，现在移除了，不能复用
-            logger.info(
-                "[模板生成] 移除标题，跳过复用: task_id=%s",
-                reusable_task.id,
-            )
-            reusable_task = None
-
-    if reusable_task:
-        # 复用预览产物：标记为正式产出，跳过渲染
-        reusable_task.mark_confirmed()
-        gen_task_repo.update(reusable_task)
-
-        # 将产物 URL 写入 plan config
-        rendered_url = _get_task_output_url(reusable_task, gen_task_repo, db)
-        plan_svc.update_plan_config(
-            plan_id,
-            {
-                "generation_task_id": reusable_task.id,
-                "rendered_storage_key": rendered_url,  # 统一用 rendered_storage_key
-            },
-        )
-        plan_svc.transition_status(plan_id, EditPlanStatus.COMPLETED)
-
-        updated_plan = plan_svc.get_plan_or_raise(plan_id)
-        logger.info(
-            "模板编辑器复用预览产物: template_id=%s plan_id=%s task_id=%s by user=%s",
-            template_id,
-            plan_id,
-            reusable_task.id,
-            current_user.user.id,
-        )
-        return EditPlanGenerateResponse(
-            plan_id=plan_id,
-            plan_status=updated_plan.status.value if hasattr(updated_plan.status, "value") else updated_plan.status,
-            generation_task_id=reusable_task.id,
-            clip_count=len((plan_check.config or {}).get("clips", [])),
-        )
-
-    # 检查是否可生成（含最后防线自动修复 + 诊断日志）
-    try:
-        can_gen, reason = plan_svc.can_generate(plan_id)
-    except ValueError as exc:
-        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail=str(exc)) from exc
-    if not can_gen:
-        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=reason)
-
-    try:
-        clip_count = plan_svc.mark_clips_ready(plan_id)
-
-        user_id = current_user.user.id
-        _check_queue_limits(gen_task_repo, user_id)
-
-        gen_task_use_case = CreateGenerationTaskUseCase(gen_task_repo)
-        plan = plan_svc.get_plan_or_raise(plan_id)
-        config_asset_ids = (plan.config or {}).get("asset_ids", [])
-        # 从 plan config 读取封面 URL（由 generate-cover 保存）
-        cover_url_from_config = (plan.config or {}).get("cover", {}).get("image_url", "")
-
-        # 处理标题配置：序列化 title_config 为 JSON 存入 custom_title
-        title_config = req.title_config or {}
-        title_text = (title_config.get("text") or "").strip()
-        custom_title_value = ""
-        if title_text:
-            custom_title_value = json.dumps(title_config, ensure_ascii=False)
-            logger.info(
-                "[模板生成] 标题配置: text=%s, config_keys=%s",
-                title_text[:30],
-                list(title_config.keys()),
-            )
-
-        gen_task = gen_task_use_case.execute(
-            CreateGenerationTaskCommand(
-                project_id=plan.project_id or "",
-                template_id=plan.template_id,
-                created_by_user_id=current_user.user.id,
-                source_edit_plan_id=plan_id,
-                asset_ids=list(config_asset_ids) if config_asset_ids else [],
-                cover_url=cover_url_from_config,
-                custom_title=custom_title_value,
-            ),
-        )
-
-        plan_svc.update_plan_config(plan_id, {"generation_task_id": gen_task.id})
-        plan_svc.transition_status(plan_id, EditPlanStatus.RENDERING)
-        celery_app.send_task("worker.render_edit_plan", args=[plan_id])
-
-        updated_plan = plan_svc.get_plan_or_raise(plan_id)
-
-        logger.info(
-            "模板编辑器触发生成: template_id=%s plan_id=%s gen_task_id=%s clips=%d by user=%s",
-            template_id,
-            plan_id,
-            gen_task.id,
-            clip_count,
-            current_user.user.id,
-        )
-
-        return EditPlanGenerateResponse(
-            plan_id=plan_id,
-            plan_status=updated_plan.status.value if hasattr(updated_plan.status, "value") else updated_plan.status,
-            generation_task_id=gen_task.id,
-            clip_count=clip_count,
-        )
-    except HTTPException:
-        raise
-    except Exception as _e:
-        logger.exception(
-            "模板编辑器触发生成失败: template_id=%s plan_id=%s",
-            template_id,
-            plan_id,
-        )
-        try:
-            plan_svc.transition_status(plan_id, EditPlanStatus.FAILED)
-        except Exception:
-            pass
-        raise HTTPException(
-            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
-            detail="生成失败，请稍后重试",
-        ) from _e
-
-
-def _find_reusable_preview_task(gen_task_repo, plan_id: str, plan) -> "object | None":
-    """查找该 plan 关联的已完成预览任务，判断是否可复用。
-
-    复用条件：
-    1. 存在 source_edit_plan_id == plan_id 的已完成预览任务
-    2. plan 在预览完成后未被修改（updated_at <= 预览完成时间）
-
-    Returns:
-        可复用的 GenerationTask，或 None
-    """
-    try:
-        tasks = gen_task_repo.list_by_source_edit_plan(plan_id)
-    except Exception:
-        return None
-
-    for task in tasks:
-        if not getattr(task, "is_preview", False):
-            continue
-        if not task.is_completed:
-            continue
-        # 检查 plan 是否在预览完成后被修改
-        completed_at = getattr(task, "completed_at", None)
-        if completed_at and hasattr(plan, "updated_at"):
-            plan_updated = plan.updated_at
-            # 如果 plan.updated_at 为空，无法判断是否修改过，跳过
-            if plan_updated is None:
-                continue
-            # 如果 plan 在预览完成后又被修改了，不能复用
-            if plan_updated > completed_at:
-                continue
-        return task
-    return None
-
-
-def _get_task_output_url(task, gen_task_repo, db) -> str:
-    """获取任务的输出视频 URL。"""
-    try:
-        video_repo = get_generated_video_repository(db)
-        use_case = ListGeneratedVideosByTaskUseCase(video_repo)
-        videos = use_case.execute(task.id)
-        if videos:
-            url = getattr(videos[0], "file_url", "") or ""
-            # 规范化：合并路径中的双斜杠（保留协议头 ://）
-            if url:
-                import re as _re
-                url = _re.sub(r"(?<!:)//", "/", url)
-            return url
-    except Exception:
-        pass
-    return ""
-
-
-@router.get("/generation-status", response_model=EditPlanGenerationStatusResponse)
-def get_editor_generation_status(
-    template_id: str,
-    plan_id: str = Depends(get_draft_plan_id),
-    services: tuple[EditTemplateService, EditPlanService] = Depends(get_editor_services),
-    storage_service: OSSStorageService = Depends(get_storage_service),
-    _: AuthenticatedUser = Depends(get_current_user),
-) -> EditPlanGenerationStatusResponse:
-    """查询草稿生成进度"""
-    _, plan_svc = services
-    try:
-        gen_status = plan_svc.get_generation_status(plan_id)
-    except ValueError as exc:
-        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail=str(exc)) from exc
-
-    plan = gen_status["plan"]
-    clips = gen_status["clips"]
-
-    clip_items = [
-        ClipStatusItem(
-            clip_id=c.id,
-            clip_type=c.clip_type,
-            order=c.order,
-            status=c.status.value if hasattr(c.status, "value") else c.status,
-            asset_id=c.asset_id or "",
-            text_content=c.text_content or "",
-            duration=c.duration,
-        )
-        for c in clips
-    ]
-
-    raw_video_url = (plan.config or {}).get("rendered_storage_key", "") or (plan.config or {}).get("rendered_url", "")
-    video_url = ""
-    if raw_video_url:
-        if raw_video_url.startswith("http"):
-            video_url = raw_video_url  # 已经是完整 URL
-        else:
-            try:
-                video_url = storage_service.get_url(raw_video_url)  # storage_key -> 完整 URL
-            except Exception as e:
-                logger.warning("生成视频URL获取失败: template_id=%s error=%s", template_id, e)
-                video_url = raw_video_url
-
-    progress = gen_status.get("progress", 0.0)
-    error_message = gen_status.get("error_message", "")
-    gen_task_status = gen_status.get("generation_task_status")
-    plan_status_val = plan.status.value if hasattr(plan.status, "value") else plan.status
-    if plan_status_val == "completed" and progress < 100:
-        progress = 100.0
-
-    return EditPlanGenerationStatusResponse(
-        plan_id=plan_id,
-        plan_status=plan_status_val,
-        generation_task_id=gen_status["generation_task_id"],
-        generation_task_status=gen_task_status,
-        progress=progress,
-        video_url=video_url,
-        error_message=error_message,
-        clips=clip_items,
-    )
-
-
-@router.get("/generations", response_model=EditPlanGenerationsResponse)
-def list_editor_generations(
-    template_id: str,
-    plan_id: str = Depends(get_draft_plan_id),
-    services: tuple[EditTemplateService, EditPlanService] = Depends(get_editor_services),
-    db: Session = Depends(get_db_session),
-    _: AuthenticatedUser = Depends(get_current_user),
-) -> EditPlanGenerationsResponse:
-    """查询草稿关联的生成记录列表"""
-    _, plan_svc = services
-    plan_svc.get_plan_or_raise(plan_id)
-
-    gen_task_repo = SQLAlchemyGenerationTaskRepository(db)
-    tasks = gen_task_repo.list_by_source_edit_plan(plan_id)
-    items = [
-        GenerationTaskResponse(
-            id=t.id,
-            project_id=t.project_id,
-            asset_library_id=t.asset_library_id,
-            strategy_id=t.strategy_id,
-            voice_library_id=t.voice_library_id,
-            template_id=t.template_id,
-            asset_ids=t.asset_ids,
-            title_ids=t.title_ids,
-            voice_ids=t.voice_ids,
-            source_edit_plan_id=t.source_edit_plan_id or "",
-            status=t.status.value if hasattr(t.status, "value") else t.status,
-            progress=t.progress,
-            result_count=t.result_count,
-            error_message=t.error_message,
-        )
-        for t in tasks
-    ]
-    return EditPlanGenerationsResponse(items=items, total=len(items))

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

@@ -6,9 +6,8 @@
 from __future__ import annotations
 
 import re as _re
-from typing import Any, Dict, List, Optional
+from typing import Any, List, Optional
 
-from app.schemas.generation_task import GenerationTaskResponse
 from pydantic import BaseModel, Field, validator
 
 _EXPORT_RESOLUTION_PATTERN = _re.compile(r"^\d+x\d+$")
@@ -16,58 +15,6 @@ _EXPORT_VALID_QUALITY_PRESETS = {"ultra_fast", "fast", "balanced", "high", "best
 _EXPORT_VALID_FORMATS = {"mp4", "mov"}
 
 
-# ── 生成状态相关 ────────────────────────────────────────────────────────────
-
-
-class ClipStatusItem(BaseModel):
-    """片段生成状态"""
-
-    clip_id: str
-    clip_type: str
-    order: int
-    status: str
-    asset_id: str
-    text_content: str
-    duration: float
-
-
-class EditPlanGenerationStatusResponse(BaseModel):
-    """剪辑计划生成进度响应体"""
-
-    plan_id: str
-    plan_status: str
-    generation_task_id: Optional[str] = None
-    generation_task_status: Optional[str] = None
-    progress: float = 0.0
-    video_url: str = ""
-    error_message: str = ""
-    clips: List[ClipStatusItem]
-
-
-class EditPlanGenerateRequest(BaseModel):
-    """模板编辑器触发生成请求体"""
-    title_config: Optional[Dict[str, Any]] = Field(
-        default_factory=dict,
-        description="标题配置（可选），渲染时烧录到视频中。支持字段: text/font/font_size/font_color/position/bold/stroke/shadow",
-    )
-
-
-class EditPlanGenerateResponse(BaseModel):
-    """剪辑计划触发生成响应体"""
-
-    plan_id: str
-    plan_status: str
-    generation_task_id: str
-    clip_count: int
-
-
-class EditPlanGenerationsResponse(BaseModel):
-    """剪辑计划关联的生成记录列表响应体"""
-
-    items: List[GenerationTaskResponse]
-    total: int
-
-
 # ── AI 推荐 ────────────────────────────────────────────────────────────────
 
 
@@ -75,12 +22,8 @@ class AIRecommendRequest(BaseModel):
     """AI 推荐片段方案请求体"""
 
     asset_ids: List[str] = Field(default_factory=list, description="素材 ID 列表")
-    editing_mode: str = Field(
-        default="one_take", description="剪辑模式: one_take / pip / voice_over / voice_pip"
-    )
-    target_duration: float = Field(
-        default=30.0, ge=1.0, le=600.0, description="目标时长（秒）"
-    )
+    editing_mode: str = Field(default="one_take", description="剪辑模式: one_take / pip / voice_over / voice_pip")
+    target_duration: float = Field(default=30.0, ge=1.0, le=600.0, description="目标时长（秒）")
 
 
 class AIRecommendClipItem(BaseModel):
@@ -107,8 +50,6 @@ class AIRecommendResponse(BaseModel):
     confidence: float = Field(..., ge=0.0, le=1.0, description="AI 推荐置信度 (0~1)")
 
 
-
-
 # ── BGM ────────────────────────────────────────────────────────────────────
 
 
@@ -224,10 +165,9 @@ class ClipBatchDeleteResponse(BaseModel):
 class ClipsFromAssetsRequest(BaseModel):
     """从素材批量创建片段请求"""
 
-    asset_ids: List[str] = Field(
-        ..., min_length=1, max_length=200, description="素材 ID 列表，按顺序追加到时间线末尾"
-    )
+    asset_ids: List[str] = Field(..., min_length=1, max_length=200, description="素材 ID 列表，按顺序追加到时间线末尾")
     clip_type: str = Field(default="main", description="片段类型，默认 main")
+    required_clips_count: Optional[int] = Field(default=None, ge=1, le=200, description="要求创建的片段数量；不传则等于素材数量")
 
 
 class ClipsFromAssetsResponse(BaseModel):
@@ -501,6 +441,28 @@ class EditorClipUpdateRequest(BaseModel):
     config: Optional[dict[str, Any]] = None
 
 
+class EditorClipBatchItem(BaseModel):
+    """批量更新clips的单个片段"""
+
+    asset_id: str = Field(default="", max_length=100, description="关联素材ID，可为空（占位片段）")
+    start_time: float = Field(default=0.0, ge=0.0)
+    duration: float = Field(default=0.0, ge=0.0)
+    order: Optional[int] = Field(default=None, ge=0, description="排序，None表示按数组顺序")
+
+
+class EditorClipBatchUpdateRequest(BaseModel):
+    """批量替换clips请求（全量覆盖）"""
+
+    clips: List[EditorClipBatchItem] = Field(default_factory=list)
+
+
+class EditorClipBatchUpdateResponse(BaseModel):
+    """批量更新clips响应"""
+
+    plan_id: str
+    clip_count: int
+
+
 class EditorPublishResponse(BaseModel):
     """发布草稿响应"""
 

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

@@ -20,6 +20,8 @@ from app.schemas.tts import (
     SaveToLibraryRequest,
     SaveToLibraryResponse,
     TTSJobResponse,
+    TTSPreviewRequest,
+    TTSPreviewResponse,
     TTSStatusResponse,
     TTSSynthesizeRequest,
     TTSSynthesizeResponse,
@@ -31,7 +33,7 @@ from packages.adapters.sqlalchemy_impl.tts_job_repository import (
     SQLAlchemyTTSJobRepository,
 )
 from packages.adapters.sqlalchemy_impl.voice_library_repository import SQLAlchemyVoiceLibraryRepository
-from packages.application.cosyvoice_service import CosyVoiceService
+from packages.application.cosyvoice_service import CosyVoiceError, CosyVoiceService
 from packages.application.tts_job.streaming_service import TTSStreamingService
 from packages.application.tts_job.use_cases import (
     CreateTTSJobUseCase,
@@ -373,6 +375,59 @@ def save_tts_job_to_library(
     )
 
 
+
+@router.post("/preview", response_model=TTSPreviewResponse)
+def preview_tts(
+    request: TTSPreviewRequest,
+    authenticated_user: AuthenticatedUser = Depends(get_current_user),
+    cosyvoice_service: CosyVoiceService = Depends(get_cosyvoice_service),
+    voice_clone_repo=Depends(get_voice_clone_profile_repository),
+) -> TTSPreviewResponse:
+    """TTS 预览（试听）——同步合成，立即返回音频 URL。
+
+    用于前端预览配音效果，限制文本长度 200 字以内。
+    支持预设音色和克隆音色：克隆音色传的是 profile UUID，需解析为 CosyVoice voice_id。
+    """
+    # 解析 voice_id：前端可能传 VoiceCloneProfile UUID 或预设音色 ID
+    actual_voice_id = request.voice_id
+    profile = voice_clone_repo.get(request.voice_id)
+    if profile is not None:
+        # 命中克隆音色 profile — 校验归属权限
+        if profile.user_id != authenticated_user.user.id:
+            raise HTTPException(
+                status_code=status.HTTP_403_FORBIDDEN,
+                detail="无权访问该音色",
+            )
+        if not profile.voice_id:
+            raise HTTPException(
+                status_code=status.HTTP_400_BAD_REQUEST,
+                detail="音色克隆尚未完成，请稍后再试",
+            )
+        actual_voice_id = profile.voice_id
+
+    try:
+        result = cosyvoice_service.synthesize_speech(
+            text=request.text,
+            voice_id=actual_voice_id,
+            speed=request.speed,
+        )
+    except CosyVoiceError as e:
+        raise HTTPException(
+            status_code=status.HTTP_502_BAD_GATEWAY,
+            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
+
+    return TTSPreviewResponse(
+        audio_url=result.audio_url,
+        duration=result.duration if result.duration and result.duration > 0 else None,
+    )
+
+
 @router.websocket("/ws/tts/stream")
 async def tts_websocket_stream(
     websocket: WebSocket,

apps/api/app/schemas/generation_task.py

@@ -10,7 +10,7 @@ class ConfirmGenerationRequest(BaseModel):
     output_width: int = Field(default=1080, ge=100, description="输出视频宽度")
     output_height: int = Field(default=1920, ge=100, description="输出视频高度")
     cover_url: str = Field(default="", description="自定义封面图片 URL")
-    custom_title: str = Field(default="", description="自定义视频标题")
+    custom_title: str = Field(default="", description="用户自定义标题文本，非空时同步到任务和编辑计划")
 
 
 class CreateGenerationTaskRequest(BaseModel):
@@ -33,6 +33,11 @@ class CreateGenerationTaskRequest(BaseModel):
     voice_ids: list[str] = Field(default_factory=list)
     # ── 来源剪辑计划 ──
     source_edit_plan_id: str = ""
+    # ── 标题配置（结构化）──
+    title_config: dict | None = Field(
+        default=None,
+        description="标题样式对象，包含 text/font/font_size/font_color/position/bold/stroke/shadow 等。为空时不影响现有行为。",
+    )
     # ── 视频标题 ──
     video_title: str = Field(default="", description="生成视频的标题/名称，为空则使用默认命名")
     # ── 批量生成 ──
@@ -72,7 +77,6 @@ class CreateGenerationTaskRequest(BaseModel):
     output_width: int = Field(default=1280, description="输出视频宽度")
     output_height: int = Field(default=720, description="输出视频高度")
     cover_url: str = Field(default="", description="封面图片 URL")
-    custom_title: str = Field(default="", description="自定义视频标题")
 
     @model_validator(mode="after")
     def _check_at_least_one_mode(self) -> "CreateGenerationTaskRequest":
@@ -108,7 +112,7 @@ class GenerationTaskResponse(BaseModel):
     output_width: int = 1280
     output_height: int = 720
     cover_url: str = ""
-    custom_title: str = ""
+    title_config: dict = Field(default_factory=dict)
     status: str
     progress: float
     result_count: int

apps/api/app/schemas/tts.py

@@ -101,3 +101,19 @@ class SaveToLibraryResponse(BaseModel):
     voice_id: str
     voice_name: str
     status: str
+
+
+class TTSPreviewRequest(BaseModel):
+    """TTS 预览（试听）请求。"""
+
+    text: str = Field(..., min_length=1, max_length=200, description="合成文本，限制 200 字")
+    voice_id: str = Field(..., min_length=1, description="音色 ID")
+    speed: float = Field(1.0, ge=0.5, le=2.0, description="语速")
+    pitch: float = Field(1.0, ge=0.5, le=2.0, description="音调（预留，当前未使用）")
+
+
+class TTSPreviewResponse(BaseModel):
+    """TTS 预览（试听）响应。"""
+
+    audio_url: str = Field(..., description="合成音频 URL")
+    duration: Optional[float] = Field(default=None, description="音频时长（秒）")

apps/api/app/services/edit_plan_service.py

@@ -371,6 +371,88 @@ class EditPlanService:
         logger.info("删除所有片段: plan_id=%s count=%d", plan_id, count)
         return count
 
+    def replace_all_clips_transactional(
+        self,
+        plan_id: str,
+        clips_data: list[dict],
+    ) -> int:
+        """事务性地替换所有片段：清空→创建→标记ready，单事务保证原子性。
+
+        Args:
+            plan_id: 计划 ID
+            clips_data: 片段数据列表，每项包含 asset_id/start_time/duration/order
+
+        Returns:
+            int: 创建的片段数量
+
+        Raises:
+            Exception: 任何步骤失败时自动回滚
+        """
+        from packages.adapters.sqlalchemy_impl.models import EditPlanClipModel
+
+        db = self._clip_repo.session
+        try:
+            # 1. 清空现有 clips（不 commit）
+            deleted_count = db.query(EditPlanClipModel).filter(EditPlanClipModel.plan_id == plan_id).delete()
+
+            # 2. 批量创建新 clips（不 commit）
+            for i, clip_item in enumerate(clips_data):
+                order = clip_item.get("order") or i
+                clip = EditPlanClip.create(
+                    plan_id=plan_id,
+                    clip_type=clip_item.get("clip_type", "main"),
+                    order=order,
+                    asset_id=clip_item.get("asset_id", ""),
+                    start_time=clip_item.get("start_time", 0.0),
+                    duration=clip_item.get("duration", 0.0),
+                )
+                model = EditPlanClipModel(
+                    id=clip.id,
+                    plan_id=clip.plan_id,
+                    clip_type=clip.clip_type,
+                    order=clip.order,
+                    asset_id=clip.asset_id,
+                    text_content=clip.text_content,
+                    start_time=clip.start_time,
+                    duration=clip.duration,
+                    transition_effect=clip.transition_effect,
+                    transition_duration=clip.transition_duration,
+                    playback_speed=clip.playback_speed,
+                    status=clip.status.value,
+                    config=clip.config,
+                )
+                db.add(model)
+
+            # flush 让新建 clip 写入当前事务（未 commit），后续查询才能找到它们
+            db.flush()
+
+            # 3. 标记有 asset_id 的 clips 为 ready（不 commit）
+            pending_with_asset = (
+                db.query(EditPlanClipModel)
+                .filter(
+                    EditPlanClipModel.plan_id == plan_id,
+                    EditPlanClipModel.status == "pending",
+                    EditPlanClipModel.asset_id != "",
+                )
+                .all()
+            )
+            for m in pending_with_asset:
+                m.status = "ready"
+
+            # 4. 一次性提交
+            db.commit()
+            logger.info(
+                "事务性替换片段: plan_id=%s deleted=%d created=%d",
+                plan_id,
+                deleted_count,
+                len(clips_data),
+            )
+            return len(clips_data)
+        except Exception:
+            db.rollback()
+            logger.exception("事务性替换片段失败: plan_id=%s", plan_id)
+            raise
+
     # ── 片段分割与合并 ──────────────────────────────────────────────────────
 
     def split_clip(self, clip_id: str, split_time: float) -> Dict[str, Any]:

apps/api/app/services/plan_generator_service.py

@@ -118,9 +118,9 @@ class PlanGeneratorService:
 
         # 4. 按 editing_mode 分配素材
         if asset_ids:
-            # 如果是随机预览模式，获取素材时长信息
+            # 获取素材时长信息，用于随机起始时间
             asset_durations = None
-            if random_preview and self._asset_repo:
+            if self._asset_repo:
                 asset_durations = self._fetch_asset_durations(asset_ids)
             self._distribute_assets(
                 clips,

apps/web/e2e/core-generation.spec.ts

@@ -185,15 +185,17 @@ test.describe("Core generation flow", () => {
     await expect(page.locator(".xx-choice-item.selected")).toBeVisible()
     await page.getByRole("button", { name: "下一步" }).click()
 
-    // Step 2: select material
+    // Step 2: select material (card grid UI)
     await expect(page.getByRole("heading", { name: /选择素材/ })).toBeVisible()
     const librarySelect = page.locator("select").first()
     await librarySelect.selectOption({ label: libraryName })
-    const materialLabel = page.getByText(sourceFileName).locator("..")
-    await expect(materialLabel.locator("input[type='checkbox']")).toBeVisible({
-      timeout: 10_000,
-    })
-    await materialLabel.locator("input[type='checkbox']").check()
+    // 新 UI: 素材以 9:16 竖屏卡片展示，点击卡片选中
+    // 注意：卡片中心是播放按钮（stopPropagation 会阻止选中），所以点击左上角避开
+    const materialCard = page.getByTestId("material-card").filter({ hasText: sourceFileName })
+    await expect(materialCard).toBeVisible({ timeout: 10_000 })
+    await materialCard.click({ position: { x: 15, y: 15 } })
+    // 验证选中：卡片应出现勾选标记（用 testid 定位，避免 ✓ 字符文本匹配不稳定）
+    await expect(materialCard.getByTestId("material-card-check")).toBeVisible({ timeout: 5_000 })
     await page.getByRole("button", { name: "下一步" }).click()
 
     // Step 3: voice (可选步骤，新注册用户无配音素材，直接跳过)
@@ -231,10 +233,7 @@ test.describe("Core generation flow", () => {
       (response) => {
         const url = response.url()
         const path = new URL(url).pathname
-        return (
-          response.request().method() === "POST" &&
-          path.endsWith("/generation/tasks")
-        )
+        return response.request().method() === "POST" && path.endsWith("/generation/tasks")
       },
       { timeout: 30_000 },
     )

apps/web/e2e/test_asset.spec.ts

@@ -37,14 +37,33 @@ async function loginWithRetry(
   })
 }
 
+async function registerWithRetry(
+  request: APIRequestContext,
+  email: string,
+  username: string,
+  password: string,
+  displayName: string,
+  maxRetries = 2,
+) {
+  for (let i = 0; i <= maxRetries; i++) {
+    const response = await request.post(`${apiBase}/auth/register`, {
+      data: { email, password, username, display_name: displayName },
+    })
+    if (response.status() !== 429) return response
+    console.log(`[register] 触发限流，等待 65s 后重试 (${i + 1}/${maxRetries})`)
+    await new Promise((r) => setTimeout(r, 65000))
+  }
+  return request.post(`${apiBase}/auth/register`, {
+    data: { email, password, username, display_name: displayName },
+  })
+}
+
 /** 注册并登录，返回 { headers, email, username, userId } */
 async function createAuthedUser(request: APIRequestContext, label: string) {
   const email = uniqueEmail(label)
   const username = uniqueUsername(label)
 
-  const reg = await request.post(`${apiBase}/auth/register`, {
-    data: { email, password: PASSWORD, username, display_name: `E2E ${label}` },
-  })
+  const reg = await registerWithRetry(request, email, username, PASSWORD, `E2E ${label}`)
   expect(reg.ok(), `注册应成功: ${await reg.text()}`).toBeTruthy()
   const regData = await reg.json()
 

apps/web/src/api/bgm/bgm.ts

@@ -4,11 +4,20 @@
 import apiClient from "../client"
 import type { BgmPreset, BgmPresetsQuery } from "./types"
 
-/** 获取 BGM 预设列表 */
-export const getBgmPresets = async (params?: BgmPresetsQuery): Promise<BgmPreset[]> => {
+/**
+ * 获取 BGM 预设列表
+ * @param templateId 模板/草稿 ID
+ * @param params 分类/关键词筛选
+ */
+export const getBgmPresets = async (
+  templateId: string,
+  params?: BgmPresetsQuery,
+): Promise<BgmPreset[]> => {
   const searchParams: Record<string, string> = {}
   if (params?.category) searchParams.category = params.category
   if (params?.keyword) searchParams.keyword = params.keyword
-  const res = await apiClient.get("/bgm/presets", { params: searchParams })
+  const res = await apiClient.get(`/templates/${templateId}/editor/bgm/presets`, {
+    params: searchParams,
+  })
   return res.data?.data ?? res.data ?? []
 }

apps/web/src/api/generation/cover.ts

@@ -1,9 +1,22 @@
 import apiClient from "../client"
 
+export interface GenerateCoverTitleConfig {
+  text?: string
+  font?: string
+  font_size?: number
+  font_color?: string
+  position?: string
+  bold?: boolean
+  stroke?: boolean
+  shadow?: boolean
+}
+
 export interface GenerateCoverRequest {
   asset_ids: string[]
   cover_type?: "ai_frame" | "manual" | "upload" | "ai_regenerate"
   frame_time?: number
+  /** 标题样式，用于在封面上叠加标题文字 */
+  title_config?: GenerateCoverTitleConfig
 }
 
 export interface GenerateCoverResponse {

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

@@ -13,6 +13,10 @@ export interface CreatePreviewRequest {
   video_title?: string
   duration?: number
   video_ratio?: string
+  /** 输出视频宽度（与 video_ratio 匹配，如 9:16 → 1080） */
+  output_width?: number
+  /** 输出视频高度（与 video_ratio 匹配，如 9:16 → 1920） */
+  output_height?: number
   /* 标题烧录配置（可选，传入后 ASS 渲染标题到预览视频中） */
   title_config?: {
     text?: string
@@ -38,6 +42,8 @@ export interface CreatePreviewResponse {
   is_preview: boolean
   resolution: string
   created_at: string
+  /** 后端自动关联的编辑计划 ID（用于 fallback 路径传递 source_edit_plan_id） */
+  source_edit_plan_id?: string
 }
 
 /** 预览任务详情响应 */

apps/web/src/api/tasks/tasks.ts

@@ -6,6 +6,7 @@ import apiClient from "../client"
 import type {
   CreateGenerationTaskRequest,
   CreateGenerationTaskResponse,
+  GenerationTaskDetail,
   TaskItem,
   TaskListParams,
   TaskListResponse,
@@ -19,6 +20,12 @@ export const createGenerationTask = async (
   return data
 }
 
+/** 获取单个生成任务详情（轮询用） */
+export const getGenerationTask = async (taskId: string): Promise<GenerationTaskDetail> => {
+  const { data } = await apiClient.get<GenerationTaskDetail>(`/generation/tasks/${taskId}`)
+  return data
+}
+
 /** 获取任务列表（支持分页和筛选） */
 export const getTasks = async (params?: TaskListParams): Promise<TaskListResponse> => {
   const { data } = await apiClient.get<TaskListResponse>("/tasks", {

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

@@ -82,10 +82,20 @@ export interface CreateGenerationTaskRequest {
     stroke?: boolean
     shadow?: boolean
   }
+  /** 关联的草稿 ID（编辑流程数据链路用） */
+  source_edit_plan_id?: string
+  /** 配音素材库 ID（用户上传的音频或 AI 配音素材） */
+  voice_library_id?: string
+  /** 自定义 BGM 配置，覆盖模板 BGM 设置 */
+  bgm_config?: {
+    enabled: boolean
+    preset_id?: string
+    volume?: number
+  }
 }
 
-/** 创建生成任务响应（对齐后端 GenerationTaskResponse） */
-export interface CreateGenerationTaskResponse {
+/** 单个生成任务详情（对齐后端 GenerationTaskResponse） */
+export interface GenerationTaskDetail {
   id: string
   project_id: string
   asset_library_id: string
@@ -95,8 +105,18 @@ export interface CreateGenerationTaskResponse {
   asset_ids: string[]
   title_ids: string[]
   voice_ids: string[]
+  source_edit_plan_id?: string
   status: string
   progress: number
   result_count: number
   error_message: string
+  error_info?: TaskErrorInfo
+  created_at?: string | null
+  updated_at?: string | null
+}
+
+/** 创建生成任务响应（后端返回批量结构 {items, total}） */
+export interface CreateGenerationTaskResponse {
+  items: GenerationTaskDetail[]
+  total: number
 }

apps/web/src/api/template-editor/clips.ts

@@ -90,10 +90,21 @@ export async function createClipsFromAssets(
   templateId: string,
   assetIds: string[],
   clipType = "main",
+  requiredClipsCount?: number,
+  opts?: { signal?: AbortSignal },
 ): Promise<ClipsFromAssetsResponse> {
+  const body: Record<string, unknown> = {
+    asset_ids: assetIds,
+    clip_type: clipType,
+  }
+  if (requiredClipsCount !== undefined) {
+    body.required_clips_count = requiredClipsCount
+  }
+  // from-assets 后端会调用 MediaKit 智能选片（最长 60s），单独延长超时
   const response = await apiClient.post<ClipsFromAssetsResponse>(
     `/templates/${templateId}/editor/clips/from-assets`,
-    { asset_ids: assetIds, clip_type: clipType },
+    body,
+    { timeout: 60000, signal: opts?.signal },
   )
   return response.data
 }

apps/web/src/api/template-editor/editPlans.ts

@@ -1,27 +1,8 @@
 /**
- * 模板草稿 CRUD + 生成相关 API
+ * 模板草稿 CRUD API
  */
 import apiClient from "../client"
-import type {
-  EditPlan,
-  EditPlanListParams,
-  EditPlanListResponse,
-  CreateEditPlanRequest,
-  UpdateEditPlanRequest,
-  GenerateResponse,
-  GenerationStatusResponse,
-  EditPlanGeneration,
-  GeneratedVideo,
-  CopyEditPlanRequest,
-} from "./types"
-
-/** 获取模板草稿列表（支持分页和筛选） */
-export async function getEditPlans(params?: EditPlanListParams): Promise<EditPlanListResponse> {
-  const response = await apiClient.get<EditPlanListResponse>("/templates/drafts", {
-    params,
-  })
-  return response.data
-}
+import type { EditPlan, UpdateEditPlanRequest, GeneratedVideo } from "./types"
 
 /** 获取单个模板草稿 */
 export async function getEditPlan(templateId: string): Promise<EditPlan> {
@@ -29,63 +10,44 @@ export async function getEditPlan(templateId: string): Promise<EditPlan> {
   return response.data
 }
 
-/** 创建模板草稿 */
-export async function createEditPlan(data: CreateEditPlanRequest): Promise<EditPlan> {
-  const response = await apiClient.post("/templates/drafts", data)
-  return response.data
-}
-
-/** 更新模板草稿 */
+/** 更新模板草稿（支持传入 AbortSignal 用于自动保存竞态取消） */
 export async function updateEditPlan(
   templateId: string,
   data: UpdateEditPlanRequest,
+  signal?: AbortSignal,
 ): Promise<EditPlan> {
-  const response = await apiClient.put(`/templates/${templateId}/editor`, data)
+  const response = await apiClient.put(`/templates/${templateId}/editor`, data, { signal })
   return response.data
 }
 
-/** 删除模板草稿 */
-export async function deleteEditPlan(templateId: string): Promise<void> {
-  await apiClient.delete(`/templates/${templateId}/editor`)
-}
-
-/** 触发生成 */
-export async function generateEditPlan(templateId: string): Promise<GenerateResponse> {
-  const response = await apiClient.post(`/templates/${templateId}/editor/generate`)
-  return response.data
-}
-
-/** 获取生成状态（轮询用） */
-export async function getGenerationStatus(templateId: string): Promise<GenerationStatusResponse> {
-  const response = await apiClient.get(`/templates/${templateId}/editor/generation-status`)
-  return response.data
-}
-
-/** 获取模板草稿关联的生成记录 */
-export async function getEditPlanGenerations(templateId: string): Promise<EditPlanGeneration[]> {
-  const response = await apiClient.get(`/templates/${templateId}/editor/generations`)
-  return response.data.items || []
-}
-
 /** 获取生成任务的视频结果列表 */
 export async function getGenerationTaskResults(taskId: string): Promise<GeneratedVideo[]> {
   const response = await apiClient.get(`/generation/tasks/${taskId}/results`)
   return response.data.items || response.data || []
 }
 
-/** 取消生成任务 */
-export async function cancelGeneration(templateId: string): Promise<void> {
-  await apiClient.post(`/templates/${templateId}/editor/cancel`)
+/** ── 草稿 clips 批量更新 ── */
+
+export interface EditPlanClipInput {
+  asset_id: string
+  start_time: number
+  duration: number
+  order: number
 }
 
-/** 复制模板草稿（含所有片段配置） */
-export async function copyEditPlan(
+/**
+ * 批量替换草稿的 clips（先全删再批量插入）
+ * 后端路由：PUT /templates/{template_id}/editor/clips
+ */
+export async function updateEditPlanClips(
   templateId: string,
-  data?: CopyEditPlanRequest,
-): Promise<EditPlan> {
-  const response = await apiClient.post<EditPlan>(
-    `/templates/${templateId}/editor/copy`,
-    data || {},
+  clips: EditPlanClipInput[],
+  signal?: AbortSignal,
+): Promise<{ count: number }> {
+  const response = await apiClient.put(
+    `/templates/${templateId}/editor/clips`,
+    { clips },
+    { signal },
   )
   return response.data
 }

apps/web/src/api/template-editor/index.ts

@@ -15,14 +15,8 @@ export type {
   EditPlanSegment,
   EditPlanConfig,
   EditPlan,
-  CreateEditPlanRequest,
   UpdateEditPlanRequest,
-  EditPlanListParams,
-  EditPlanListResponse,
-  GenerateResponse,
-  EditPlanGeneration,
   ClipStatusItem,
-  GenerationStatusResponse,
   GeneratedVideo,
   AIRecommendRequest,
   AIRecommendClipItem,
@@ -37,7 +31,6 @@ export type {
   ClipReorderResponse,
   ClipBatchDeleteResponse,
   ClipsFromAssetsResponse,
-  CopyEditPlanRequest,
   TransitionEffect,
   MediaAsset,
 } from "./types"
@@ -53,18 +46,12 @@ export {
 
 // 模板草稿 CRUD + 生成
 export {
-  getEditPlans,
   getEditPlan,
-  createEditPlan,
   updateEditPlan,
-  deleteEditPlan,
-  generateEditPlan,
-  getGenerationStatus,
-  getEditPlanGenerations,
+  updateEditPlanClips,
   getGenerationTaskResults,
-  cancelGeneration,
-  copyEditPlan,
 } from "./editPlans"
+export type { EditPlanClipInput } from "./editPlans"
 
 // 片段 CRUD + 批量操作
 export {

apps/web/src/api/template-editor/types.ts

@@ -114,10 +114,19 @@ export interface EditPlanConfig {
   auto_subtitles?: boolean
   /** 是否启用 BGM */
   bgm?: boolean
-  /** 生成数量 */
-  generate_count?: number
   /** 素材模式 */
   material_mode?: string
+  /** 前端标题设置（Step4 自动保存，与 title_config 字段分离，不影响后端渲染） */
+  title?: {
+    text?: string
+    font?: string
+    font_size?: number
+    color?: string
+    position?: string
+    bold?: boolean
+    stroke?: boolean
+    shadow?: boolean
+  }
   /** 预览视频 URL（封面生成用） */
   rendered_storage_key?: string
   /** 生成任务 ID */
@@ -176,31 +185,6 @@ export interface EditPlanListResponse {
 
 /* ── 生成相关 ── */
 
-/** 生成响应 */
-export interface GenerateResponse {
-  plan_id: string
-  plan_status: EditPlanStatus
-  generation_task_id: string
-  clip_count: number
-}
-
-/** 模板草稿关联的生成记录 */
-export interface EditPlanGeneration {
-  id: string
-  source_edit_plan_id: string
-  template_id: string
-  asset_ids: string[]
-  status: EditPlanStatus
-  progress: number
-  result_count: number
-  error_message: string
-  error_info: Record<string, unknown>
-  logs: Array<Record<string, unknown>>
-  retry_count: number
-  created_at?: string
-  updated_at?: string
-}
-
 /** 片段生成状态 */
 export interface ClipStatusItem {
   clip_id: string
@@ -213,17 +197,6 @@ export interface ClipStatusItem {
   error_message?: string
 }
 
-/** 生成状态轮询响应 */
-export interface GenerationStatusResponse {
-  plan_id: string
-  plan_status: EditPlanStatus
-  generation_task_id?: string
-  error_message?: string
-  clips: ClipStatusItem[]
-  error?: string
-  message?: string
-}
-
 /** 生成视频详情 */
 export interface GeneratedVideo {
   id: string

apps/web/src/api/templates/index.ts

@@ -9,8 +9,6 @@ export type {
   TemplateSegment,
   TemplateListParams,
   TemplateListResponse,
-  GenerateFromTemplateRequest,
-  GenerateFromTemplateResponse,
   CopyTemplateResponse,
 } from "./types"
 
@@ -24,5 +22,4 @@ export {
   getTemplate,
   toggleFavoriteTemplate,
   copyTemplate,
-  generateFromTemplate,
 } from "./templates"

apps/web/src/api/templates/templates.ts

@@ -5,8 +5,6 @@
 import apiClient from "../client"
 import type {
   CopyTemplateResponse,
-  GenerateFromTemplateRequest,
-  GenerateFromTemplateResponse,
   TemplateItem,
   TemplateListParams,
   TemplateListResponse,
@@ -45,15 +43,3 @@ export const copyTemplate = async (templateId: string): Promise<CopyTemplateResp
   const response = await apiClient.post<CopyTemplateResponse>(`/templates/${templateId}/copy`)
   return response.data
 }
-
-/** 从模板生成 */
-export const generateFromTemplate = async (
-  templateId: string,
-  data?: GenerateFromTemplateRequest,
-): Promise<GenerateFromTemplateResponse> => {
-  const response = await apiClient.post<GenerateFromTemplateResponse>(
-    `/templates/${templateId}/generate`,
-    data,
-  )
-  return response.data
-}

apps/web/src/components/asset-selector/AssetCard.tsx

@@ -59,8 +59,23 @@ const AssetCard: React.FC<AssetCardProps> = ({
     >
       {/* 缩略图 */}
       <div className="as-card-thumb">
-        {asset.thumbnail_url ? (
-          <img src={asset.thumbnail_url} alt={asset.name} loading="lazy" />
+        {asset.thumbnail_url && !asset.thumbnail_url.match(/\.(mp4|mov|avi|webm|mkv)(\?|$)/i) ? (
+          <img
+            src={asset.thumbnail_url}
+            alt={asset.name}
+            loading="lazy"
+            onError={(e) => {
+              // 图片加载失败时降级显示类型图标
+              const parent = (e.target as HTMLElement).parentElement
+              if (parent) {
+                ;(e.target as HTMLElement).style.display = "none"
+                const icon = document.createElement("span")
+                icon.className = "as-card-thumb-icon"
+                icon.textContent = MATERIAL_TYPE_ICONS[asset.type] || "🎬"
+                parent.appendChild(icon)
+              }
+            }}
+          />
         ) : (
           <span className="as-card-thumb-icon">{MATERIAL_TYPE_ICONS[asset.type]}</span>
         )}

apps/web/src/config/navigation.ts

@@ -71,7 +71,7 @@ export const NAV_ITEMS: NavItem[] = [
   },
   {
     key: "editing-planner",
-    label: "剪辑编辑器",
+    label: "剪辑模板",
     path: "/app/editing-planner",
     icon: React.createElement(EditOutlined),
   },
@@ -133,7 +133,7 @@ export const NAV_GROUPS: NavGroup[] = [
       },
       {
         key: "editing-planner",
-        label: "剪辑编辑器",
+        label: "剪辑模板",
         path: "/app/editing-planner",
         icon: React.createElement(EditOutlined),
       },

apps/web/src/pages/editing-planner/EditingPlanner.tsx

@@ -15,8 +15,6 @@ import React, { useState } from "react"
 import { useSearchParams } from "react-router-dom"
 import { MODE_LABELS } from "@/api/editing-planner"
 import { MODE_LIST } from "./constants"
-import type { MediaAsset } from "@/api/template-editor"
-
 import MediaPanel from "./components/MediaPanel"
 import PreviewPlayer from "./components/PreviewPlayer"
 import TimelinePanel from "./components/TimelinePanel"
@@ -79,14 +77,6 @@ const EditingPlanner: React.FC = () => {
   /* ── 右侧栏 Tab ── */
   const [rightTab, setRightTab] = useState<"properties" | "clips">("properties")
 
-  /* ── 素材库 ── */
-  const [mediaAssets, setMediaAssets] = useState<MediaAsset[]>([])
-  const [selectedAssetIds, setSelectedAssetIds] = useState<string[]>([])
-
-  const handleAssetSelect = (ids: string[]) => {
-    setSelectedAssetIds(ids)
-  }
-
   /* ── 配音素材 ── */
   const {
     voiceMaterials,
@@ -113,7 +103,6 @@ const EditingPlanner: React.FC = () => {
     resetClips,
     setClips,
     setSelectedClipId: clipOps.setSelectedClipId,
-    setMediaAssets,
     setTitleConfig,
     setSubtitleSettings,
     setBgmSettings,
@@ -164,9 +153,6 @@ const EditingPlanner: React.FC = () => {
           onLoadTemplate={tpl.handleLoadTemplate}
           onSearchChange={tpl.setSearchQuery}
           onFilterChange={tpl.setCurrentFilter}
-          mediaAssets={mediaAssets}
-          onAssetSelect={handleAssetSelect}
-          selectedAssetIds={selectedAssetIds}
         />
 
         {/* 中栏 flex-1 */}

apps/web/src/pages/editing-planner/components/GenerationHistoryModal.tsx

@@ -1,116 +0,0 @@
-/**
- * 生成历史弹窗 — 展示当前模板草稿的生成任务记录
- * 从 EditingPlanner 拆分，避免主文件过大
- */
-import React from "react"
-import { CloseOutlined, InboxOutlined } from "@ant-design/icons"
-import type { EditPlanGeneration } from "@/api/template-editor"
-import { PLAN_STATUS_LABELS } from "@/api/template-editor"
-
-interface GenerationHistoryModalProps {
-  open: boolean
-  loading: boolean
-  history: EditPlanGeneration[]
-  onClose: () => void
-  onCancel?: (taskId: string) => void
-  cancelLoading?: boolean
-}
-
-const GenerationHistoryModal: React.FC<GenerationHistoryModalProps> = ({
-  open,
-  loading,
-  history,
-  onClose,
-  onCancel,
-  cancelLoading,
-}) => {
-  if (!open) return null
-
-  return (
-    <div className="ep-modal-overlay" onClick={onClose}>
-      <div className="ep-modal ep-gh-modal" onClick={(e) => e.stopPropagation()}>
-        <div className="ep-modal-header">
-          <h3>生成历史</h3>
-          <button className="ep-modal-close" onClick={onClose}>
-            <CloseOutlined />
-          </button>
-        </div>
-        <div className="ep-modal-body ep-gh-body">
-          {loading ? (
-            <div className="ep-gh-empty">
-              <div className="ep-skeleton">
-                <div className="ep-skeleton-item ep-skeleton-item--header" />
-                <div className="ep-skeleton-item" />
-                <div className="ep-skeleton-item" />
-                <div className="ep-skeleton-item" />
-              </div>
-            </div>
-          ) : history.length === 0 ? (
-            <div className="ep-gh-empty">
-              <InboxOutlined style={{ fontSize: 32, opacity: 0.4 }} />
-              <span>暂无生成记录</span>
-            </div>
-          ) : (
-            <table className="ep-gh-table">
-              <thead>
-                <tr className="ep-gh-table-header-row">
-                  <th className="ep-gh-th">任务ID</th>
-                  <th className="ep-gh-th">状态</th>
-                  <th className="ep-gh-th">创建时间</th>
-                  <th className="ep-gh-th">更新时间</th>
-                  {onCancel && <th className="ep-gh-th">操作</th>}
-                </tr>
-              </thead>
-              <tbody>
-                {history.map((gen) => {
-                  const statusClass = `ep-gh-status-tag--${gen.status}`
-                  const canCancel = gen.status === "rendering" || gen.status === "editing"
-                  return (
-                    <tr key={gen.id} className="ep-gh-table-row">
-                      <td className="ep-gh-td ep-gh-td-id">
-                        {gen.id ? `${gen.id.slice(0, 8)}...` : "—"}
-                      </td>
-                      <td className="ep-gh-td">
-                        <span className={`ep-gh-status-tag ${statusClass}`}>
-                          {PLAN_STATUS_LABELS[gen.status] || gen.status}
-                        </span>
-                      </td>
-                      <td className="ep-gh-td ep-gh-td-time">
-                        {gen.created_at ? new Date(gen.created_at).toLocaleString("zh-CN") : "—"}
-                      </td>
-                      <td className="ep-gh-td ep-gh-td-time">
-                        {gen.updated_at ? new Date(gen.updated_at).toLocaleString("zh-CN") : "—"}
-                      </td>
-                      {onCancel && (
-                        <td className="ep-gh-td ep-gh-td-action">
-                          {canCancel ? (
-                            <button
-                              className="ep-gh-cancel-btn"
-                              onClick={() => onCancel(gen.id)}
-                              disabled={cancelLoading}
-                            >
-                              取消
-                            </button>
-                          ) : (
-                            <span className="ep-gh-action-placeholder">—</span>
-                          )}
-                        </td>
-                      )}
-                    </tr>
-                  )
-                })}
-              </tbody>
-            </table>
-          )}
-        </div>
-        <div className="ep-modal-footer">
-          <button className="ep-btn ep-btn-secondary" onClick={onClose}>
-            关闭
-          </button>
-        </div>
-      </div>
-    </div>
-  )
-}
-
-export default GenerationHistoryModal

apps/web/src/pages/editing-planner/components/GenerationProgressModal.tsx

@@ -1,6 +0,0 @@
-/**
- * 生成进度弹窗 — 入口文件（向后兼容）
- * 实际实现已移至 ./generation-progress-modal/ 目录
- */
-export { default } from "./generation-progress-modal"
-export type { GenPhase, GenerationProgressModalProps } from "./generation-progress-modal"

apps/web/src/pages/editing-planner/components/MediaPanel.tsx

@@ -1,12 +1,10 @@
 /**
- * 左侧面板 — V8 原型 1:1 还原
- * Tab 切换：模板列表 + 素材库
+ * 左侧面板 — 模板列表
+ * 模板编辑器只负责定义模板规则（片段数量、时长范围），不承载素材管理。
  */
-import React, { useState } from "react"
+import React from "react"
 import type { EditingTemplate } from "@/api/editing-planner"
 import { MODE_LABELS } from "@/api/editing-planner"
-import type { MediaAsset } from "@/api/template-editor"
-import AssetSelector from "@/components/asset-selector/AssetSelector"
 
 interface MediaPanelProps {
   templates: EditingTemplate[]
@@ -18,10 +16,6 @@ interface MediaPanelProps {
   onLoadTemplate: (id: string) => void
   onSearchChange: (q: string) => void
   onFilterChange: (f: string) => void
-  // 素材相关
-  mediaAssets?: MediaAsset[]
-  onAssetSelect?: (ids: string[]) => void
-  selectedAssetIds?: string[]
 }
 
 const MediaPanel: React.FC<MediaPanelProps> = ({
@@ -34,113 +28,73 @@ const MediaPanel: React.FC<MediaPanelProps> = ({
   onLoadTemplate,
   onSearchChange,
   onFilterChange,
-  mediaAssets = [],
-  onAssetSelect,
-  selectedAssetIds = [],
 }) => {
-  const [activeTab, setActiveTab] = useState<"templates" | "assets">("templates")
-
   return (
     <div className="ep-left-panel">
-      {/* Tab 切换 */}
-      <div className="ep-left-tabs">
-        <button
-          className={`ep-left-tab ${activeTab === "templates" ? "active" : ""}`}
-          onClick={() => setActiveTab("templates")}
-        >
-          📋 模板
-        </button>
-        <button
-          className={`ep-left-tab ${activeTab === "assets" ? "active" : ""}`}
-          onClick={() => setActiveTab("assets")}
-        >
-          📁 素材
-        </button>
+      {/* 搜索 */}
+      <div className="ep-search-wrap ep-media-panel-inner">
+        <span className="ep-search-icon">🔍</span>
+        <input
+          className="ep-search-input"
+          placeholder="搜索模板..."
+          value={searchQuery}
+          onChange={(e) => onSearchChange(e.target.value)}
+        />
       </div>
 
-      {/* 模板 Tab */}
-      {activeTab === "templates" && (
-        <>
-          {/* 搜索 */}
-          <div className="ep-search-wrap ep-media-panel-inner">
-            <span className="ep-search-icon">🔍</span>
-            <input
-              className="ep-search-input"
-              placeholder="搜索模板..."
-              value={searchQuery}
-              onChange={(e) => onSearchChange(e.target.value)}
-            />
-          </div>
+      {/* Chip 分类筛选 */}
+      <div className="ep-filter-chips">
+        {filterCategories.map((cat) => (
+          <button
+            key={cat}
+            className={`ep-filter-chip ${currentFilter === cat ? "active" : ""}`}
+            onClick={() => onFilterChange(cat)}
+          >
+            {cat}
+          </button>
+        ))}
+      </div>
 
-          {/* Chip 分类筛选 */}
-          <div className="ep-filter-chips">
-            {filterCategories.map((cat) => (
-              <button
-                key={cat}
-                className={`ep-filter-chip ${currentFilter === cat ? "active" : ""}`}
-                onClick={() => onFilterChange(cat)}
-              >
-                {cat}
-              </button>
-            ))}
+      {/* 模板列表 */}
+      <div className="ep-template-list">
+        {loading ? (
+          <div className="ep-loading">
+            <span>⏳</span>
+            <span>加载中...</span>
           </div>
-
-          {/* 模板列表 */}
-          <div className="ep-template-list">
-            {loading ? (
-              <div className="ep-loading">
-                <span>⏳</span>
-                <span>加载中...</span>
+        ) : templates.length === 0 ? (
+          <div className="ep-empty">
+            <span>📭</span>
+            <span>暂无模板</span>
+          </div>
+        ) : (
+          templates.map((tpl) => (
+            <div
+              key={tpl.id}
+              className={`ep-template-card ${loadedTemplateId === tpl.id ? "active" : ""}`}
+              onClick={() => onLoadTemplate(tpl.id)}
+            >
+              <div className="ep-template-card-header">
+                <span className="ep-template-card-name">{tpl.name}</span>
+                <span className="ep-template-card-mode">{MODE_LABELS[tpl.mode]}</span>
               </div>
-            ) : templates.length === 0 ? (
-              <div className="ep-empty">
-                <span>📭</span>
-                <span>暂无模板</span>
+              <div className="ep-template-card-meta">
+                <span>⏱️ {tpl.estimated_duration}s</span>
+                <span>📐 {tpl.segments.length}片段</span>
               </div>
-            ) : (
-              templates.map((tpl) => (
-                <div
-                  key={tpl.id}
-                  className={`ep-template-card ${loadedTemplateId === tpl.id ? "active" : ""}`}
-                  onClick={() => onLoadTemplate(tpl.id)}
-                >
-                  <div className="ep-template-card-header">
-                    <span className="ep-template-card-name">{tpl.name}</span>
-                    <span className="ep-template-card-mode">{MODE_LABELS[tpl.mode]}</span>
-                  </div>
-                  <div className="ep-template-card-meta">
-                    <span>⏱️ {tpl.estimated_duration}s</span>
-                    <span>📐 {tpl.segments.length}片段</span>
-                  </div>
-                  {tpl.tags.length > 0 && (
-                    <div className="ep-template-card-tags">
-                      {tpl.tags.map((tag) => (
-                        <span key={tag} className="ep-template-tag">
-                          {tag}
-                        </span>
-                      ))}
-                    </div>
-                  )}
+              {tpl.tags.length > 0 && (
+                <div className="ep-template-card-tags">
+                  {tpl.tags.map((tag) => (
+                    <span key={tag} className="ep-template-tag">
+                      {tag}
+                    </span>
+                  ))}
                 </div>
-              ))
-            )}
-          </div>
-        </>
-      )}
-
-      {/* 素材 Tab */}
-      {activeTab === "assets" && (
-        <div className="ep-assets-tab">
-          <AssetSelector
-            assets={mediaAssets}
-            selectedIds={selectedAssetIds}
-            onSelectionChange={onAssetSelect}
-            showQualityFilter={false}
-            showBatchSelect={false}
-            compact
-          />
-        </div>
-      )}
+              )}
+            </div>
+          ))
+        )}
+      </div>
     </div>
   )
 }

apps/web/src/pages/editing-planner/components/bgm-selector/index.tsx

@@ -16,9 +16,17 @@ interface BgmSelectorProps {
   onClose: () => void
   config: BgmMixConfig
   onChange: (config: BgmMixConfig) => void
+  /** 模板/草稿 ID，用于请求 BGM 预设 */
+  templateId?: string
 }
 
-const BgmSelector: React.FC<BgmSelectorProps> = ({ open, onClose, config, onChange }) => {
+const BgmSelector: React.FC<BgmSelectorProps> = ({
+  open,
+  onClose,
+  config,
+  onChange,
+  templateId,
+}) => {
   const {
     presets,
     loading,
@@ -30,7 +38,7 @@ const BgmSelector: React.FC<BgmSelectorProps> = ({ open, onClose, config, onChan
     loadPresets,
     handlePreview,
     stopPreview,
-  } = useBgmSelector(open)
+  } = useBgmSelector(open, templateId)
 
   /* ── 选中 BGM ── */
   const handleSelect = useCallback(

apps/web/src/pages/editing-planner/components/bgm-selector/useBgmSelector.ts

@@ -19,7 +19,7 @@ export const CATEGORY_LIST: {
  * BGM 选择器数据与交互 Hook
  * 封装列表加载、搜索、分类筛选、试听播放逻辑
  */
-export function useBgmSelector(open: boolean) {
+export function useBgmSelector(open: boolean, templateId?: string) {
   const [presets, setPresets] = useState<BgmPreset[]>([])
   const [loading, setLoading] = useState(false)
   const [activeCategory, setActiveCategory] = useState<BgmCategory | "all">("all")
@@ -30,19 +30,23 @@ export function useBgmSelector(open: boolean) {
 
   /* ── 加载 BGM 列表 ── */
   const loadPresets = useCallback(async () => {
+    if (!templateId) {
+      setPresets([])
+      return
+    }
     setLoading(true)
     try {
       const params: { category?: string; keyword?: string } = {}
       if (activeCategory !== "all") params.category = activeCategory
       if (keyword.trim()) params.keyword = keyword.trim()
-      const data = await getBgmPresets(params)
+      const data = await getBgmPresets(templateId, params)
       setPresets(data)
     } catch {
       message.error("加载 BGM 列表失败")
     } finally {
       setLoading(false)
     }
-  }, [activeCategory, keyword])
+  }, [activeCategory, keyword, templateId])
 
   useEffect(() => {
     if (open) loadPresets()

apps/web/src/pages/editing-planner/components/generation-progress-modal/ProgressPhase.tsx

@@ -1,65 +0,0 @@
-import React from "react"
-import { Modal } from "@/components/ui"
-import type { TaskItem } from "@/api/tasks"
-import { getStepLabel, getStatusColor } from "./constants"
-
-interface ProgressPhaseProps {
-  open: boolean
-  task: TaskItem | null
-  onCancel: () => void
-}
-
-/** progress（进度轮询）阶段弹窗 */
-export const ProgressPhase: React.FC<ProgressPhaseProps> = ({ open, task, onCancel }) => {
-  const progress = task?.progress ?? 0
-  const status = task?.status ?? ""
-  const currentStep = task?.current_step ?? ""
-  const userMessage = task?.user_message ?? ""
-  const stepColor = getStatusColor(status, currentStep)
-
-  return (
-    <Modal open={open} title="视频生成中" footer={null} onCancel={onCancel} closable width={480}>
-      <div className="ep-gen-progress">
-        {/* 进度环 */}
-        <div className="ep-gen-progress-ring-wrap">
-          <svg className="ep-gen-progress-ring" viewBox="0 0 120 120">
-            <circle className="ep-gen-progress-ring-bg" cx="60" cy="60" r="52" />
-            <circle
-              className="ep-gen-progress-ring-fill"
-              cx="60"
-              cy="60"
-              r="52"
-              style={{
-                strokeDasharray: `${2 * Math.PI * 52}`,
-                strokeDashoffset: `${2 * Math.PI * 52 * (1 - progress / 100)}`,
-                stroke: stepColor,
-              }}
-            />
-          </svg>
-          <span className="ep-gen-progress-pct" style={{ color: stepColor }}>
-            {progress}%
-          </span>
-        </div>
-
-        {/* 当前步骤 */}
-        <div className="ep-gen-step-text">
-          {userMessage || getStepLabel(currentStep) || "处理中…"}
-        </div>
-
-        {/* 进度条 */}
-        <div className="ep-gen-progress-bar">
-          <div
-            className="ep-gen-progress-bar-fill"
-            style={{
-              width: `${progress}%`,
-              backgroundColor: stepColor,
-            }}
-          />
-        </div>
-
-        {/* 任务 ID */}
-        {task?.id && <div className="ep-gen-task-id">任务 ID: {task.id}</div>}
-      </div>
-    </Modal>
-  )
-}

apps/web/src/pages/editing-planner/components/generation-progress-modal/ResultPhase.tsx

@@ -1,82 +0,0 @@
-import React from "react"
-import { Modal, Button } from "@/components/ui"
-import type { TaskItem } from "@/api/tasks"
-
-interface ResultPhaseProps {
-  open: boolean
-  phase: "completed" | "failed"
-  task: TaskItem | null
-  onCancel: () => void
-  onRetry?: () => void
-  onClose?: () => void
-}
-
-/** completed / failed（结果）阶段弹窗 */
-export const ResultPhase: React.FC<ResultPhaseProps> = ({
-  open,
-  phase,
-  task,
-  onCancel,
-  onRetry,
-  onClose,
-}) => {
-  const userMessage = task?.user_message ?? ""
-  const errorMessage = task?.error_message ?? ""
-  const retryable = task?.retryable ?? false
-  const handleClose = onClose || onCancel
-
-  if (phase === "completed") {
-    return (
-      <Modal
-        open={open}
-        title="✅ 生成完成"
-        footer={null}
-        onCancel={handleClose}
-        closable
-        width={440}
-      >
-        <div className="ep-gen-result">
-          <div className="ep-gen-result-icon">🎉</div>
-          <div className="ep-gen-result-title">视频生成完成！</div>
-          {userMessage && <div className="ep-gen-result-msg">{userMessage}</div>}
-          <div className="ep-gen-result-actions">
-            <Button buttonType="primary" onClick={handleClose}>
-              查看结果
-            </Button>
-          </div>
-        </div>
-      </Modal>
-    )
-  }
-
-  return (
-    <Modal
-      open={open}
-      title="❌ 生成失败"
-      footer={null}
-      onCancel={handleClose}
-      closable
-      width={440}
-    >
-      <div className="ep-gen-result ep-gen-result--error">
-        <div className="ep-gen-result-icon">😥</div>
-        <div className="ep-gen-result-title">视频生成失败</div>
-        {(errorMessage || userMessage) && (
-          <div className="ep-gen-result-msg ep-gen-result-msg--error">
-            {errorMessage || userMessage}
-          </div>
-        )}
-        <div className="ep-gen-result-actions">
-          {retryable && onRetry && (
-            <Button buttonType="primary" onClick={onRetry}>
-              🔄 重试
-            </Button>
-          )}
-          <Button buttonType="secondary" onClick={handleClose}>
-            关闭
-          </Button>
-        </div>
-      </div>
-    </Modal>
-  )
-}

apps/web/src/pages/editing-planner/components/generation-progress-modal/SetupPhase.tsx

@@ -1,53 +0,0 @@
-import React from "react"
-import { Modal } from "@/components/ui"
-
-interface SetupPhaseProps {
-  open: boolean
-  voiceoverDuration: number | null
-  estimatedDuration: number
-  submitting: boolean
-  onDurationChange: (v: number | null) => void
-  onGenerate: () => void
-  onCancel: () => void
-}
-
-/** setup（配置）阶段弹窗 */
-export const SetupPhase: React.FC<SetupPhaseProps> = ({
-  open,
-  voiceoverDuration,
-  estimatedDuration,
-  submitting,
-  onDurationChange,
-  onGenerate,
-  onCancel,
-}) => (
-  <Modal
-    open={open}
-    title="使用模板生成视频"
-    confirmLoading={submitting}
-    onOk={onGenerate}
-    onCancel={onCancel}
-    okText="开始生成"
-    cancelText="取消"
-    width={440}
-  >
-    <div className="ep-gen-setup">
-      <label className="ep-gen-field-label">配音时长（秒）</label>
-      <input
-        type="number"
-        className="ep-gen-duration-input"
-        placeholder="请输入配音时长"
-        value={voiceoverDuration ?? ""}
-        onChange={(e) => {
-          const v = e.target.value ? Number(e.target.value) : null
-          onDurationChange(v)
-        }}
-        min={1}
-        max={600}
-      />
-      <div className="ep-gen-estimate">
-        预估总时长：<strong>{estimatedDuration}s</strong>
-      </div>
-    </div>
-  </Modal>
-)

apps/web/src/pages/editing-planner/components/generation-progress-modal/constants.ts

@@ -1,31 +0,0 @@
-/* ──────────── 步骤文案映射 ──────────── */
-
-export const STEP_LABELS: Record<string, string> = {
-  queued: "排队中…",
-  preparing: "准备素材…",
-  generating_video: "渲染视频中…",
-  adding_effects: "添加特效…",
-  composing: "合成中…",
-  encoding: "编码输出中…",
-  completed: "生成完成！",
-  failed: "生成失败",
-}
-
-export const getStepLabel = (step: string) => STEP_LABELS[step] || step.replace(/_/g, " ")
-
-/* ──────────── 状态徽标颜色 ──────────── */
-
-export const STATUS_COLOR: Record<string, string> = {
-  queued: "#6b7280",
-  pending: "#6b7280",
-  preparing: "#f59e0b",
-  generating_video: "#4f46e5",
-  adding_effects: "#7c3aed",
-  composing: "#2563eb",
-  encoding: "#0891b2",
-  completed: "#10b981",
-  failed: "#ef4444",
-}
-
-export const getStatusColor = (status: string, currentStep: string) =>
-  STATUS_COLOR[status] || STATUS_COLOR[currentStep] || "#4f46e5"

apps/web/src/pages/editing-planner/components/generation-progress-modal/index.tsx

@@ -1,70 +0,0 @@
-/**
- * 生成进度弹窗 — 任务 2.17
- * 三阶段 UI：setup（配置）→ progress（进度轮询）→ completed / failed（结果）
- * V21 设计系统，CSS 类名前缀 ep-gen-
- */
-import React, { useEffect, useRef } from "react"
-import type { GenPhase, GenerationProgressModalProps } from "./types"
-import { SetupPhase } from "./SetupPhase"
-import { ProgressPhase } from "./ProgressPhase"
-import { ResultPhase } from "./ResultPhase"
-
-/* 重新导出类型，保持向后兼容 */
-export type { GenPhase, GenerationProgressModalProps }
-
-const GenerationProgressModal: React.FC<GenerationProgressModalProps> = ({
-  open,
-  phase,
-  voiceoverDuration,
-  estimatedDuration,
-  onDurationChange,
-  onGenerate,
-  task,
-  submitting,
-  onCancel,
-  onRetry,
-  onClose,
-}) => {
-  /* 关闭弹窗时重置（避免下次打开残留旧状态） */
-  const prevOpen = useRef(false)
-  useEffect(() => {
-    if (prevOpen.current && !open) {
-      /* modal just closed — parent handles reset */
-    }
-    prevOpen.current = open
-  }, [open])
-
-  /* setup 阶段 */
-  if (phase === "setup") {
-    return (
-      <SetupPhase
-        open={open}
-        voiceoverDuration={voiceoverDuration}
-        estimatedDuration={estimatedDuration}
-        submitting={submitting}
-        onDurationChange={onDurationChange}
-        onGenerate={onGenerate}
-        onCancel={onCancel}
-      />
-    )
-  }
-
-  /* progress 阶段 */
-  if (phase === "progress") {
-    return <ProgressPhase open={open} task={task} onCancel={onCancel} />
-  }
-
-  /* completed / failed 阶段 */
-  return (
-    <ResultPhase
-      open={open}
-      phase={phase as "completed" | "failed"}
-      task={task}
-      onCancel={onCancel}
-      onRetry={onRetry}
-      onClose={onClose}
-    />
-  )
-}
-
-export default GenerationProgressModal

apps/web/src/pages/editing-planner/components/generation-progress-modal/types.ts

@@ -1,23 +0,0 @@
-import type { TaskItem } from "@/api/tasks"
-
-export type GenPhase = "setup" | "progress" | "completed" | "failed"
-
-export interface GenerationProgressModalProps {
-  open: boolean
-  phase: GenPhase
-
-  /* setup 阶段 */
-  voiceoverDuration: number | null
-  estimatedDuration: number
-  onDurationChange: (v: number | null) => void
-  onGenerate: () => void
-
-  /* progress / 结果阶段 */
-  task: TaskItem | null
-
-  /* 通用 */
-  submitting: boolean
-  onCancel: () => void
-  onRetry?: () => void
-  onClose?: () => void
-}

apps/web/src/pages/editing-planner/hooks/template-management/useTemplateList.ts

@@ -6,16 +6,13 @@ import {
   type EditingTemplate,
   type TemplateCategory,
 } from "@/api/editing-planner"
-import { getMediaAssets, type MediaAsset } from "@/api/template-editor"
 import { FILTER_CATEGORIES } from "../../constants"
 
 /**
  * 模板列表 + 分类 + 筛选搜索
+ * 模板编辑器只负责模板规则定义，不再加载/管理业务素材。
  */
-export function useTemplateList(
-  setMediaAssets: (assets: MediaAsset[]) => void,
-  initialTemplateId: string | null,
-) {
+export function useTemplateList(initialTemplateId: string | null) {
   const [templates, setTemplates] = useState<EditingTemplate[]>([])
   const [categories, setCategories] = useState<TemplateCategory[]>([])
   const [loadingTemplates, setLoadingTemplates] = useState(false)
@@ -24,26 +21,20 @@ export function useTemplateList(
   const [loadedTemplateId, setLoadedTemplateId] = useState<string | null>(initialTemplateId)
 
   /**
-   * 并行加载模板列表、分类、素材库
-   * 三个接口无依赖关系，用 Promise.all 并发
+   * 并行加载模板列表和分类（两者无依赖关系）
    */
   const loadTemplates = useCallback(async () => {
     setLoadingTemplates(true)
     try {
-      const [tpls, cats, assets] = await Promise.all([
-        getEditingTemplates(),
-        getTemplateCategories(),
-        getMediaAssets(),
-      ])
+      const [tpls, cats] = await Promise.all([getEditingTemplates(), getTemplateCategories()])
       setTemplates(tpls)
       setCategories(cats)
-      setMediaAssets(assets)
     } catch {
       message.error("加载模板失败")
     } finally {
       setLoadingTemplates(false)
     }
-  }, [setMediaAssets])
+  }, [])
 
   useEffect(() => {
     loadTemplates()

apps/web/src/pages/editing-planner/hooks/useTemplateManagement.ts

@@ -1,6 +1,6 @@
 import { useState, useCallback, type Dispatch, type SetStateAction } from "react"
 import type { TemplateMode } from "@/api/editing-planner"
-import type { MediaAsset, TitleConfig } from "@/api/template-editor"
+import type { TitleConfig } from "@/api/template-editor"
 import type {
   ClipData,
   WatermarkConfig,
@@ -24,7 +24,6 @@ interface UseTemplateManagementParams {
   resetClips: (clips: ClipData[]) => void
   setClips: (updater: (prev: ClipData[]) => ClipData[]) => void
   setSelectedClipId: (id: string | null) => void
-  setMediaAssets: (assets: MediaAsset[]) => void
   setTitleConfig: Dispatch<SetStateAction<TitleConfig>>
   setSubtitleSettings: Dispatch<SetStateAction<SubtitleStyleConfig>>
   setBgmSettings: Dispatch<SetStateAction<BgmMixConfig>>
@@ -52,7 +51,6 @@ export const useTemplateManagement = (params: UseTemplateManagementParams) => {
     resetClips,
     setClips,
     setSelectedClipId,
-    setMediaAssets,
     setTitleConfig,
     setSubtitleSettings,
     setBgmSettings,
@@ -86,7 +84,7 @@ export const useTemplateManagement = (params: UseTemplateManagementParams) => {
     filteredTemplates,
     currentTemplate,
     loadTemplates,
-  } = useTemplateList(setMediaAssets, urlTemplateId || null)
+  } = useTemplateList(urlTemplateId || null)
 
   /* ── 保存 ── */
   const {

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

@@ -1,30 +1,27 @@
 /**
- * 智能剪辑页面 — V24 前端预览播放器架构改造
- * 7 步向导：选择模板 → 素材 → 配音 → 标题 → 预览 → 封面 → 确认生成
+ * 智能剪辑页面 — 前端实时预览架构
+ * 6 步向导：选择模板 → 素材 → 配音 → 标题(含预览) → 封面 → 确认生成
  * 左右布局：左侧 generate-form + 右侧 generate-preview
  *
- * 架构改造：
- * - Step5 预览改为前端素材切片播放（FrontendPreviewPlayer）
- * - 完全去除后端 FFmpeg 预览依赖
- * - 标题样式通过 CSS 层实时叠加，所见即所得
- * - 最终成片仍走后端 FFmpeg 渲染（Step7 确认生成）
+ * 架构：
+ * - 步骤 4-5 右侧显示 FrontendPreviewPlayer 实时预览
+ * - 步骤 6 右侧内联播放生成的最终视频
+ * - 点"确认生成"时调用 createGenerationTask 创建一次服务器渲染任务
  */
-import React, { useMemo } from "react"
-import { Modal, message } from "antd"
+import React, { useMemo, useState, useEffect, useRef } from "react"
+import { message } from "antd"
 import { useNavigate } from "react-router-dom"
 import type { VoiceClone } from "@/api/voice-clone"
 import { useQuery } from "@tanstack/react-query"
 import { useCloneProgress } from "@/hooks/useCloneProgress"
-import { getAssetsByKind } from "@/api/assets"
 import CloneModal from "@/components/voice/CloneModal"
 import GenerateHeader from "./components/GenerateHeader"
 import {
   calculateTotalVideoDuration,
   estimateTotalVideoDuration,
 } from "./utils/calculateTotalVideoDuration"
+import FrontendPreviewPlayer from "./components/FrontendPreviewPlayer"
 import GenerateStepsBar from "./components/GenerateStepsBar"
-import GenerateResultPanel from "./components/GenerateResultPanel"
-import PreviewVideoPanel from "./components/PreviewVideoPanel"
 import GenerateStepContent from "./components/GenerateStepContent"
 import GenerateStepActions from "./components/GenerateStepActions"
 import { useGenerateFormState } from "./hooks/useGenerateFormState"
@@ -32,6 +29,8 @@ import { useStepNavigation } from "./hooks/useStepNavigation"
 import { useGenerateVideo } from "./hooks/useGenerateVideo"
 import { usePreviewAssets } from "./hooks/usePreviewAssets"
 import { useTitleStyleUpdaters } from "./hooks/useStep4Title/useTitleStyleUpdaters"
+import { getAssetsByKind } from "@/api/assets"
+import { previewTts } from "@/api/tts"
 import "./generate.css"
 
 const GeneratePage: React.FC = () => {
@@ -64,26 +63,74 @@ const GeneratePage: React.FC = () => {
     presetVoices,
     cloneModalOpen,
     setCloneModalOpen,
-    generateCount,
-    setGenerateCount,
     videoRatio,
     duration,
     style,
     autoSubtitles,
     bgm,
     editPlanId,
-    previewVideo,
-    setPreviewVideo,
-    previewModalOpen,
-    setPreviewModalOpen,
+    sourceEditPlanId,
+    previewTaskId,
+    setPreviewTaskId,
+    storedSourceEditPlanId,
+    setStoredSourceEditPlanId,
+    serverClips,
+    setServerClips,
   } = formState
 
-  /* ── 标题样式回调（Step5 样式面板 + 右侧预览 CSS 层共用） ── */
+  /* ── 标题样式回调 ── */
   const styleUpdaters = useTitleStyleUpdaters({
     titleSettings,
     onTitleSettingsChange: setTitleSettings,
   })
 
+  /* ── 配音素材库（TTS 试听）── */
+  const { data: voiceMaterials = [] } = useQuery({
+    queryKey: ["assets", "voice"],
+    queryFn: () => getAssetsByKind("voice", { limit: 50 }),
+  })
+
+  const [previewVoiceAudioUrl, setPreviewVoiceAudioUrl] = useState<string | null>(null)
+  const ttsAbortRef = useRef<AbortController | null>(null)
+
+  useEffect(() => {
+    const voiceAsset = voiceMaterials.find((m) => m.id === selectedVoice)
+    if (voiceAsset?.file_url) {
+      setPreviewVoiceAudioUrl(voiceAsset.file_url)
+      return
+    }
+
+    const voiceId = selectedClonedVoice || selectedVoice
+    if (!voiceId || !titleSettings.title) {
+      setPreviewVoiceAudioUrl(null)
+      return
+    }
+
+    ttsAbortRef.current?.abort()
+    const controller = new AbortController()
+    ttsAbortRef.current = controller
+    let cancelled = false
+
+    previewTts({ text: titleSettings.title, voice_id: voiceId })
+      .then((res) => {
+        if (!cancelled && res.audio_url) {
+          setPreviewVoiceAudioUrl(res.audio_url)
+        }
+      })
+      .catch((err) => {
+        if (!cancelled) {
+          console.warn("[预览配音生成失败]", err)
+          setPreviewVoiceAudioUrl(null)
+        }
+      })
+
+    return () => {
+      cancelled = true
+      controller.abort()
+    }
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [selectedVoice, selectedClonedVoice, titleSettings.title, voiceMaterials])
+
   /* ── 克隆声音 ── */
   const { clones: clonedVoices, addClone, hasProcessing } = useCloneProgress()
 
@@ -93,44 +140,47 @@ const GeneratePage: React.FC = () => {
     message.success("音色克隆成功！")
   }
 
-  /* ── 前端预览：加载选中素材的视频文件信息 ── */
+  /* ── 素材 ID 列表 ── */
   const previewAssetIds = useMemo(
     () => (materialMode === "auto" ? smartSelectedIds : selectedMaterials),
     [materialMode, smartSelectedIds, selectedMaterials],
   )
-  const previewAssetsEnabled = currentStep >= 4 && previewAssetIds.length > 0
-  const {
-    assets: previewAssets,
-    loading: previewAssetsLoading,
-    ready: previewAssetsReady,
-  } = usePreviewAssets(previewAssetIds, previewAssetsEnabled)
 
-  /* ── 当前模板对象（传给前端预览播放器） ── */
+  /* ── 当前模板对象 ── */
   const currentTemplate = useMemo(
     () => userTemplates.find((t) => t.id === selectedTemplate) || null,
     [userTemplates, selectedTemplate],
   )
 
-  /* ── 视频总时长计算（用于配音时长校验） ── */
+  /* ── BGM 配置 ── */
+  const bgmConfig = useMemo(
+    () => ({
+      enabled: bgm,
+      music_id: currentTemplate?.bgm_config?.music_id || "",
+    }),
+    [bgm, currentTemplate],
+  )
+
+  /* ── 加载素材详情（供前端预览播放器使用 + 配音时长校验） ── */
+  const previewAssetsEnabled = previewAssetIds.length > 0
+  const { assets: previewAssets, ready: previewAssetsReady } = usePreviewAssets(
+    previewAssetIds,
+    previewAssetsEnabled,
+  )
+
+  /* ── 预览就绪：素材已加载，且有模板 ── */
+  const previewReady = useMemo(
+    () => previewAssetsReady && !!currentTemplate,
+    [previewAssetsReady, currentTemplate],
+  )
+
+  /* ── 视频总时长计算 ── */
   const totalVideoDuration = useMemo(() => {
-    // 优先用素材精确时长；素材未加载时用模板 segments 的 duration_max 之和估算
     const exact = calculateTotalVideoDuration(previewAssets, currentTemplate ?? undefined)
     if (exact > 0) return exact
     return estimateTotalVideoDuration(currentTemplate ?? undefined)
   }, [previewAssets, currentTemplate])
 
-  /* ── 配音音频 URL ── */
-  const { data: voiceMaterials = [] } = useQuery({
-    queryKey: ["assets", "voice"],
-    queryFn: () => getAssetsByKind("voice", { limit: 50 }),
-  })
-
-  const voiceAudioUrl = useMemo(() => {
-    if (!selectedVoice) return undefined
-    const asset = voiceMaterials.find((v) => v.id === selectedVoice)
-    return asset?.file_url || undefined
-  }, [selectedVoice, voiceMaterials])
-
   /* ── 步骤导航 ── */
   const { goNext, goPrev } = useStepNavigation({
     currentStep,
@@ -140,7 +190,7 @@ const GeneratePage: React.FC = () => {
     selectedMaterials,
     smartSelectedIds,
     titleSettings,
-    previewReady: previewAssetsReady,
+    previewReady,
   })
 
   /* ── 视频生成核心逻辑 ── */
@@ -170,23 +220,26 @@ const GeneratePage: React.FC = () => {
     duration,
     autoSubtitles,
     bgm,
-    generateCount,
+    sourceEditPlanId: storedSourceEditPlanId || sourceEditPlanId,
+    previewTaskId,
+    bgmConfig,
+    onGenerationSuccess: () => {
+      setPreviewTaskId(null)
+      setStoredSourceEditPlanId(null)
+    },
   })
 
   /* ================================================================
-     渲染 — 主页面
+     渲染
      ================================================================ */
 
   return (
     <div className="xx-generate-page">
-      {/* ── 页头 ── */}
       <GenerateHeader fromEditPlan={!!editPlanId} />
 
-      {/* ── 步骤条 ── */}
       <GenerateStepsBar currentStep={currentStep} onStepClick={setCurrentStep} />
 
-      {/* ── 主布局 ── */}
-      <div className="xx-generate-layout">
+      <div className={`xx-generate-layout${currentStep < 4 ? " full-width" : ""}`}>
         {/* ════ 左侧：表单区 ════ */}
         <div className="xx-generate-form">
           <GenerateStepContent
@@ -202,7 +255,6 @@ const GeneratePage: React.FC = () => {
             onSmartSelectedIdsChange={setSmartSelectedIds}
             titleSettings={titleSettings}
             onTitleSettingsChange={setTitleSettings}
-            /* 标题样式回调 */
             onUpdatePosition={styleUpdaters.updatePosition}
             onUpdateFont={styleUpdaters.updateFont}
             onUpdateSize={styleUpdaters.updateSize}
@@ -213,12 +265,17 @@ const GeneratePage: React.FC = () => {
             onApplyPreset={styleUpdaters.applyPreset}
             activePreset={styleUpdaters.activePreset}
             titlePresets={styleUpdaters.titlePresets}
+            onPreviewTaskCreated={setPreviewTaskId}
+            onSourceEditPlanIdExtracted={setStoredSourceEditPlanId}
+            bgm={bgm}
+            bgmConfig={bgmConfig}
             coverSettings={coverSettings}
             onCoverSettingsChange={setCoverSettings}
             duration={duration}
             selectedVoice={selectedVoice}
             onSelectedVoiceChange={setSelectedVoice}
             totalVideoDuration={totalVideoDuration}
+            onServerClipsChange={setServerClips}
             voiceMode={voiceMode}
             onVoiceModeChange={setVoiceMode}
             selectedClonedVoice={selectedClonedVoice}
@@ -228,8 +285,6 @@ const GeneratePage: React.FC = () => {
             hasProcessing={hasProcessing}
             cloneModalOpen={cloneModalOpen}
             onCloneModalOpenChange={setCloneModalOpen}
-            generateCount={generateCount}
-            onGenerateCountChange={setGenerateCount}
             generating={generating}
             generated={generated}
             generateError={generateError}
@@ -251,63 +306,58 @@ const GeneratePage: React.FC = () => {
           />
         </div>
 
-        {/* ════ 右侧：预览 + 生成结果 ════ */}
+        {/* ════ 右侧：步骤 4-5 实时预览，步骤 6 最终视频 ════ */}
         <div className="xx-generate-right-col">
-          {/* 预览视频面板（Step4+ 显示，含 CSS 标题实时预览层） */}
-          {currentStep >= 4 && (
-            <PreviewVideoPanel
+          {currentStep >= 4 && currentStep <= 5 && !!currentTemplate && (
+            <FrontendPreviewPlayer
               assets={previewAssets}
               template={currentTemplate}
               videoRatio={videoRatio}
-              assetsReady={previewAssetsReady}
-              assetsLoading={previewAssetsLoading}
-              titleSettings={titleSettings}
-              voiceAudioUrl={voiceAudioUrl}
+              ready={previewAssets.length > 0}
+              serverClips={serverClips}
+              voiceAudioUrl={previewVoiceAudioUrl || undefined}
+              titleSettings={{
+                title: titleSettings.title,
+                size: titleSettings.size,
+                font: titleSettings.font,
+                color: titleSettings.color,
+                position: titleSettings.position as "top" | "center" | "bottom",
+                bold: titleSettings.bold,
+                italic: titleSettings.italic,
+                stroke: titleSettings.stroke,
+                shadow: titleSettings.shadow,
+              }}
             />
           )}
-          {currentStep >= 6 && (
-            <GenerateResultPanel
-              generated={generated}
-              generating={generating}
-              progress={progress}
-              generateError={generateError}
-              generatedVideos={generatedVideos}
-              onVideoPreview={(video) => {
-                setPreviewVideo(video)
-                setPreviewModalOpen(true)
-              }}
-              onDownload={handleDownload}
-              onShare={handleShare}
-              onGoToLibrary={() => navigate("/app/products")}
-            />
+          {currentStep === 6 && generated && generatedVideos.length > 0 && (
+            <div className="xx-inline-video-player">
+              <video
+                src={generatedVideos[0].download_url || generatedVideos[0].file_url}
+                controls
+                autoPlay
+                style={{ width: "100%", maxHeight: "70vh", objectFit: "contain", borderRadius: 12 }}
+                poster={generatedVideos[0].thumbnail_url || undefined}
+              />
+              <div style={{ display: "flex", gap: 8, marginTop: 12, justifyContent: "center" }}>
+                <button className="xx-btn xx-btn-ghost xx-btn-sm" onClick={handleDownload}>
+                  ⬇️ 下载
+                </button>
+                <button className="xx-btn xx-btn-ghost xx-btn-sm" onClick={handleShare}>
+                  🔗 分享
+                </button>
+                <button
+                  className="xx-btn xx-btn-ghost xx-btn-sm"
+                  onClick={() => navigate("/app/products")}
+                >
+                  📁 前往成片库
+                </button>
+              </div>
+            </div>
           )}
         </div>
       </div>
 
-      {/* ── 视频预览弹窗 ── */}
-      <Modal
-        className="xx-preview-modal"
-        open={previewModalOpen}
-        onCancel={() => setPreviewModalOpen(false)}
-        footer={null}
-        width="80vw"
-        centered
-        destroyOnClose
-      >
-        {previewVideo && (
-          <div className="xx-preview-modal-content">
-            <video
-              src={previewVideo.download_url || previewVideo.file_url}
-              controls
-              autoPlay
-              style={{ width: "100%", maxHeight: "70vh", objectFit: "contain" }}
-              poster={previewVideo.thumbnail_url || undefined}
-            />
-          </div>
-        )}
-      </Modal>
-
-      {/* ── 音色克隆弹窗 ── */}
+      {/* 音色克隆弹窗 */}
       <CloneModal
         open={cloneModalOpen}
         onClose={() => setCloneModalOpen(false)}

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

@@ -16,14 +16,16 @@ import {
 } from "@ant-design/icons"
 import type { AssetItem } from "@/api/assets"
 import type { EditingTemplate } from "@/api/editing-planner"
+import type { EditPlanClip } from "@/api/template-editor"
 import { useSegmentScheduler, type PlaybackSegment } from "../hooks/useSegmentScheduler"
-import { useCanvasPlayer, isWebCodecsSupported } from "../hooks/useCanvasPlayer"
+import { useCanvasPlayer } from "../hooks/useCanvasPlayer"
 
 interface FrontendPreviewPlayerProps {
   assets: AssetItem[]
   template: EditingTemplate | null
   videoRatio: string
   ready: boolean
+  serverClips?: EditPlanClip[]
   voiceAudioUrl?: string
   titleSettings?: {
     title: string
@@ -50,9 +52,31 @@ function formatTime(seconds: number): string {
 function buildPlaybackSegments(
   assets: AssetItem[],
   template: EditingTemplate | null,
+  serverClips?: EditPlanClip[],
 ): PlaybackSegment[] {
   if (!assets.length) return []
 
+  // Build asset lookup map
+  const assetMap = new Map(assets.map((a) => [a.id, a]))
+
+  // 优先使用服务端 clips（含随机 start_time 和正确数量），与最终生成结果一致
+  if (serverClips && serverClips.length > 0) {
+    const segments: PlaybackSegment[] = []
+    for (const clip of serverClips) {
+      const asset = assetMap.get(clip.asset_id)
+      if (!asset) continue
+      const assetDuration = asset.duration || asset.metadata?.duration || 30
+      const startTime = clip.start_time || 0
+      const endTime = Math.min(startTime + (clip.duration || assetDuration), assetDuration)
+      const videoUrl = asset.file_url || asset.storage_key
+      segments.push({ assetId: asset.id, videoUrl, startTime, endTime, order: clip.order })
+    }
+    if (segments.length > 0) {
+      return segments.sort((a, b) => a.order - b.order)
+    }
+  }
+
+  // Fallback: 本地构建片段（与旧行为一致）
   const templateSegments = template?.segments || []
   const segments: PlaybackSegment[] = []
 
@@ -76,13 +100,66 @@ function buildPlaybackSegments(
 const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
   assets,
   template,
-  videoRatio: _videoRatio,
+  videoRatio,
   ready,
+  serverClips,
   voiceAudioUrl,
   titleSettings,
 }) => {
-  const segments = useMemo(() => buildPlaybackSegments(assets, template), [assets, template])
-  const useWebCodecs = isWebCodecsSupported()
+  const segments = useMemo(
+    () => buildPlaybackSegments(assets, template, serverClips),
+    [assets, template, serverClips],
+  )
+
+  // ── ASS 坐标系参数（与后端 ass_subtitle_builder.py 一致） ──
+  const TITLE_MARGIN_TOP = 120
+  const TITLE_MARGIN_BOTTOM = 60
+  const TITLE_MARGIN_SIDE = 40
+  const playRes = (() => {
+    switch (videoRatio) {
+      case "16:9":
+        return { width: 1920, height: 1080 }
+      case "1:1":
+        return { width: 1080, height: 1080 }
+      case "9:16":
+      default:
+        return { width: 1080, height: 1920 }
+    }
+  })()
+  const playerContainerRef = useRef<HTMLDivElement>(null)
+  const [containerHeight, setContainerHeight] = useState(0)
+  useEffect(() => {
+    const el = playerContainerRef.current
+    if (!el) return
+    const ro = new ResizeObserver((entries) => {
+      for (const entry of entries) {
+        const h = entry.contentRect.height
+        if (h > 0) setContainerHeight(h)
+      }
+    })
+    ro.observe(el)
+    const rect = el.getBoundingClientRect()
+    if (rect.height > 0) setContainerHeight(rect.height)
+    return () => ro.disconnect()
+  }, [])
+
+  // 标题字号按容器高度与 PlayResY 的比例缩放
+  const titleFontSizePx =
+    containerHeight > 0
+      ? ((titleSettings?.size ?? 36) / playRes.height) * containerHeight
+      : (titleSettings?.size ?? 36)
+  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)
@@ -94,7 +171,7 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
         fontSize: titleSettings.size,
         fontFamily: titleSettings.font || "思源黑体",
         color: titleSettings.color || "#ffffff",
-        position: titleSettings.position || "bottom",
+        position: titleSettings.position || "top",
         bold: titleSettings.bold,
         stroke: titleSettings.stroke,
         shadow: titleSettings.shadow,
@@ -122,9 +199,10 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
 
   const { state: canvasState, controls: canvasControls } = useCanvasPlayer(
     canvasRef,
-    canvasSegments,
+    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
@@ -195,7 +273,9 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
     const audio = audioRef.current
     if (!audio || !audio.src || !isPlaying) return
     audio.currentTime = currentTime
-  }, [segmentSyncKey, isPlaying, 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])
 
   const handleSeekTo = useCallback(
     (time: number) => {
@@ -265,51 +345,36 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
 
   const progressPercent = totalDuration > 0 ? (currentTime / totalDuration) * 100 : 0
 
-  // ── Canvas ResizeObserver ──
+  // ── Canvas 容器 ref（保留声明，WebCodecs 兜底路径仍引用） ──
   const canvasContainerRef = useRef<HTMLDivElement>(null)
-  useEffect(() => {
-    if (!effectiveUseWebCodecs || !canPlay) return
-    const container = canvasContainerRef.current
-    const canvas = canvasRef.current
-    if (!container || !canvas) return
-    // 立即设置一次 canvas 像素分辨率，避免默认 300×150 导致首帧变形
-    const initRect = container.getBoundingClientRect()
-    if (initRect.width > 0 && initRect.height > 0) {
-      const dpr = window.devicePixelRatio || 1
-      canvas.width = initRect.width * dpr
-      canvas.height = initRect.height * dpr
-    }
-    const ro = new ResizeObserver((entries) => {
-      for (const entry of entries) {
-        const { width, height } = entry.contentRect
-        if (width > 0 && height > 0) {
-          canvas.width = width * window.devicePixelRatio
-          canvas.height = height * window.devicePixelRatio
-        }
-      }
-    })
-    ro.observe(container)
-    return () => ro.disconnect()
-  }, [effectiveUseWebCodecs, canPlay])
 
   // ── 未就绪 ──
   if (!ready || !assets.length) {
     return (
       <div
-        className="xx-preview-empty"
         style={{
-          position: "absolute",
-          inset: 0,
+          position: "relative",
+          width: "100%",
+          maxWidth: 280,
+          aspectRatio: "9 / 16",
+          background: "#0a0a0a",
+          borderRadius: 24,
+          overflow: "hidden",
+          boxShadow:
+            "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)",
           display: "flex",
           flexDirection: "column",
           alignItems: "center",
           justifyContent: "center",
-          zIndex: 1,
         }}
       >
-        <SoundOutlined style={{ fontSize: 48, color: "var(--text-tertiary)", marginBottom: 12 }} />
-        <p className="xx-preview-empty-title">准备预览素材...</p>
-        <p className="xx-preview-empty-desc">加载素材后即可预览播放</p>
+        <SoundOutlined 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" }}>
+          准备预览素材...
+        </p>
+        <p style={{ color: "rgba(255,255,255,0.35)", fontSize: 12, margin: 0 }}>
+          加载素材后即可预览播放
+        </p>
       </div>
     )
   }
@@ -319,31 +384,48 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
     const showDecodeError = forceVideoFallback && canvasState.hasDecodeError
     return (
       <div
-        className="xx-preview-empty"
         style={{
-          position: "absolute",
-          inset: 0,
+          position: "relative",
+          width: "100%",
+          maxWidth: 280,
+          aspectRatio: "9 / 16",
+          background: "#0a0a0a",
+          borderRadius: 24,
+          overflow: "hidden",
+          boxShadow:
+            "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)",
           display: "flex",
           flexDirection: "column",
           alignItems: "center",
           justifyContent: "center",
-          zIndex: 1,
+          padding: 24,
         }}
       >
         {isBuffering ? (
           <>
-            <LoadingOutlined style={{ fontSize: 48, color: "#fff", marginBottom: 12 }} spin />
-            <p style={{ color: "rgba(255,255,255,0.8)" }}>加载中...</p>
+            <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: 48, color: "#ef4444", marginBottom: 12 }} />
-            <p className="xx-preview-empty-title" style={{ color: "rgba(255,255,255,0.9)" }}>
+            <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
-              className="xx-preview-empty-desc"
-              style={{ color: "rgba(255,255,255,0.6)", maxWidth: 300, textAlign: "center" }}
+              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>
@@ -351,10 +433,14 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
         ) : (
           <>
             <PlayCircleOutlined
-              style={{ fontSize: 48, color: "var(--text-tertiary)", marginBottom: 12 }}
+              style={{ fontSize: 40, color: "rgba(255,255,255,0.3)", marginBottom: 12 }}
             />
-            <p className="xx-preview-empty-title">暂无可播放素材</p>
-            <p className="xx-preview-empty-desc">请先在左侧选择素材</p>
+            <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>
@@ -362,7 +448,20 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
   }
 
   return (
-    <>
+    <div
+      ref={playerContainerRef}
+      style={{
+        position: "relative",
+        width: "100%",
+        maxWidth: 280,
+        aspectRatio: "9 / 16",
+        background: "#0a0a0a",
+        borderRadius: 24,
+        overflow: "hidden",
+        boxShadow:
+          "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
@@ -385,7 +484,7 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
         </div>
       )}
 
-      {/* ── Video 渲染层（fallback 路径，或 WebCodecs 解码失败时自动切换） ── */}
+      {/* ── Video 渲染层（默认路径，浏览器原生硬件解码） ── */}
       {!effectiveUseWebCodecs &&
         segments.map((seg, i) => (
           <video
@@ -394,9 +493,7 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
             ref={(el) => {
               videoRefs.current[i] = el
             }}
-            preload={
-              i === videoCurrentSegIdx ? "auto" : i === videoCurrentSegIdx + 1 ? "metadata" : "none"
-            }
+            preload="auto"
             src={seg.videoUrl}
             style={{
               position: "absolute",
@@ -413,58 +510,121 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
           />
         ))}
 
-      {/* 播放按钮 */}
+      {/* 标题CSS叠加层 — 与后端 ASS 烧录坐标系 1:1 对齐 */}
+      {titleSettings?.title && (
+        <div
+          style={{
+            position: "absolute",
+            inset: 0,
+            zIndex: 5,
+            pointerEvents: "none",
+            overflow: "hidden",
+          }}
+        >
+          <div
+            style={{
+              position: "absolute",
+              left: `${titleSidePct}%`,
+              right: `${titleSidePct}%`,
+              textAlign: "center",
+              ...(titleSettings.position === "top"
+                ? { top: `${titleTopPct}%` }
+                : titleSettings.position === "center"
+                  ? { top: "50%", transform: "translateY(-50%)" }
+                  : { bottom: `${titleBottomPct}%` }),
+            }}
+          >
+            <span
+              style={{
+                fontSize: `${titleFontSizePx}px`,
+                fontFamily: titleSettings.font || "思源黑体",
+                color: titleSettings.color || "#ffffff",
+                fontWeight: titleSettings.bold ? 700 : 400,
+                fontStyle: titleSettings.italic ? "italic" : "normal",
+                lineHeight: 1.05,
+                wordBreak: "break-word",
+                WebkitTextStroke: titleSettings.stroke
+                  ? `${titleStrokeWidth}px #000000`
+                  : undefined,
+                textShadow: titleSettings.shadow
+                  ? `${titleShadowOffset}px ${titleShadowOffset}px ${titleShadowBlur}px rgba(0,0,0,0.8)`
+                  : undefined,
+              }}
+            >
+              {titleSettings.title.split(/[/／]/).map((part, i) => (
+                <span key={i}>
+                  {i > 0 && <br />}
+                  {part}
+                </span>
+              ))}
+            </span>
+          </div>
+        </div>
+      )}
+
+      {/* 中央播放按钮 */}
       {!isPlaying && (
         <button
-          className="xx-preview-play-btn"
           onClick={handleTogglePlay}
           style={{
             position: "absolute",
             top: "50%",
             left: "50%",
             transform: "translate(-50%, -50%)",
-            background: "rgba(0,0,0,0.5)",
-            border: "none",
+            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: 56,
-            height: 56,
+            width: 52,
+            height: 52,
             cursor: "pointer",
             display: "flex",
             alignItems: "center",
             justifyContent: "center",
             color: "#fff",
-            fontSize: 28,
+            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>
       )}
 
-      {/* 片段指示器 */}
+      {/* 片段指示器 — 右上角胶囊 */}
       <div
         style={{
           position: "absolute",
           top: 8,
-          left: 8,
-          background: "rgba(0,0,0,0.6)",
-          color: "#fff",
-          fontSize: 11,
+          right: 8,
+          background: "rgba(0,0,0,0.45)",
+          backdropFilter: "blur(8px)",
+          WebkitBackdropFilter: "blur(8px)",
+          color: "rgba(255,255,255,0.9)",
+          fontSize: 10,
+          fontWeight: 500,
           padding: "2px 8px",
-          borderRadius: 4,
+          borderRadius: 999,
           zIndex: 10,
+          border: "1px solid rgba(255,255,255,0.1)",
+          letterSpacing: 0.3,
         }}
       >
-        {effectiveUseWebCodecs
-          ? "Canvas"
-          : forceVideoFallback
-            ? "Canvas 解码失败，已切换原生播放"
-            : `片段 ${videoCurrentSegIdx + 1}/${segments.length}`}
+        {`${videoCurrentSegIdx + 1} / ${segments.length}`}
       </div>
 
-      {/* 控制条 */}
+      {/* 控制条 — 手机风格毛玻璃 */}
       <div
-        className="xx-preview-controls"
         style={{
           position: "absolute",
           bottom: 0,
@@ -472,23 +632,36 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
           right: 0,
           display: "flex",
           alignItems: "center",
-          gap: 8,
-          padding: "8px 12px",
-          background: "linear-gradient(transparent, rgba(0,0,0,0.6))",
+          gap: 10,
+          padding: "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: "none",
+            background: "rgba(255,255,255,0.15)",
             border: "none",
             color: "#fff",
-            fontSize: 18,
+            fontSize: 16,
             cursor: "pointer",
-            padding: 4,
+            width: 32,
+            height: 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 />}
@@ -496,10 +669,11 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
 
         <span
           style={{
-            fontSize: 12,
-            color: "rgba(255,255,255,0.8)",
-            minWidth: 80,
+            fontSize: 11,
+            color: "rgba(255,255,255,0.85)",
+            minWidth: 72,
             fontVariantNumeric: "tabular-nums",
+            letterSpacing: 0.2,
           }}
         >
           {formatTime(currentTime)} / {formatTime(totalDuration)}
@@ -510,8 +684,8 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
           onMouseDown={handleMouseDown}
           style={{
             flex: 1,
-            height: 4,
-            background: "rgba(255,255,255,0.15)",
+            height: 3,
+            background: "rgba(255,255,255,0.2)",
             borderRadius: 2,
             cursor: "pointer",
             position: "relative",
@@ -521,7 +695,7 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
             style={{
               height: "100%",
               width: `${progressPercent}%`,
-              background: "#3b82f6",
+              background: "#fff",
               borderRadius: 2,
               transition: isDragging ? "none" : "width 0.1s linear",
             }}
@@ -535,15 +709,15 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
               width: 10,
               height: 10,
               borderRadius: "50%",
-              background: "#3b82f6",
-              border: "2px solid #fff",
+              background: "#fff",
+              boxShadow: "0 0 6px rgba(255,255,255,0.5)",
               opacity: isDragging ? 1 : 0,
               transition: "opacity 0.15s",
             }}
           />
         </div>
       </div>
-    </>
+    </div>
   )
 }
 

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

@@ -28,7 +28,7 @@ export const GenerateStepActions: React.FC<GenerateStepActionsProps> = ({
       <button className="xx-btn xx-btn-ghost" onClick={onPrev} disabled={currentStep === 1}>
         ← 上一步
       </button>
-      {currentStep < 7 ? (
+      {currentStep < 6 ? (
         <button className="xx-btn xx-btn-primary" onClick={onNext}>
           下一步 →
         </button>

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

@@ -1,12 +1,13 @@
 /**
  * GeneratePage 步骤内容渲染
  * 根据当前步骤渲染对应的 Step 组件
- * 步骤顺序：模板(1) → 素材(2) → 配音(3) → 标题(4) → 预览(5) → 封面(6) → 确认(7)
+ * 步骤顺序（6步）：模板(1) → 素材(2) → 配音(3) → 标题(4) → 封面(5) → 确认(6)
  *
- * V24: 移除 Step5 预览生成相关 props，改为纯标题样式编辑
+ * 原 Step4(标题) + Step5(预览样式) 已合并为 Step4
  */
 import React from "react"
 import type { EditingTemplate } from "@/api/editing-planner"
+import type { EditPlanClip } from "@/api/template-editor"
 import type { PresetVoiceItem } from "@/api/voices"
 import type { VoiceClone } from "@/api/voice-clone"
 import type { CoverConfig } from "../types/cover"
@@ -14,10 +15,9 @@ import type { TitleSettings } from "../types"
 import Step1TemplateSelect from "../components/Step1TemplateSelect"
 import Step2MaterialSelect from "../components/Step2MaterialSelect"
 import Step3VoiceSelect from "../components/Step5VoiceSelect"
-import Step5GeneratePreview from "../components/Step5GeneratePreview"
 import Step4TitleSettings from "../components/Step4TitleSettings"
-import Step6CoverSettings from "../components/Step6CoverSettings"
-import Step7ConfirmGenerate from "../components/Step7ConfirmGenerate"
+import Step5CoverSettings from "../components/Step6CoverSettings"
+import Step6ConfirmGenerate from "../components/Step7ConfirmGenerate"
 import type { GeneratedVideo } from "@/api/template-editor"
 
 export interface GenerateStepContentProps {
@@ -36,7 +36,7 @@ export interface GenerateStepContentProps {
   /* 标题 */
   titleSettings: TitleSettings
   onTitleSettingsChange: (settings: TitleSettings) => void
-  /* 标题样式回调 — Step5 样式面板使用 */
+  /* 标题样式回调 — Step4 合并后直接使用 */
   onUpdatePosition: (position: string) => void
   onUpdateFont: (font: string) => void
   onUpdateSize: (size: number) => void
@@ -55,6 +55,7 @@ export interface GenerateStepContentProps {
   selectedVoice: string
   onSelectedVoiceChange: (id: string) => void
   totalVideoDuration?: number
+  onServerClipsChange: (clips: EditPlanClip[]) => void
   voiceMode: "preset" | "custom" | "clone"
   onVoiceModeChange: (mode: "preset" | "custom" | "clone") => void
   selectedClonedVoice: string
@@ -65,8 +66,6 @@ export interface GenerateStepContentProps {
   cloneModalOpen: boolean
   onCloneModalOpenChange: (open: boolean) => void
   /* 生成 */
-  generateCount: number
-  onGenerateCountChange: (n: number) => void
   generating: boolean
   generated: boolean
   generateError: string | null
@@ -76,6 +75,14 @@ export interface GenerateStepContentProps {
   onDismissError: () => void
   /* 其他 */
   presetVoices: PresetVoiceItem[]
+  /** 预览任务创建回调——传递给封面步骤 */
+  onPreviewTaskCreated?: (taskId: string) => void
+  /** 从预览响应中提取到 source_edit_plan_id 时的回调 */
+  onSourceEditPlanIdExtracted?: (planId: string) => void
+  /** BGM 开关 */
+  bgm: boolean
+  /** BGM 配置（来自模板） */
+  bgmConfig?: { enabled: boolean; music_id?: string }
 }
 
 export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) => {
@@ -108,11 +115,10 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
     selectedVoice,
     onSelectedVoiceChange,
     totalVideoDuration,
+    onServerClipsChange,
     voiceMode,
     selectedClonedVoice,
     clonedVoices,
-    generateCount,
-    onGenerateCountChange,
     generating,
     generated,
     generateError,
@@ -121,8 +127,16 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
     onRetry,
     onDismissError,
     presetVoices,
+    onPreviewTaskCreated,
+    onSourceEditPlanIdExtracted,
+    bgm,
+    bgmConfig,
   } = props
 
+  /* 当前模板的 segments，传给 Step2 构建 clips */
+  const currentTemplate = userTemplates.find((t) => t.id === selectedTemplate)
+  const templateSegments = currentTemplate?.segments
+
   switch (currentStep) {
     case 1:
       return (
@@ -141,6 +155,9 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
           onSelectedMaterialsChange={onSelectedMaterialsChange}
           smartSelectedIds={smartSelectedIds}
           onSmartSelectedIdsChange={onSmartSelectedIdsChange}
+          selectedTemplate={selectedTemplate}
+          templateSegments={templateSegments}
+          onServerClipsChange={onServerClipsChange}
         />
       )
     case 3:
@@ -156,12 +173,7 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
         <Step4TitleSettings
           titleSettings={titleSettings}
           onTitleSettingsChange={onTitleSettingsChange}
-        />
-      )
-    case 5:
-      return (
-        <Step5GeneratePreview
-          titleSettings={titleSettings}
+          selectedTemplate={selectedTemplate}
           onUpdatePosition={onUpdatePosition}
           onUpdateFont={onUpdateFont}
           onUpdateSize={onUpdateSize}
@@ -174,19 +186,27 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
           titlePresets={titlePresets}
         />
       )
-    case 6:
+    case 5:
       return (
-        <Step6CoverSettings
+        <Step5CoverSettings
           coverSettings={coverSettings}
           onCoverSettingsChange={onCoverSettingsChange}
           duration={duration}
           assetIds={materialMode === "auto" ? smartSelectedIds : selectedMaterials}
           selectedTemplate={selectedTemplate}
+          titleSettings={titleSettings}
+          onPreviewTaskCreated={onPreviewTaskCreated}
+          onSourceEditPlanIdExtracted={onSourceEditPlanIdExtracted}
+          voiceMode={voiceMode}
+          selectedVoice={selectedVoice}
+          selectedClonedVoice={selectedClonedVoice}
+          bgm={bgm}
+          bgmConfig={bgmConfig}
         />
       )
-    case 7:
+    case 6:
       return (
-        <Step7ConfirmGenerate
+        <Step6ConfirmGenerate
           templates={userTemplates}
           selectedTemplate={selectedTemplate}
           materialMode={materialMode}
@@ -199,8 +219,6 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
           presetVoices={presetVoices}
           clonedVoices={clonedVoices}
           coverSettings={coverSettings}
-          generateCount={generateCount}
-          onGenerateCountChange={onGenerateCountChange}
           generating={generating}
           generated={generated}
           generateError={generateError}

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

@@ -1,57 +1,72 @@
 /**
- * 右侧预览视频面板
- * Step4+: 显示预览视频面板
- * Step5: 前端实时预览 — 用原生 video 播放素材片段 + CSS 标题叠加
+ * 右侧预览视频面板 — 服务器渲染预览架构
  *
- * 架构改造：完全去除后端 FFmpeg 预览依赖
- * - 使用 FrontendPreviewPlayer 直接播放素材片段
- * - TitleOverlay CSS 层实时响应标题样式变化
+ * Step4+: 显示预览面板
+ * Step5: 播放服务器渲染的真实视频（POST /generation/preview）
  *
- * 布局：本组件提供 .xx-preview-video 容器（position: relative + overflow: hidden）
- * FrontendPreviewPlayer 的内容通过 absolute 定位填充容器
- * TitleOverlay 通过 absolute 定位 + z-index: 30 覆盖在最上层
+ * 架构：
+ * - 进入 Step4/5 时自动创建服务器预览渲染任务
+ * - 轮询完成后用 <video> 标签播放返回的 video_url
+ * - 标题样式编辑时 CSS TitleOverlay 实时叠加预览
+ * - 素材/配音/BGM 变更自动重新渲染
+ * - 标题文字/样式变更标记 stale，保留旧视频 + 显示"重新预览"按钮
+ *
+ * 点"确认生成"时走 confirm 路径，成品就是预览视频本身，100% 一致。
  */
 import React, { useMemo, useRef, useState, useEffect } from "react"
-import { LoadingOutlined } from "@ant-design/icons"
-import type { AssetItem } from "@/api/assets"
-import type { EditingTemplate } from "@/api/editing-planner"
+import { LoadingOutlined, ReloadOutlined, ExclamationCircleOutlined } from "@ant-design/icons"
+import { Button } from "antd"
 import type { TitleSettings } from "../types"
-import FrontendPreviewPlayer from "./FrontendPreviewPlayer"
+import { getFontFamily } from "../constants"
+import type { ServerPreviewStatus } from "../hooks/useServerPreview"
 
 interface PreviewVideoPanelProps {
-  /** 已加载的素材列表 */
-  assets: AssetItem[]
-  /** 当前模板 */
-  template: EditingTemplate | null
+  /** 服务器预览状态 */
+  previewStatus: ServerPreviewStatus
+  /** 服务器渲染视频 URL */
+  videoUrl: string | null
+  /** 渲染进度 0-100 */
+  progress: number
+  /** 错误信息 */
+  error: string | null
+  /** 重新预览回调 */
+  onRetry: () => void
   /** 视频比例 */
   videoRatio: string
-  /** 素材是否已加载就绪 */
-  assetsReady: boolean
-  /** 素材是否正在加载 */
-  assetsLoading: boolean
-  /** 标题设置 — 用于 CSS 实时预览层 */
+  /** 标题设置 — CSS 实时预览层 */
   titleSettings?: TitleSettings
-  /** 配音音频 URL */
-  voiceAudioUrl?: string
+  /** 素材数量 */
+  assetCount?: number
 }
 
 /* ── ASS 坐标系参数（与后端 ass_subtitle_builder.py 一致） ── */
-const ASS_VIDEO_HEIGHT = 720
-const ASS_TITLE_MARGIN_TOP = 60
-const ASS_TITLE_MARGIN_BOTTOM = 60
-const ASS_TITLE_MARGIN_SIDE = 40
+const TITLE_MARGIN_TOP = 120
+const TITLE_MARGIN_BOTTOM = 60
+const TITLE_MARGIN_SIDE = 40
 
-/**
- * 根据 position 计算 CSS 垂直定位
- * 与后端 position_to_ass_alignment() 对齐：top→8, center→5, bottom→2
- */
-function getPositionStyle(position: string): React.CSSProperties {
-  const sidePercent = (ASS_TITLE_MARGIN_SIDE / 1280) * 100
+/** 根据视频比例返回后端实际渲染分辨率（PlayResX × PlayResY） */
+function getResolution(ratio: string): { width: number; height: number } {
+  switch (ratio) {
+    case "16:9":
+      return { width: 1920, height: 1080 }
+    case "1:1":
+      return { width: 1080, height: 1080 }
+    case "9:16":
+    default:
+      return { width: 1080, height: 1920 }
+  }
+}
 
+function getPositionStyle(
+  position: string,
+  playResX: number,
+  playResY: number,
+): React.CSSProperties {
+  const sidePercent = (TITLE_MARGIN_SIDE / playResX) * 100
   switch (position) {
     case "bottom":
       return {
-        bottom: `${(ASS_TITLE_MARGIN_BOTTOM / ASS_VIDEO_HEIGHT) * 100}%`,
+        bottom: `${(TITLE_MARGIN_BOTTOM / playResY) * 100}%`,
         left: `${sidePercent}%`,
         right: `${sidePercent}%`,
         textAlign: "center",
@@ -67,7 +82,7 @@ function getPositionStyle(position: string): React.CSSProperties {
     case "top":
     default:
       return {
-        top: `${(ASS_TITLE_MARGIN_TOP / ASS_VIDEO_HEIGHT) * 100}%`,
+        top: `${(TITLE_MARGIN_TOP / playResY) * 100}%`,
         left: `${sidePercent}%`,
         right: `${sidePercent}%`,
         textAlign: "center",
@@ -75,52 +90,41 @@ function getPositionStyle(position: string): React.CSSProperties {
   }
 }
 
-/**
- * 构建 CSS 标题层的样式
- * 所有渲染参数与后端 FFmpeg ASS 字幕一致
- */
-function buildTitleStyle(settings: TitleSettings, containerHeight: number): React.CSSProperties {
-  // 用 px 计算 fontSize，不再依赖父元素 font-size 的百分比
+function buildTitleStyle(
+  settings: TitleSettings,
+  containerHeight: number,
+  playResY: number,
+): React.CSSProperties {
+  // 字号按容器高度与 PlayResY 的比例缩放，不设上限（与后端一致）
   const fontSizePx =
     containerHeight > 0
-      ? (Math.min(settings.size, 96) / ASS_VIDEO_HEIGHT) * containerHeight
-      : (Math.min(settings.size, 96) / ASS_VIDEO_HEIGHT) * 400 // fallback
+      ? (settings.size / playResY) * containerHeight
+      : (settings.size / playResY) * 400
 
   const base: React.CSSProperties = {
-    fontFamily: settings.font || "思源黑体",
+    fontFamily: getFontFamily(settings.font),
     fontSize: `${fontSizePx}px`,
     color: settings.color || "#ffffff",
     fontWeight: settings.bold ? 700 : 400,
     fontStyle: settings.italic ? "italic" : "normal",
-    lineHeight: 1.3,
+    lineHeight: 1.05,
     wordBreak: "break-word",
     pointerEvents: "none",
     userSelect: "none",
-    paddingLeft: `${(ASS_TITLE_MARGIN_SIDE / 1280) * 100}%`,
-    paddingRight: `${(ASS_TITLE_MARGIN_SIDE / 1280) * 100}%`,
   }
-
-  if (settings.stroke) {
-    base.WebkitTextStroke = "1px #000000"
-  }
-
-  if (settings.shadow) {
-    base.textShadow = "2px 2px 4px rgba(0,0,0,0.8)"
-  }
-
+  if (settings.stroke) base.WebkitTextStroke = "1px #000000"
+  if (settings.shadow) base.textShadow = "2px 2px 4px rgba(0,0,0,0.8)"
   return base
 }
 
-/**
- * CSS 标题预览覆盖层
- * 始终渲染：有标题显示标题，无标题显示占位文本"标题预览"
- * z-index: 20（在视频 z-index:1 和控制条 z-index:10 之上）
- */
-const TitleOverlay: React.FC<{ titleSettings: TitleSettings }> = ({ titleSettings }) => {
+/** CSS 标题实时预览覆盖层 */
+const TitleOverlay: React.FC<{ titleSettings: TitleSettings; videoRatio: string }> = ({
+  titleSettings,
+  videoRatio,
+}) => {
   const containerRef = useRef<HTMLDivElement>(null)
-  const [containerHeight, setContainerHeight] = useState(400) // fallback
+  const [containerHeight, setContainerHeight] = useState(400)
 
-  // ResizeObserver 获取容器实际高度
   useEffect(() => {
     const el = containerRef.current
     if (!el) return
@@ -131,19 +135,21 @@ const TitleOverlay: React.FC<{ titleSettings: TitleSettings }> = ({ titleSetting
       }
     })
     ro.observe(el)
-    // 初始化也读一次
     const rect = el.getBoundingClientRect()
     if (rect.height > 0) setContainerHeight(rect.height)
     return () => ro.disconnect()
   }, [])
 
+  const { width: playResX, height: playResY } = getResolution(videoRatio)
+
   const positionStyle = useMemo(
-    () => getPositionStyle(titleSettings.position),
-    [titleSettings.position],
+    () => getPositionStyle(titleSettings.position, playResX, playResY),
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+    [titleSettings.position, playResX, playResY],
   )
   const titleStyle = useMemo(
-    () => buildTitleStyle(titleSettings, containerHeight),
-    // eslint-disable-next-line react-hooks/exhaustive-deps -- 已逐字段列出 titleSettings 依赖
+    () => buildTitleStyle(titleSettings, containerHeight, playResY),
+    // eslint-disable-next-line react-hooks/exhaustive-deps
     [
       containerHeight,
       titleSettings.font,
@@ -153,6 +159,7 @@ const TitleOverlay: React.FC<{ titleSettings: TitleSettings }> = ({ titleSetting
       titleSettings.italic,
       titleSettings.stroke,
       titleSettings.shadow,
+      playResY,
     ],
   )
 
@@ -169,14 +176,13 @@ const TitleOverlay: React.FC<{ titleSettings: TitleSettings }> = ({ titleSetting
         overflow: "hidden",
       }}
     >
-      <div
-        style={{
-          ...positionStyle,
-          ...titleStyle,
-          position: "absolute",
-        }}
-      >
-        {displayTitle}
+      <div style={{ ...positionStyle, ...titleStyle, position: "absolute" }}>
+        {displayTitle.split(/[/／]/).map((part, i) => (
+          <span key={i}>
+            {i > 0 && <br />}
+            {part}
+          </span>
+        ))}
       </div>
     </div>
   )
@@ -185,30 +191,47 @@ const TitleOverlay: React.FC<{ titleSettings: TitleSettings }> = ({ titleSetting
 /* ── 主组件 ── */
 
 export const PreviewVideoPanel: React.FC<PreviewVideoPanelProps> = ({
-  assets,
-  template,
+  previewStatus,
+  videoUrl,
+  progress,
+  error,
+  onRetry,
   videoRatio,
-  assetsReady,
-  assetsLoading,
   titleSettings,
-  voiceAudioUrl,
+  assetCount,
 }) => {
   const videoAspectStyle = { aspectRatio: (videoRatio || "9:16").replace(":", "/") }
+  const isLoading = previewStatus === "loading"
+  const isReady = previewStatus === "ready" || previewStatus === "stale"
+  const isFailed = previewStatus === "failed"
+  const isIdle = previewStatus === "idle"
+  const isStale = previewStatus === "stale"
 
   return (
     <div className="xx-generate-preview">
       <div className="xx-preview-header">
         <h3>预览视频</h3>
-        {assetsReady && assets.length > 0 && <span className="xx-preview-badge">实时预览</span>}
+        {isReady && !isStale && <span className="xx-preview-badge">服务器渲染</span>}
+        {isStale && (
+          <span
+            style={{
+              fontSize: 12,
+              color: "#faad14",
+              display: "flex",
+              alignItems: "center",
+              gap: 4,
+            }}
+          >
+            <ExclamationCircleOutlined /> 配置已变更
+          </span>
+        )}
+        {isLoading && <span className="xx-preview-badge">渲染中</span>}
       </div>
 
-      {/* ✅ 预览容器 — 唯一的 .xx-preview-video 容器
-          内部所有内容（视频、控制条、标题叠加层）通过 absolute 定位填充 */}
       <div className="xx-preview-video" style={{ ...videoAspectStyle, position: "relative" }}>
-        {/* 加载中状态 */}
-        {assetsLoading && (
+        {/* 加载中 */}
+        {isLoading && (
           <div
-            className="xx-preview-loading-center"
             style={{
               position: "absolute",
               inset: 0,
@@ -217,34 +240,132 @@ export const PreviewVideoPanel: React.FC<PreviewVideoPanelProps> = ({
               alignItems: "center",
               justifyContent: "center",
               zIndex: 5,
+              background: "rgba(0,0,0,0.6)",
             }}
           >
             <LoadingOutlined style={{ fontSize: 36, color: "#fff" }} spin />
-            <p style={{ marginTop: 12, color: "rgba(255,255,255,0.8)", fontSize: 14 }}>
-              加载素材中...
+            <p style={{ marginTop: 12, color: "rgba(255,255,255,0.9)", fontSize: 14 }}>
+              正在渲染预览视频{progress > 0 ? `...${progress}%` : "..."}
+            </p>
+            <p style={{ marginTop: 4, color: "rgba(255,255,255,0.5)", fontSize: 12 }}>
+              首次渲染约需 30-60 秒
             </p>
           </div>
         )}
 
-        {/* 前端播放器（视频 + 控制条 + 播放按钮）*/}
-        <FrontendPreviewPlayer
-          assets={assets}
-          template={template}
-          videoRatio={videoRatio}
-          ready={assetsReady}
-          voiceAudioUrl={voiceAudioUrl}
-        />
+        {/* 空闲状态（尚未触发预览） */}
+        {isIdle && (
+          <div
+            style={{
+              position: "absolute",
+              inset: 0,
+              display: "flex",
+              flexDirection: "column",
+              alignItems: "center",
+              justifyContent: "center",
+              background: "rgba(0,0,0,0.3)",
+              zIndex: 5,
+            }}
+          >
+            <p style={{ color: "rgba(255,255,255,0.7)", fontSize: 14 }}>等待素材选择...</p>
+          </div>
+        )}
 
-        {/* CSS 标题实时预览层 — z-index: 20，始终渲染在内容层之上 */}
-        {titleSettings && <TitleOverlay titleSettings={titleSettings} />}
+        {/* 服务器渲染的真实视频 */}
+        {isReady && videoUrl && (
+          <video
+            key={videoUrl}
+            src={videoUrl}
+            controls
+            autoPlay
+            loop
+            playsInline
+            style={{
+              width: "100%",
+              height: "100%",
+              objectFit: "contain",
+              position: "absolute",
+              inset: 0,
+              zIndex: 1,
+            }}
+          />
+        )}
+
+        {/* 标题样式实时预览层（仅在有视频时叠加） */}
+        {isReady && titleSettings && (
+          <TitleOverlay titleSettings={titleSettings} videoRatio={videoRatio} />
+        )}
+
+        {/* stale 遮罩：配置变更提示 */}
+        {isStale && (
+          <div
+            style={{
+              position: "absolute",
+              bottom: 0,
+              left: 0,
+              right: 0,
+              padding: "10px 16px",
+              background: "linear-gradient(transparent, rgba(0,0,0,0.85))",
+              display: "flex",
+              alignItems: "center",
+              justifyContent: "space-between",
+              gap: 8,
+              zIndex: 30,
+            }}
+          >
+            <span style={{ color: "rgba(255,255,255,0.9)", fontSize: 12 }}>
+              配置已变更，预览内容可能不是最新
+            </span>
+            <Button size="small" type="primary" icon={<ReloadOutlined />} onClick={onRetry}>
+              重新预览
+            </Button>
+          </div>
+        )}
+
+        {/* 错误状态 */}
+        {isFailed && (
+          <div
+            style={{
+              position: "absolute",
+              inset: 0,
+              display: "flex",
+              flexDirection: "column",
+              alignItems: "center",
+              justifyContent: "center",
+              background: "rgba(0,0,0,0.7)",
+              zIndex: 10,
+              padding: 24,
+            }}
+          >
+            <ExclamationCircleOutlined style={{ fontSize: 40, color: "#ff4d4f" }} />
+            <p
+              style={{
+                marginTop: 12,
+                color: "rgba(255,255,255,0.9)",
+                fontSize: 14,
+                textAlign: "center",
+              }}
+            >
+              {error || "预览渲染失败"}
+            </p>
+            <Button
+              type="primary"
+              icon={<ReloadOutlined />}
+              onClick={onRetry}
+              style={{ marginTop: 12 }}
+            >
+              重新预览
+            </Button>
+          </div>
+        )}
       </div>
 
       {/* 素材信息 */}
-      {assetsReady && assets.length > 0 && (
+      {assetCount !== undefined && assetCount > 0 && (
         <div className="xx-preview-info">
           <div className="xx-preview-info-row">
             <span>素材数</span>
-            <span>{assets.length} 个</span>
+            <span>{assetCount} 个</span>
           </div>
           <div className="xx-preview-info-row">
             <span>比例</span>

apps/web/src/pages/generate/components/Step2MaterialSelect.tsx

@@ -2,6 +2,8 @@
  * Step 2 素材选择组件
  */
 import React from "react"
+import type { TemplateSegment } from "@/api/templates/types"
+import type { EditPlanClip } from "@/api/template-editor"
 import { useStep2Materials } from "../hooks/useStep2Materials"
 import MaterialModeTabs from "./material/MaterialModeTabs"
 import ManualMaterialList from "./material/ManualMaterialList"
@@ -15,6 +17,12 @@ interface Step2MaterialSelectProps {
   onSelectedMaterialsChange: (ids: string[]) => void
   smartSelectedIds: string[]
   onSmartSelectedIdsChange: (ids: string[]) => void
+  /** 当前选中的模板/草稿 ID，用于自动保存 */
+  selectedTemplate?: string
+  /** 当前模板的 segments（用于构建 clips duration） */
+  templateSegments?: TemplateSegment[]
+  /** 服务端 clips 创建成功后的回调 */
+  onServerClipsChange?: (clips: EditPlanClip[]) => void
 }
 
 const Step2MaterialSelect: React.FC<Step2MaterialSelectProps> = (props) => {

apps/web/src/pages/generate/components/Step4TitleSettings.tsx

@@ -1,21 +1,50 @@
 /**
- * Step 4 标题设置组件
- * 仅包含标题文字输入 + AI 标题生成
- * 标题样式面板已迁移到 Step5（生成预览页面）
+ * Step 4 选择标题（合并原 Step4 标题输入 + Step5 标题样式面板）
+ *
+ * 左侧：标题文字输入 + AI生成标题 + 样式设置（位置/字号/字体/颜色/样式/预设）
+ * 右侧：FrontendPreviewPlayer 实时预览（由 GeneratePage 统一渲染）
  */
 import React from "react"
 import { AutoComplete } from "antd"
+import { PlayCircleOutlined } from "@ant-design/icons"
 import type { TitleSettings } from "../types"
+import { POSITION_OPTIONS, FONT_OPTIONS } from "../constants"
 import { useStep4Title } from "../hooks/useStep4Title"
 import AiTitleGenerator from "./title/AiTitleGenerator"
+import TitleStylePanel from "./title/TitleStylePanel"
 
 interface Step4TitleSettingsProps {
   titleSettings: TitleSettings
   onTitleSettingsChange: (settings: TitleSettings) => void
+  /** 当前选中的模板/草稿 ID，用于自动保存 */
+  selectedTemplate?: string
+  /* 标题样式回调 */
+  onUpdatePosition: (position: string) => void
+  onUpdateFont: (font: string) => void
+  onUpdateSize: (size: number) => void
+  onToggleBold: () => void
+  onToggleItalic: () => void
+  onToggleStroke: () => void
+  onToggleShadow: () => void
+  onApplyPreset: (presetKey: string) => void
+  activePreset: string | null
+  titlePresets: { key: string; label: string; previewStyle: React.CSSProperties }[]
 }
 
 const Step4TitleSettings: React.FC<Step4TitleSettingsProps> = (props) => {
   const t = useStep4Title(props)
+  const {
+    onUpdatePosition,
+    onUpdateFont,
+    onUpdateSize,
+    onToggleBold,
+    onToggleItalic,
+    onToggleStroke,
+    onToggleShadow,
+    onApplyPreset,
+    activePreset,
+    titlePresets,
+  } = props
 
   return (
     <div className="xx-form-section">
@@ -110,6 +139,42 @@ const Step4TitleSettings: React.FC<Step4TitleSettingsProps> = (props) => {
           </div>
         </>
       )}
+
+      {/* 标题样式面板（原 Step5） */}
+      <div
+        style={{
+          display: "flex",
+          alignItems: "center",
+          gap: 8,
+          padding: "10px 14px",
+          background: "rgba(59, 130, 246, 0.08)",
+          borderRadius: 8,
+          marginTop: 16,
+          marginBottom: 12,
+          border: "1px solid rgba(59, 130, 246, 0.15)",
+        }}
+      >
+        <PlayCircleOutlined style={{ fontSize: 16, color: "#3b82f6" }} />
+        <span style={{ fontSize: 12, color: "var(--text-secondary, #666)" }}>
+          右侧为实时预览，调整样式即时生效
+        </span>
+      </div>
+
+      <TitleStylePanel
+        settings={t.titleSettings}
+        onUpdatePosition={onUpdatePosition}
+        onUpdateFont={onUpdateFont}
+        onUpdateSize={onUpdateSize}
+        onToggleBold={onToggleBold}
+        onToggleItalic={onToggleItalic}
+        onToggleStroke={onToggleStroke}
+        onToggleShadow={onToggleShadow}
+        onApplyPreset={onApplyPreset}
+        activePreset={activePreset}
+        titlePresets={titlePresets}
+        POSITION_OPTIONS={POSITION_OPTIONS}
+        FONT_OPTIONS={FONT_OPTIONS}
+      />
     </div>
   )
 }

apps/web/src/pages/generate/components/Step5GeneratePreview.tsx

@@ -1,84 +0,0 @@
-/**
- * Step 5 生成预览组件
- * 架构改造：移除后端预览生成，改为前端实时预览
- * 左侧仅保留标题样式面板，视频在右侧 PreviewVideoPanel 实时播放
- */
-import React from "react"
-import { PlayCircleOutlined } from "@ant-design/icons"
-import { POSITION_OPTIONS, FONT_OPTIONS } from "../constants"
-import type { TitleSettings } from "../types"
-import TitleStylePanel from "./title/TitleStylePanel"
-
-interface Step5GeneratePreviewProps {
-  /* 标题样式 */
-  titleSettings: TitleSettings
-  onUpdatePosition: (position: string) => void
-  onUpdateFont: (font: string) => void
-  onUpdateSize: (size: number) => void
-  onToggleBold: () => void
-  onToggleItalic: () => void
-  onToggleStroke: () => void
-  onToggleShadow: () => void
-  onApplyPreset: (presetKey: string) => void
-  activePreset: string | null
-  titlePresets: { key: string; label: string; previewStyle: React.CSSProperties }[]
-}
-
-const Step5GeneratePreview: React.FC<Step5GeneratePreviewProps> = ({
-  titleSettings,
-  onUpdatePosition,
-  onUpdateFont,
-  onUpdateSize,
-  onToggleBold,
-  onToggleItalic,
-  onToggleStroke,
-  onToggleShadow,
-  onApplyPreset,
-  activePreset,
-  titlePresets,
-}) => {
-  return (
-    <div className="xx-form-section">
-      <h3>🎬 预览设置</h3>
-
-      {/* 前端实时预览提示 */}
-      <div
-        className="xx-preview-tip"
-        style={{
-          display: "flex",
-          alignItems: "center",
-          gap: 8,
-          padding: "12px 16px",
-          background: "rgba(59, 130, 246, 0.08)",
-          borderRadius: 8,
-          marginBottom: 16,
-          border: "1px solid rgba(59, 130, 246, 0.15)",
-        }}
-      >
-        <PlayCircleOutlined style={{ fontSize: 18, color: "#3b82f6" }} />
-        <span style={{ fontSize: 13, color: "var(--text-secondary, #666)" }}>
-          右侧面板直接播放素材片段，调整标题样式可实时预览效果
-        </span>
-      </div>
-
-      {/* 标题样式面板 */}
-      <TitleStylePanel
-        settings={titleSettings}
-        onUpdatePosition={onUpdatePosition}
-        onUpdateFont={onUpdateFont}
-        onUpdateSize={onUpdateSize}
-        onToggleBold={onToggleBold}
-        onToggleItalic={onToggleItalic}
-        onToggleStroke={onToggleStroke}
-        onToggleShadow={onToggleShadow}
-        onApplyPreset={onApplyPreset}
-        activePreset={activePreset}
-        titlePresets={titlePresets}
-        POSITION_OPTIONS={POSITION_OPTIONS}
-        FONT_OPTIONS={FONT_OPTIONS}
-      />
-    </div>
-  )
-}
-
-export default Step5GeneratePreview

apps/web/src/pages/generate/components/Step5VoiceSelect.tsx

@@ -17,6 +17,16 @@ interface Step5VoiceSelectProps {
 }
 
 /** 格式化时长 mm:ss */
+/** 获取素材实际时长（优先顶层 duration，fallback 到 metadata.duration） */
+const getDuration = (item: AssetItem): number => {
+  return item.duration ?? (item.metadata?.duration as number) ?? 0
+}
+
+/** 获取素材实际文件大小 */
+const getFileSize = (item: AssetItem): number => {
+  return item.file_size ?? (item.metadata?.file_size as number) ?? 0
+}
+
 const formatDuration = (seconds?: number): string => {
   if (!seconds || seconds <= 0) return "00:00"
   const m = Math.floor(seconds / 60)
@@ -89,7 +99,7 @@ const Step5VoiceSelect: React.FC<Step5VoiceSelectProps> = ({
       // 如果启用了时长校验，且配音时长不足
       if (totalVideoDuration > 0) {
         const material = materials.find((m) => m.id === id)
-        if (material && (material.duration || 0) < totalVideoDuration) {
+        if (material && getDuration(material) < totalVideoDuration) {
           setPendingVoiceId(id)
           setDurationWarningOpen(true)
           return
@@ -271,25 +281,24 @@ const Step5VoiceSelect: React.FC<Step5VoiceSelectProps> = ({
                 }}
               >
                 <span style={{ display: "flex", alignItems: "center", gap: 4 }}>
-                  {formatDuration(item.duration)}
-                  {totalVideoDuration > 0 &&
-                    (Number(item.duration) || 0) < Number(totalVideoDuration) && (
-                      <span
-                        style={{
-                          color: "#ff4d4f",
-                          fontSize: 11,
-                          fontWeight: 500,
-                          display: "inline-flex",
-                          alignItems: "center",
-                          gap: 2,
-                        }}
-                      >
-                        <WarningOutlined />
-                        时长不足
-                      </span>
-                    )}
+                  {formatDuration(getDuration(item))}
+                  {totalVideoDuration > 0 && getDuration(item) < Number(totalVideoDuration) && (
+                    <span
+                      style={{
+                        color: "#ff4d4f",
+                        fontSize: 11,
+                        fontWeight: 500,
+                        display: "inline-flex",
+                        alignItems: "center",
+                        gap: 2,
+                      }}
+                    >
+                      <WarningOutlined />
+                      时长不足
+                    </span>
+                  )}
                 </span>
-                <span>{formatFileSize(item.file_size)}</span>
+                <span>{formatFileSize(getFileSize(item))}</span>
               </div>
             </div>
           )
@@ -318,7 +327,9 @@ const Step5VoiceSelect: React.FC<Step5VoiceSelectProps> = ({
           return (
             <p>
               该配音时长（
-              <strong>{pendingMaterial ? formatDuration(pendingMaterial.duration) : "--"}</strong>
+              <strong>
+                {pendingMaterial ? formatDuration(getDuration(pendingMaterial)) : "--"}
+              </strong>
               ）短于视频总时长（
               <strong>{formatDuration(totalVideoDuration)}</strong>
               ），播放时配音可能提前结束，建议选择更长的配音素材。

apps/web/src/pages/generate/components/Step6CoverSettings.tsx

@@ -14,6 +14,22 @@ interface Step6CoverSettingsProps {
   assetIds?: string[]
   /** 当前选中的模板 ID */
   selectedTemplate?: string
+  /** Step4 标题设置，用于预览视频烧录标题 & 封面叠加标题 */
+  titleSettings?: import("../types").TitleSettings
+  /** 预览任务创建回调——将 task_id 暴露给父组件供 confirmGeneration 复用 */
+  onPreviewTaskCreated?: (taskId: string) => void
+  /** 从预览响应中提取到 source_edit_plan_id 时的回调 */
+  onSourceEditPlanIdExtracted?: (planId: string) => void
+  /** 配音模式 */
+  voiceMode?: "preset" | "custom" | "clone"
+  /** 选中的配音素材 ID */
+  selectedVoice?: string
+  /** 选中的克隆音色 ID */
+  selectedClonedVoice?: string
+  /** BGM 开关 */
+  bgm?: boolean
+  /** BGM 配置 */
+  bgmConfig?: { enabled: boolean; music_id?: string }
 }
 
 const Step6CoverSettings: React.FC<Step6CoverSettingsProps> = (props) => {
@@ -40,6 +56,14 @@ const Step6CoverSettings: React.FC<Step6CoverSettingsProps> = (props) => {
     duration: props.duration,
     assetIds: props.assetIds,
     selectedTemplate: props.selectedTemplate,
+    titleSettings: props.titleSettings,
+    onPreviewTaskCreated: props.onPreviewTaskCreated,
+    onSourceEditPlanIdExtracted: props.onSourceEditPlanIdExtracted,
+    voiceMode: props.voiceMode,
+    selectedVoice: props.selectedVoice,
+    selectedClonedVoice: props.selectedClonedVoice,
+    bgm: props.bgm,
+    bgmConfig: props.bgmConfig,
   })
 
   const handleAutoGenerate = () => {

apps/web/src/pages/generate/components/Step7ConfirmGenerate.tsx

@@ -24,8 +24,6 @@ interface Step7ConfirmGenerateProps {
   presetVoices: PresetVoiceItem[]
   clonedVoices: VoiceClone[]
   coverSettings: CoverConfig
-  generateCount: number
-  onGenerateCountChange: (count: number) => void
   generating: boolean
   generated: boolean
   generateError: string | null
@@ -42,9 +40,6 @@ const Step7ConfirmGenerate: React.FC<Step7ConfirmGenerateProps> = (props) => {
     title,
     voiceName,
     coverSummary,
-    generateCount,
-    handleDecrement,
-    handleIncrement,
     generating,
     generated,
     generateError,
@@ -65,10 +60,6 @@ const Step7ConfirmGenerate: React.FC<Step7ConfirmGenerateProps> = (props) => {
         title={title}
         voiceName={voiceName}
         coverSummary={coverSummary}
-        generateCount={generateCount}
-        generating={generating}
-        onDecrement={handleDecrement}
-        onIncrement={handleIncrement}
       />
       <GenerationStatus
         generating={generating}

apps/web/src/pages/generate/components/material/ManualMaterialList.tsx

@@ -1,7 +1,8 @@
 /**
- * 手动选择素材列表
+ * 手动选择素材列表 — 竖屏 9:16 卡片网格
+ * 交互：默认显示封面，点击播放按钮播放，播放中隐藏按钮，点击视频区域暂停
  */
-import React, { useRef, useCallback } from "react"
+import React, { useRef, useState, useCallback } from "react"
 import { Typography } from "antd"
 import type { AssetItem } from "@/api/assets"
 
@@ -14,39 +15,241 @@ interface ManualMaterialListProps {
   onToggle: (materialId: string) => void
 }
 
+/** 秒数格式化为 mm:ss */
+const fmtDuration = (seconds?: number): string => {
+  if (!seconds && seconds !== 0) return "--:--"
+  const m = Math.floor(seconds / 60)
+  const s = Math.floor(seconds % 60)
+  return `${String(m).padStart(2, "0")}:${String(s).padStart(2, "0")}`
+}
+
+/** 单个素材卡片 */
+const MaterialCard: React.FC<{
+  asset: AssetItem
+  checked: boolean
+  onToggle: () => void
+}> = ({ asset, checked, onToggle }) => {
+  const videoRef = useRef<HTMLVideoElement>(null)
+  const [isPlaying, setIsPlaying] = useState(false)
+  const isVideo = asset.mime_type?.startsWith("video/") ?? false
+  const thumbSrc = asset.thumbnail_url || undefined
+
+  const handlePlayToggle = useCallback(
+    (e: React.MouseEvent) => {
+      e.stopPropagation()
+      const video = videoRef.current
+      if (!video || !isVideo) return
+
+      if (isPlaying) {
+        video.pause()
+        setIsPlaying(false)
+      } else {
+        video.play().catch(() => {})
+        setIsPlaying(true)
+      }
+    },
+    [isPlaying, isVideo],
+  )
+
+  const handleVideoEnded = useCallback(() => {
+    setIsPlaying(false)
+  }, [])
+
+  const handleCardClick = useCallback(() => {
+    // 如果视频正在播放，点击卡片空白区域暂停视频
+    if (isPlaying) {
+      const video = videoRef.current
+      if (video) {
+        video.pause()
+        setIsPlaying(false)
+      }
+      return
+    }
+    onToggle()
+  }, [isPlaying, onToggle])
+
+  return (
+    <div
+      data-testid="material-card"
+      onClick={handleCardClick}
+      style={{
+        position: "relative",
+        aspectRatio: "9 / 16",
+        borderRadius: 10,
+        overflow: "hidden",
+        cursor: "pointer",
+        border: checked ? "2px solid var(--primary-color, #4f46e5)" : "2px solid transparent",
+        boxShadow: checked ? "0 0 0 2px rgba(79, 70, 229, 0.2)" : "0 1px 3px rgba(0, 0, 0, 0.1)",
+        background: "#1e293b",
+        transition: "all 0.15s ease",
+      }}
+    >
+      {/* 视频元素 */}
+      {isVideo && asset.file_url ? (
+        <video
+          ref={videoRef}
+          src={asset.file_url}
+          poster={thumbSrc}
+          muted
+          loop
+          playsInline
+          preload="metadata"
+          onEnded={handleVideoEnded}
+          style={{
+            width: "100%",
+            height: "100%",
+            objectFit: "cover",
+            display: "block",
+          }}
+        />
+      ) : thumbSrc ? (
+        <img
+          src={thumbSrc}
+          alt={asset.name}
+          style={{
+            width: "100%",
+            height: "100%",
+            objectFit: "cover",
+            display: "block",
+          }}
+          onError={(e) => {
+            const target = e.target as HTMLImageElement
+            target.style.display = "none"
+          }}
+        />
+      ) : (
+        <div
+          style={{
+            width: "100%",
+            height: "100%",
+            display: "flex",
+            alignItems: "center",
+            justifyContent: "center",
+            background: "linear-gradient(135deg, #334155, #1e293b)",
+            color: "rgba(255,255,255,0.5)",
+            fontSize: 28,
+          }}
+        >
+          {isVideo ? "🎬" : "🎵"}
+        </div>
+      )}
+
+      {/* 底部渐变遮罩 */}
+      <div
+        style={{
+          position: "absolute",
+          bottom: 0,
+          left: 0,
+          right: 0,
+          height: "50%",
+          background: "linear-gradient(0deg, rgba(0,0,0,0.6) 0%, transparent 100%)",
+          pointerEvents: "none",
+        }}
+      />
+
+      {/* 播放按钮 — 播放中隐藏 */}
+      {!isPlaying && (
+        <div
+          onClick={handlePlayToggle}
+          style={{
+            position: "absolute",
+            top: "50%",
+            left: "50%",
+            transform: "translate(-50%, -50%)",
+            width: 36,
+            height: 36,
+            borderRadius: "50%",
+            background: "rgba(99, 102, 241, 0.85)",
+            display: "flex",
+            alignItems: "center",
+            justifyContent: "center",
+            zIndex: 3,
+            transition: "opacity 0.2s ease",
+          }}
+        >
+          <svg width="16" height="16" viewBox="0 0 24 24" fill="white">
+            <path d="M8 5v14l11-7z" />
+          </svg>
+        </div>
+      )}
+
+      {/* 文件名（左下角） */}
+      <div
+        style={{
+          position: "absolute",
+          bottom: 6,
+          left: 6,
+          right: 50,
+          color: "white",
+          fontSize: 11,
+          fontWeight: 500,
+          whiteSpace: "nowrap",
+          overflow: "hidden",
+          textOverflow: "ellipsis",
+          textShadow: "0 1px 2px rgba(0,0,0,0.5)",
+          pointerEvents: "none",
+          zIndex: 1,
+        }}
+      >
+        {asset.name}
+      </div>
+
+      {/* 时长（右下角） */}
+      <div
+        style={{
+          position: "absolute",
+          bottom: 6,
+          right: 6,
+          background: "rgba(0, 0, 0, 0.7)",
+          color: "white",
+          padding: "1px 5px",
+          borderRadius: 3,
+          fontSize: 10,
+          fontWeight: 600,
+          fontVariantNumeric: "tabular-nums",
+          pointerEvents: "none",
+          zIndex: 1,
+        }}
+      >
+        {fmtDuration(asset.duration)}
+      </div>
+
+      {/* 选中勾选标记（左上角） */}
+      {checked && (
+        <div
+          data-testid="material-card-check"
+          aria-label="已选中"
+          style={{
+            position: "absolute",
+            top: 6,
+            left: 6,
+            width: 20,
+            height: 20,
+            borderRadius: "50%",
+            background: "var(--primary-color, #4f46e5)",
+            display: "flex",
+            alignItems: "center",
+            justifyContent: "center",
+            color: "white",
+            fontSize: 12,
+            fontWeight: 700,
+            zIndex: 2,
+            pointerEvents: "none",
+          }}
+        >
+          ✓
+        </div>
+      )}
+    </div>
+  )
+}
+
 const ManualMaterialList: React.FC<ManualMaterialListProps> = ({
   materials,
   materialsLoading,
   selectedMaterials,
   onToggle,
 }) => {
-  // 追踪当前正在播放的视频元素，确保同时只有一个视频播放
-  const activeVideoRef = useRef<HTMLVideoElement | null>(null)
-
-  const handleVideoMouseEnter = useCallback((e: React.MouseEvent<HTMLVideoElement>) => {
-    const video = e.currentTarget
-    // 暂停之前正在播放的视频（检查是否仍在 DOM 中）
-    if (
-      activeVideoRef.current &&
-      activeVideoRef.current !== video &&
-      document.body.contains(activeVideoRef.current)
-    ) {
-      activeVideoRef.current.pause()
-      activeVideoRef.current.currentTime = 0
-    }
-    activeVideoRef.current = video
-    video.play().catch(() => {})
-  }, [])
-
-  const handleVideoMouseLeave = useCallback((e: React.MouseEvent<HTMLVideoElement>) => {
-    const video = e.currentTarget
-    video.pause()
-    video.currentTime = 0
-    if (activeVideoRef.current === video) {
-      activeVideoRef.current = null
-    }
-  }, [])
-
   return (
     <div style={{ marginTop: 14 }}>
       {materialsLoading ? (
@@ -56,145 +259,21 @@ const ManualMaterialList: React.FC<ManualMaterialListProps> = ({
           暂无素材，请先在视频库中上传
         </Text>
       ) : (
-        <div style={{ display: "flex", flexDirection: "column", gap: 6 }}>
-          {materials.items.map((m) => {
-            const checked = selectedMaterials.includes(m.id)
-            const isVideo = m.mime_type?.startsWith("video/") ?? false
-            const thumbSrc = m.thumbnail_url || undefined
-            return (
-              <label
-                key={m.id}
-                style={{
-                  display: "flex",
-                  alignItems: "center",
-                  gap: 10,
-                  padding: "8px 12px",
-                  background: checked ? "var(--primary-soft, #eef2ff)" : "#f8fafc",
-                  borderRadius: 10,
-                  cursor: "pointer",
-                  border: checked
-                    ? "1px solid var(--primary-color, #4f46e5)"
-                    : "1px solid transparent",
-                  transition: "all 0.15s ease",
-                }}
-              >
-                <input
-                  type="checkbox"
-                  checked={checked}
-                  onChange={() => onToggle(m.id)}
-                  style={{
-                    accentColor: "var(--primary-color, #4f46e5)",
-                    flexShrink: 0,
-                  }}
-                />
-                {/* 缩略图预览 48×48 */}
-                <div
-                  style={{
-                    width: 48,
-                    height: 48,
-                    borderRadius: 6,
-                    overflow: "hidden",
-                    background: "#e2e8f0",
-                    flexShrink: 0,
-                    display: "flex",
-                    alignItems: "center",
-                    justifyContent: "center",
-                  }}
-                >
-                  {isVideo && m.file_url ? (
-                    <video
-                      src={m.file_url}
-                      poster={m.thumbnail_url || undefined}
-                      muted
-                      loop
-                      playsInline
-                      preload="none"
-                      onMouseEnter={handleVideoMouseEnter}
-                      onMouseLeave={handleVideoMouseLeave}
-                      style={{
-                        width: "100%",
-                        height: "100%",
-                        objectFit: "cover",
-                        cursor: "pointer",
-                      }}
-                    />
-                  ) : thumbSrc ? (
-                    <img
-                      src={thumbSrc}
-                      alt={m.name}
-                      style={{
-                        width: "100%",
-                        height: "100%",
-                        objectFit: "cover",
-                      }}
-                      onError={(e) => {
-                        const target = e.target as HTMLImageElement
-                        target.style.display = "none"
-                        const fallback = target.nextElementSibling as HTMLElement | null
-                        if (fallback) fallback.style.display = "flex"
-                      }}
-                    />
-                  ) : null}
-                  {!thumbSrc && !isVideo && (
-                    <span
-                      style={{
-                        fontSize: 20,
-                        opacity: 0.5,
-                        display: "flex",
-                      }}
-                    >
-                      🎵
-                    </span>
-                  )}
-                  {!thumbSrc && isVideo && !m.file_url && (
-                    <span
-                      style={{
-                        fontSize: 20,
-                        opacity: 0.5,
-                        display: "flex",
-                      }}
-                    >
-                      🎬
-                    </span>
-                  )}
-                  {/* img onError 时显示的 fallback（初始隐藏） */}
-                  {thumbSrc && !(isVideo && m.file_url) && (
-                    <span
-                      style={{
-                        fontSize: 20,
-                        opacity: 0.5,
-                        display: "none",
-                      }}
-                    >
-                      {isVideo ? "🎬" : "🎵"}
-                    </span>
-                  )}
-                </div>
-                <span
-                  style={{
-                    fontSize: 13,
-                    color: "var(--text-primary)",
-                    flex: 1,
-                    minWidth: 0,
-                    overflow: "hidden",
-                    textOverflow: "ellipsis",
-                    whiteSpace: "nowrap",
-                  }}
-                >
-                  {m.name}
-                </span>
-                <span
-                  style={{
-                    fontSize: 11,
-                    color: "var(--text-tertiary, #94a3b8)",
-                    flexShrink: 0,
-                  }}
-                >
-                  {m.mime_type?.split("/")?.[1]?.toUpperCase() ?? "FILE"}
-                </span>
-              </label>
-            )
-          })}
+        <div
+          style={{
+            display: "grid",
+            gridTemplateColumns: "repeat(auto-fill, minmax(110px, 1fr))",
+            gap: 10,
+          }}
+        >
+          {materials.items.map((asset) => (
+            <MaterialCard
+              key={asset.id}
+              asset={asset}
+              checked={selectedMaterials.includes(asset.id)}
+              onToggle={() => onToggle(asset.id)}
+            />
+          ))}
         </div>
       )}
     </div>

apps/web/src/pages/generate/components/step7-confirm/SummaryCard.tsx

@@ -1,5 +1,4 @@
 import React from "react"
-import { MinusOutlined, PlusOutlined } from "@ant-design/icons"
 
 interface SummaryCardProps {
   templateName: string
@@ -7,10 +6,6 @@ interface SummaryCardProps {
   title: string
   voiceName: string
   coverSummary: string
-  generateCount: number
-  generating: boolean
-  onDecrement: () => void
-  onIncrement: () => void
 }
 
 const SummaryCard: React.FC<SummaryCardProps> = ({
@@ -19,10 +14,6 @@ const SummaryCard: React.FC<SummaryCardProps> = ({
   title,
   voiceName,
   coverSummary,
-  generateCount,
-  generating,
-  onDecrement,
-  onIncrement,
 }) => {
   return (
     <div className="xx-summary-card">
@@ -46,29 +37,6 @@ const SummaryCard: React.FC<SummaryCardProps> = ({
         <span className="xx-summary-label">封面</span>
         <span className="xx-summary-value">{coverSummary}</span>
       </div>
-      <div className="xx-summary-row">
-        <span className="xx-summary-label">生成数量</span>
-        <span className="xx-summary-value">
-          <div className="xx-count-stepper">
-            <button
-              className="xx-count-stepper-btn"
-              disabled={generateCount <= 1 || generating}
-              onClick={onDecrement}
-            >
-              <MinusOutlined />
-            </button>
-            <span className="xx-count-stepper-value">{generateCount}</span>
-            <button
-              className="xx-count-stepper-btn"
-              disabled={generateCount >= 10 || generating}
-              onClick={onIncrement}
-            >
-              <PlusOutlined />
-            </button>
-            <span className="xx-count-stepper-hint">条视频</span>
-          </div>
-        </span>
-      </div>
     </div>
   )
 }

apps/web/src/pages/generate/components/title/TitlePresetsGrid.tsx

@@ -2,6 +2,7 @@
  * 标题预设样式网格
  */
 import React from "react"
+import { getFontFamily } from "../../constants"
 
 interface TitlePresetItem {
   key: string
@@ -35,7 +36,7 @@ const TitlePresetsGrid: React.FC<TitlePresetsGridProps> = ({
           >
             <span
               className="xx-title-preset-preview-text"
-              style={{ ...p.previewStyle, ...(fontFamily ? { fontFamily } : {}) }}
+              style={{ ...p.previewStyle, fontFamily: getFontFamily(fontFamily || "思源黑体") }}
             >
               标题
             </span>

apps/web/src/pages/generate/components/title/TitleStylePanel.tsx

@@ -94,7 +94,7 @@ const TitleStylePanel: React.FC<TitleStylePanelProps> = ({
           className="xx-slider"
           type="range"
           min={12}
-          max={48}
+          max={128}
           value={settings.size}
           onChange={(e) => onUpdateSize(Number(e.target.value))}
         />

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

@@ -33,9 +33,8 @@ export const STEPS = [
   { key: 2, label: "选择素材" },
   { key: 3, label: "选择配音" },
   { key: 4, label: "选择标题" },
-  { key: 5, label: "生成预览" },
-  { key: 6, label: "选择封面" },
-  { key: 7, label: "确认生成" },
+  { key: 5, label: "选择封面" },
+  { key: 6, label: "确认生成" },
 ]
 
 /* ── 标题位置选项 ── */
@@ -56,6 +55,21 @@ export const FONT_OPTIONS = [
   "华康俪金黑",
 ]
 
+/* ── 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> = {
+  思源黑体: '"Source Han Sans SC", "Noto Sans SC", "PingFang SC", "Microsoft YaHei", sans-serif',
+  思源宋体: '"Source Han Serif SC", "Noto Serif SC", "Songti SC", "SimSun", serif',
+  苹方: '"PingFang SC", -apple-system, "Helvetica Neue", sans-serif',
+  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 {
+  return FONT_FAMILY_MAP[font] || FONT_FAMILY_MAP["思源黑体"]
+}
+
 /* ── 标题样式预设 ── */
 export const TITLE_PRESETS = [
   {

apps/web/src/pages/generate/generate.css

@@ -113,6 +113,14 @@
   align-items: start;
 }
 
+.xx-generate-layout.full-width {
+  grid-template-columns: 1fr;
+}
+
+.xx-generate-layout.full-width .xx-generate-right-col {
+  display: none;
+}
+
 /* ============================================================
    左侧表单区 generate-form
    ============================================================ */
@@ -186,16 +194,16 @@
    ============================================================ */
 .xx-choice-list {
   display: grid;
-  grid-template-columns: repeat(3, 1fr);
-  gap: 12px;
+  grid-template-columns: repeat(6, 1fr);
+  gap: 8px;
 }
 
 .xx-choice-item {
   position: relative;
   background: var(--bg-primary);
   border: 2px solid var(--border-color);
-  border-radius: var(--radius-md);
-  padding: 14px;
+  border-radius: var(--radius-sm);
+  padding: 6px;
   cursor: pointer;
   transition: 0.18s ease;
   text-align: center;
@@ -211,18 +219,20 @@
 }
 
 .xx-choice-thumb {
-  height: 60px;
+  width: 33%;
+  max-width: 52px;
+  height: 24px;
   border-radius: var(--radius-sm);
   display: grid;
   place-items: center;
   color: var(--text-inverse);
-  font-size: 24px;
+  font-size: 12px;
   font-weight: 700;
-  margin-bottom: 10px;
+  margin: 0 auto 4px;
 }
 
 .xx-choice-item h4 {
-  margin: 0 0 4px;
+  margin: 0 0 2px;
   font-size: 13px;
   font-weight: 600;
   color: var(--text-primary);
@@ -230,7 +240,7 @@
 
 .xx-choice-item p {
   margin: 0;
-  font-size: 11px;
+  font-size: 12px;
   color: var(--text-tertiary);
 }
 
@@ -1055,7 +1065,7 @@
   }
 
   .xx-choice-list {
-    grid-template-columns: repeat(2, 1fr);
+    grid-template-columns: repeat(4, 1fr);
   }
 
   .xx-voice-choice-list {
@@ -1280,53 +1290,6 @@
   }
 }
 
-/* ── 生成数量步进器 ── */
-.xx-count-stepper {
-  display: inline-flex;
-  align-items: center;
-  gap: 6px;
-}
-
-.xx-count-stepper-btn {
-  display: inline-flex;
-  align-items: center;
-  justify-content: center;
-  width: 28px;
-  height: 28px;
-  border: 1px solid var(--border-primary, #e2e8f0);
-  border-radius: 8px;
-  background: var(--bg-surface, #fff);
-  color: var(--text-secondary, #64748b);
-  font-size: 13px;
-  cursor: pointer;
-  transition: all 0.15s;
-}
-
-.xx-count-stepper-btn:hover:not(:disabled) {
-  border-color: var(--primary-400, #818cf8);
-  color: var(--primary-600, #4f46e5);
-  background: var(--primary-50, #eef2ff);
-}
-
-.xx-count-stepper-btn:disabled {
-  opacity: 0.35;
-  cursor: not-allowed;
-}
-
-.xx-count-stepper-value {
-  min-width: 24px;
-  text-align: center;
-  font-size: 16px;
-  font-weight: 600;
-  color: var(--text-primary, #1e293b);
-}
-
-.xx-count-stepper-hint {
-  font-size: 12px;
-  color: var(--text-tertiary, #94a3b8);
-  margin-left: 2px;
-}
-
 /* ── 素材选择模式切换 Tab ── */
 .xx-material-mode-tabs {
   display: flex;
@@ -2309,6 +2272,8 @@
 .xx-cover-preview-box {
   position: relative;
   aspect-ratio: 9 / 16;
+  max-width: 180px;
+  margin: 0 auto;
   background: var(--bg-tertiary);
   border-radius: var(--radius-md);
   overflow: hidden;
@@ -2523,9 +2488,25 @@
 .xx-generate-right-col {
   display: flex;
   flex-direction: column;
+  align-items: center;
   gap: 16px;
 }
 
+/* ── 内联视频播放器（右侧） ── */
+.xx-inline-video-player {
+  width: 100%;
+  max-width: 320px;
+  background: var(--bg-surface, #fff);
+  border: 1px solid var(--border-primary, #e2e8f0);
+  border-radius: 16px;
+  padding: 16px;
+  box-shadow: 0 2px 8px rgba(0, 0, 0, 0.04);
+}
+
+.xx-inline-video-player video {
+  background: #000;
+}
+
 .xx-preview-header {
   display: flex;
   align-items: center;

apps/web/src/pages/generate/hooks/generate-video/types.ts

@@ -18,7 +18,17 @@ export interface UseGenerateVideoProps {
   duration: number
   autoSubtitles: boolean
   bgm: boolean
-  generateCount: number
+  /** 当前草稿 ID（URL 参数 edit_plan_id，用于后端回写任务关联） */
+  sourceEditPlanId?: string | null
+  /** 预览任务 ID（由 useStep6Cover 创建后写入，供 confirmGeneration 复用预览产物） */
+  previewTaskId?: string | null
+  /** BGM 配置（来自模板 bgm_config，受 bgm 开关控制） */
+  bgmConfig?: {
+    enabled: boolean
+    music_id?: string
+  }
+  /** 生成成功后的回调（用于清除持久化的 previewTaskId 等状态） */
+  onGenerationSuccess?: () => void
 }
 
 /** 生成阶段 */

apps/web/src/pages/generate/hooks/generate-video/useGenerationPolling.ts

@@ -1,87 +1,146 @@
 import { useRef, useCallback } from "react"
 import { message } from "antd"
-import { getGenerationStatus, getGenerationTaskResults } from "@/api/template-editor"
+import axios from "axios"
+import { getGenerationTask } from "@/api/tasks/tasks"
+import { getGenerationTaskResults } from "@/api/template-editor"
 import { safeExtractError } from "./errorUtils"
 
 interface UseGenerationPollingOptions {
-  templateId: string
   onProgress: (progress: number) => void
   onComplete: (videos: unknown[]) => void
   onFailed: (errorMsg: string) => void
 }
 
+/** 最大连续错误次数（仅对可重试错误），超过后终止轮询 */
+const MAX_RETRYABLE_ERRORS = 10
+/** 获取结果的最大重试次数 */
+const MAX_RESULTS_RETRIES = 3
+
 /**
- * 生成状态轮询 Hook
- * 轮询生成状态，更新进度，处理完成/失败
+ * 生成状态轮询 Hook（v2 — 改用 /generation/tasks/{task_id}）
+ *
+ * 旧版轮询 GET /templates/{id}/editor/generation-status 依赖 plan 维度状态，
+ * 在编辑流程数据链路断裂时拿不到 task_id。新版直接使用 POST /generation/tasks
+ * 返回的 task_id 轮询任务详情，不再依赖 plan。
+ *
+ * 错误处理：
+ * - 4xx（尤其 404）视为不可恢复，立即 onFailed，不再重试
+ * - 5xx / 网络错误重试，最多连续 MAX_RETRYABLE_ERRORS 次
+ * - 任务完成后获取结果失败会重试 MAX_RESULTS_RETRIES 次，仍失败则 onFailed
  */
 export const useGenerationPolling = ({
-  templateId,
   onProgress,
   onComplete,
   onFailed,
 }: UseGenerationPollingOptions) => {
   const progressTimer = useRef<ReturnType<typeof setTimeout>>()
+  const cancelledRef = useRef(false)
 
   const clearTimer = useCallback(() => {
+    cancelledRef.current = true
     if (progressTimer.current) {
       clearTimeout(progressTimer.current)
       progressTimer.current = undefined
     }
   }, [])
 
-  const startPolling = useCallback(() => {
-    const poll = async () => {
+  /** 任务完成后拉取结果列表，带重试 */
+  const fetchResultsWithRetry = useCallback(
+    async (taskId: string, attempt = 0): Promise<unknown[] | null> => {
       try {
-        const data = await getGenerationStatus(templateId)
-
-        if (data.plan_status === "completed") {
-          onProgress(100)
-          // 获取生成的视频结果
-          let videos: unknown[] = []
-          if (data.generation_task_id) {
-            try {
-              videos = await getGenerationTaskResults(data.generation_task_id)
-            } catch (err) {
-              console.error("[获取生成结果失败]", err)
-            }
-          }
-          onComplete(videos)
-          message.success("视频生成完成！")
-          return
+        return await getGenerationTaskResults(taskId)
+      } catch (err) {
+        if (cancelledRef.current) return null
+        console.error(`[获取生成结果失败] 第 ${attempt + 1} 次`, err)
+        if (attempt < MAX_RESULTS_RETRIES - 1) {
+          await new Promise((resolve) => setTimeout(resolve, 1000 * (attempt + 1)))
+          return fetchResultsWithRetry(taskId, attempt + 1)
         }
-        if (data.plan_status === "failed") {
-          const dataAny = data as unknown as Record<string, unknown>
-          const rawMsg =
-            dataAny.error_message ||
-            dataAny.error ||
-            dataAny.message ||
-            (Array.isArray(data.clips)
-              ? (data.clips as { status: string; error_message?: string }[]).find(
-                  (c) => c.status === "failed",
-                )?.error_message
-              : undefined) ||
-            "视频生成失败，请联系管理员或重试"
-          const errorMsg = safeExtractError(rawMsg)
-          console.error("[生成失败] templateId:", templateId, "响应:", data)
-          onFailed(errorMsg)
-          message.error(errorMsg)
-          return
-        }
-
-        const clips = data.clips || []
-        const total = clips.length || 1
-        const done = (clips as { status: string }[]).filter((c) => c.status === "completed").length
-        onProgress(Math.round((done / total) * 100))
-
-        progressTimer.current = setTimeout(poll, 2000)
-      } catch (pollErr) {
-        console.error("[轮询出错] templateId:", templateId, pollErr)
-        progressTimer.current = setTimeout(poll, 3000)
+        return null
       }
-    }
+    },
+    [],
+  )
 
-    progressTimer.current = setTimeout(poll, 2000)
-  }, [templateId, onProgress, onComplete, onFailed])
+  const startPolling = useCallback(
+    (taskId: string) => {
+      cancelledRef.current = false
+      let consecutiveErrors = 0
+
+      const poll = async () => {
+        if (cancelledRef.current) return
+        try {
+          const task = await getGenerationTask(taskId)
+          consecutiveErrors = 0
+
+          if (task.status === "completed") {
+            onProgress(100)
+            const videos = await fetchResultsWithRetry(taskId)
+            if (cancelledRef.current) return
+            if (videos === null) {
+              const errorMsg = "视频已生成，但获取结果列表失败，请稍后在任务列表查看"
+              console.error("[生成结果获取失败] taskId:", taskId)
+              onFailed(errorMsg)
+              message.error(errorMsg)
+              return
+            }
+            onComplete(videos)
+            message.success("视频生成完成！")
+            return
+          }
+
+          if (task.status === "failed" || task.status === "cancelled") {
+            const rawMsg =
+              task.error_info?.error_message ||
+              task.error_message ||
+              (task.status === "cancelled" ? "任务已取消" : "视频生成失败，请联系管理员或重试")
+            const errorMsg = safeExtractError(rawMsg)
+            console.error("[生成失败] taskId:", taskId, "响应:", task)
+            onFailed(errorMsg)
+            message.error(errorMsg)
+            return
+          }
+
+          // pending / waiting / running — 继续轮询
+          const pct = Math.max(0, Math.min(99, Math.round(Number(task.progress) || 0)))
+          onProgress(pct)
+          progressTimer.current = setTimeout(poll, 2000)
+        } catch (pollErr) {
+          if (cancelledRef.current) return
+          console.error("[轮询出错] taskId:", taskId, pollErr)
+
+          // 4xx 不可恢复，立即失败
+          const status = axios.isAxiosError(pollErr) ? pollErr.response?.status : undefined
+          if (status && status >= 400 && status < 500) {
+            const msg =
+              (axios.isAxiosError(pollErr) &&
+                (pollErr.response?.data as { detail?: string; message?: string } | undefined)
+                  ?.detail) ||
+              (axios.isAxiosError(pollErr) &&
+                (pollErr.response?.data as { detail?: string; message?: string } | undefined)
+                  ?.message) ||
+              `查询任务失败 (${status})`
+            const errorMsg = safeExtractError(msg)
+            onFailed(errorMsg)
+            message.error(errorMsg)
+            return
+          }
+
+          consecutiveErrors += 1
+          if (consecutiveErrors >= MAX_RETRYABLE_ERRORS) {
+            const errorMsg = "任务状态查询连续失败，请稍后在任务列表查看结果"
+            onFailed(errorMsg)
+            message.error(errorMsg)
+            return
+          }
+          progressTimer.current = setTimeout(poll, 3000)
+        }
+      }
+
+      progressTimer.current = setTimeout(poll, 1500)
+    },
+    [onProgress, onComplete, onFailed, fetchResultsWithRetry],
+  )
 
   return { startPolling, clearTimer }
 }

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

@@ -9,8 +9,6 @@ import { createFile } from "mp4box"
 import type { Movie, Sample } from "mp4box"
 
 // ── 常量 ──
-/** 初始化预解码最大帧数（约 2 秒 @30fps），后续帧通过 decodeAroundPosition 按需解码 */
-const MAX_INIT_FRAMES = 60
 
 /**
  * 规范化 mp4box 提取的 codec 字符串为 WebCodecs 兼容格式
@@ -111,7 +109,7 @@ class FrameQueue {
   private frames: FrameEntry[] = []
   private maxSize: number
 
-  constructor(maxSize = 5) {
+  constructor(maxSize = 200) {
     this.maxSize = maxSize
   }
 
@@ -123,24 +121,36 @@ class FrameQueue {
     this.frames.push(entry)
   }
 
-  /** 获取当前时间戳应显示的帧 */
+  /** 获取当前时间戳应显示的帧（二分查找，O(log n)） */
   getCurrentFrame(timestamp: number): VideoFrame | null {
-    let best: FrameEntry | null = null
-    let bestIdx = -1
-    for (let i = 0; i < this.frames.length; i++) {
-      const f = this.frames[i]
-      if (f.pts <= timestamp + 0.01) {
-        best = f
-        bestIdx = i
+    if (this.frames.length === 0) return null
+
+    const target = timestamp + 0.01
+
+    // 找到最后一个 pts <= target 的帧（右边界）
+    let lo = 0,
+      hi = this.frames.length - 1,
+      bestIdx = -1
+    while (lo <= hi) {
+      const mid = (lo + hi) >> 1
+      if (this.frames[mid].pts <= target) {
+        bestIdx = mid
+        lo = mid + 1
+      } else {
+        hi = mid - 1
       }
     }
+
+    if (bestIdx < 0) return null
+
+    // 关闭并移除 bestIdx 之前的所有已播放帧
     for (let i = 0; i < bestIdx; i++) {
       this.frames[i].frame.close()
     }
-    if (bestIdx >= 0) {
-      this.frames = this.frames.slice(bestIdx)
-    }
-    return best?.frame ?? null
+    this.frames.splice(0, bestIdx)
+
+    // 此时 bestIdx 对应帧已在索引 0
+    return this.frames[0]?.frame ?? null
   }
 
   clear() {
@@ -229,9 +239,10 @@ export function useCanvasPlayer(
     shadow?: boolean
   },
   onError?: (error: Error) => void,
+  enabled: boolean = true,
 ) {
   const [state, setState] = useState<CanvasPlayerState>({
-    hasSupport: isWebCodecsSupported(),
+    hasSupport: enabled && isWebCodecsSupported(),
     isPlaying: false,
     currentTime: 0,
     duration: 0,
@@ -243,9 +254,13 @@ export function useCanvasPlayer(
 
   // ── 内部引用 ──
   const decoderRef = useRef<VideoDecoder | null>(null)
-  const frameQueueRef = useRef(new FrameQueue(600))
-  /** 已解码的片段索引集合，用于按需解码（先标记防重入，失败时移除允许重试） */
-  const decodedSegmentsRef = useRef(new Set<number>())
+  const frameQueueRef = useRef(new FrameQueue(200))
+  /** 每个片段持久化解码器，避免每次新建导致关键帧错误 */
+  const segmentDecodersRef = useRef(new Map<number, VideoDecoder>())
+  /** 每个片段已送入解码器的 sample 游标（用于续解码） */
+  const segmentSampleCursorRef = useRef(new Map<number, number>())
+  /** 后台补充解码是否正在运行（防重入） */
+  const isFeedingRef = useRef(false)
   /** 解码代数计数器，seek 时递增以作废正在进行的异步解码 */
   const decodeGenerationRef = useRef(0)
   const rafRef = useRef<number>(0)
@@ -475,115 +490,91 @@ export function useCanvasPlayer(
     [segments, extractCodecDescription],
   )
 
-  // ── 初始化 VideoDecoder 并解码指定片段 ──
-  const decodeSegment = useCallback(
-    async (_buffer: ArrayBuffer, meta: SegmentMeta, maxFrames?: number): Promise<void> => {
-      if (isDestroyedRef.current) return
+  // ── 解码片段的一批帧（使用持久化解码器，支持从断点续解码） ──
+  /**
+   * @param segIdx 片段索引
+   * @param maxFrames 本次最多解码多少帧
+   */
+  const decodeSegmentBatch = useCallback(
+    async (segIdx: number, maxFrames: number = 60): Promise<number> => {
+      if (isDestroyedRef.current) return 0
+      const metas = segmentMetaRef.current
+      const meta = metas[segIdx]
+      if (!meta) return 0
+      const buffer = segmentDataRef.current.get(meta.assetId)
+      if (!buffer) return 0
 
-      let decoderReady = false
+      const gen = decodeGenerationRef.current
+      let decoder = segmentDecodersRef.current.get(segIdx)
+      let cursor = segmentSampleCursorRef.current.get(segIdx) ?? 0
+      const samples = meta.samples
 
-      // 配置解码器（每个片段可能需要不同的 codec/分辨率）
-      const decoder = new VideoDecoder({
-        output: (frame: VideoFrame) => {
-          // 从第一帧获取实际尺寸
-          if (videoDimRef.current.width === 0 || videoDimRef.current.height === 0) {
-            videoDimRef.current = { width: frame.codedWidth, height: frame.codedHeight }
-            console.log(
-              `[useCanvasPlayer] Actual frame size: ${frame.codedWidth}x${frame.codedHeight}`,
-            )
-          }
-          const localTime = frame.timestamp / 1_000_000
-          const globalTime = localTime + meta.globalStartTime
-          frameQueueRef.current.push({
-            frame,
-            pts: globalTime,
-            duration: (frame.duration ?? 0) / 1_000_000,
+      // 如果还没有解码器，新建一个（从关键帧开始，不会报 key frame 错误）
+      if (!decoder || decoder.state === "closed") {
+        decoder = new VideoDecoder({
+          output: (frame: VideoFrame) => {
+            if (videoDimRef.current.width === 0 || videoDimRef.current.height === 0) {
+              videoDimRef.current = { width: frame.codedWidth, height: frame.codedHeight }
+            }
+            const localTime = frame.timestamp / 1_000_000
+            const globalTime = localTime + meta.globalStartTime
+            frameQueueRef.current.push({
+              frame,
+              pts: globalTime,
+              duration: (frame.duration ?? 0) / 1_000_000,
+            })
+          },
+          error: (e: DOMException) => {
+            console.error(`[useCanvasPlayer] Segment ${segIdx} decoder error:`, e)
+            // 重置该片段的解码器和游标，允许重试
+            segmentDecodersRef.current.delete(segIdx)
+            segmentSampleCursorRef.current.set(segIdx, 0)
+          },
+        })
+
+        try {
+          await decoder.configure({
+            codec: meta.codec,
+            ...(meta.description ? { description: meta.description } : {}),
           })
-        },
-        error: (e: DOMException) => {
-          console.error("[useCanvasPlayer] Decoder error callback:", e)
-          const error = new Error(`VideoDecoder error: ${e.message || e.name || "unknown"}`)
+        } catch (err) {
+          console.error(`[useCanvasPlayer] Segment ${segIdx} configure failed:`, err)
+          const error = err instanceof Error ? err : new Error(String(err))
           setState((s) => ({
             ...s,
             isBuffering: false,
             hasDecodeError: true,
-            errorMessage: `视频解码器错误: ${e.message || "解码异常"}`,
+            errorMessage: `视频解码失败: ${error.message || "不支持的编解码器"}`,
           }))
           onErrorRef.current?.(error)
-        },
-      })
+          return 0
+        }
 
-      console.log("[useCanvasPlayer] configure:", {
-        codec: meta.codec,
-        description: meta.description,
-        descriptionByteLength: meta.description?.byteLength,
-        videoWidth: meta.videoWidth,
-        videoHeight: meta.videoHeight,
-        codecCharCodes: meta.codec.split("").map((c) => c.charCodeAt(0)),
-      })
-
-      try {
-        await decoder.configure({
-          codec: meta.codec,
-          ...(meta.description ? { description: meta.description } : {}),
-        })
-        decoderRef.current = decoder
-        decoderReady = true
-        // 标记缓冲结束，让 UI 开始渲染
+        segmentDecodersRef.current.set(segIdx, decoder)
+        cursor = 0
+        // 标记缓冲结束
         setState((s) => ({ ...s, isBuffering: false }))
-      } catch (err) {
-        const error = err instanceof Error ? err : new Error(String(err))
-        console.error(
-          "[useCanvasPlayer] Decoder configure failed for segment:",
-          meta.assetId,
-          error,
-        )
-        console.error("[useCanvasPlayer] Failed codec config:", {
-          codec: meta.codec,
-          descriptionByteLength: meta.description?.byteLength,
-          videoWidth: meta.videoWidth,
-          videoHeight: meta.videoHeight,
-        })
-        setState((s) => ({
-          ...s,
-          isBuffering: false,
-          isReady: false,
-          hasDecodeError: true,
-          errorMessage: `视频解码失败: ${error.message || "不支持的编解码器"}`,
-        }))
-        onErrorRef.current?.(error)
-        return
       }
 
-      if (!decoderReady) return
+      if (decoder.state !== "configured") return 0
 
-      // 使用 demuxSegment 中已提取并过滤的 samples（前端切片）
-      const samplesCollected = meta.samples
-      console.log(
-        `[useCanvasPlayer] Segment ${meta.assetId}: ${samplesCollected.length} samples to decode`,
-      )
-      if (samplesCollected.length === 0) {
-        console.warn("[useCanvasPlayer] No samples to decode for segment", meta.assetId)
-        return
-      }
-
-      // 送入解码器
-      let decodedCount = 0
-      let skippedCount = 0
-      let decodeErrors = 0
-      for (const sample of samplesCollected) {
-        if (!sample.data || isDestroyedRef.current) {
-          skippedCount++
+      // 从 cursor 继续喂 sample（流水线批量提交，不 await 单个 decode）
+      let decoded = 0
+      let si = cursor
+      while (si < samples.length && decoded < maxFrames) {
+        if (decodeGenerationRef.current !== gen || isDestroyedRef.current) break
+        if ((decoder.state as string) === "closed") break
+        // 帧队列快满时停止提交（这才是真正的背压）
+        if (frameQueueRef.current.size >= 180) break
+        // 解码器内部队列积压过多时短暂让出线程（阈值64，给硬件足够流水线深度）
+        if (decoder.decodeQueueSize > 64) {
+          await new Promise((r) => setTimeout(r, 5))
           continue
         }
-        if (decoder.state === "closed") break
-        // 初始化阶段限制解码帧数，避免帧缓冲溢出
-        if (maxFrames && decodedCount >= maxFrames) {
-          console.log(
-            `[useCanvasPlayer] Segment ${meta.assetId}: init decode limited to ${maxFrames} frames`,
-          )
-          break
-        }
+
+        const sample = samples[si]
+        si++
+        if (!sample.data) continue
 
         const chunk = new EncodedVideoChunk({
           type: sample.is_sync ? "key" : "delta",
@@ -593,93 +584,73 @@ export function useCanvasPlayer(
         })
 
         try {
-          await decoder.decode(chunk) // 修复：await 捕获异步错误
-          decodedCount++
+          decoder.decode(chunk)
+          decoded++
         } catch (e) {
-          decodeErrors++
-          console.warn(`[useCanvasPlayer] Decode chunk error (${decodeErrors}):`, e)
-          // 连续 3 次解码失败，放弃当前片段并报告错误
-          if (decodeErrors >= 3) {
-            console.error("[useCanvasPlayer] Too many decode errors, aborting segment")
-            const error = new Error(`视频解码连续失败 ${decodeErrors} 次，片段: ${meta.assetId}`)
-            setState((s) => ({
-              ...s,
-              isBuffering: false,
-              hasDecodeError: true,
-              errorMessage: `视频解码失败: 连续 ${decodeErrors} 次错误`,
-            }))
-            onErrorRef.current?.(error)
-            break
-          }
+          console.warn(`[useCanvasPlayer] Segment ${segIdx} decode error:`, e)
+          break
+        }
+      }
+
+      cursor = si
+      segmentSampleCursorRef.current.set(segIdx, cursor)
+      // 等待解码器输出帧（最多500ms）
+      if (decoded > 0 && (decoder.state as string) === "configured") {
+        let waited = 0
+        while (frameQueueRef.current.size < Math.min(decoded, 10) && waited < 500) {
+          await new Promise((r) => setTimeout(r, 20))
+          waited += 20
+          if (isDestroyedRef.current) break
         }
       }
       console.log(
-        `[useCanvasPlayer] Segment ${meta.assetId}: decoded ${decodedCount}, skipped ${skippedCount}, errors ${decodeErrors}, decoder.state=${decoder.state}`,
+        `[useCanvasPlayer] Segment ${segIdx} decoded ${decoded} frames, queue size: ${frameQueueRef.current.size}`,
       )
-
-      // flush 仅在解码器状态正常时执行
-      if (decoder.state === "configured") {
-        try {
-          await decoder.flush()
-          console.log(`[useCanvasPlayer] Segment ${meta.assetId}: flush complete`)
-        } catch (e) {
-          console.warn("[useCanvasPlayer] Decoder flush error:", e)
-        }
-      }
+      return decoded
     },
     [],
   )
 
+  // ── 后台持续补充帧 ──
   /**
-   * 按需解码当前播放位置 ±1 个片段。
-   * 在渲染循环中定期调用，避免一次性解码所有片段导致环形缓冲区溢出丢帧。
-   * 使用"先标记再解码"模式防止并发重复解码，失败时移除标记允许重试。
+   * 根据当前播放时间，确保队列中有足够缓冲
+   * 播放循环每 200ms 调用一次
    */
-  const decodeAroundPosition = useCallback(
+  const feedFrames = useCallback(
     async (currentTime: number) => {
-      const metas = segmentMetaRef.current
-      if (!metas || metas.length === 0) return
+      if (isFeedingRef.current) return
+      isFeedingRef.current = true
+      try {
+        const metas = segmentMetaRef.current
+        if (!metas || metas.length === 0) return
 
-      // 记录当前代数，seek 后代数变化则中止
-      const gen = decodeGenerationRef.current
+        // 队列帧数充足时不解码（目标：保持 >= 80 帧缓冲）
+        if (frameQueueRef.current.size >= 80) return
 
-      let targetIdx = -1
-      let acc = 0
-      for (let i = 0; i < metas.length; i++) {
-        const dur = metas[i].globalEndTime - metas[i].globalStartTime
-        if (currentTime < acc + dur) {
-          targetIdx = i
-          break
+        // 找到当前播放的片段
+        let targetIdx = 0
+        let acc = 0
+        for (let i = 0; i < metas.length; i++) {
+          const dur = metas[i].globalEndTime - metas[i].globalStartTime
+          if (currentTime < acc + dur) {
+            targetIdx = i
+            break
+          }
+          acc += dur
         }
-        acc += dur
-      }
-      if (targetIdx === -1) targetIdx = metas.length - 1
 
-      for (
-        let i = Math.max(0, targetIdx - 1);
-        i <= Math.min(metas.length - 1, targetIdx + 1);
-        i++
-      ) {
-        // seek 已作废当前解码任务
-        if (decodeGenerationRef.current !== gen) return
-        if (decodedSegmentsRef.current.has(i)) continue
-        const meta = metas[i]
-        const buffer = segmentDataRef.current.get(meta.assetId)
-        if (!buffer) continue
-        // 先标记为解码中，防止下一帧渲染时重复发起解码
-        decodedSegmentsRef.current.add(i)
-        try {
-          await decodeSegment(buffer, meta, 300)
-        } catch (e) {
-          // 解码失败则移除标记，允许后续重试
-          decodedSegmentsRef.current.delete(i)
-          console.warn(`[useCanvasPlayer] 按需解码片段 ${i} 失败:`, e)
+        // 依次补充：当前片段 → 下一个片段 → 再下一个
+        for (let offset = 0; offset <= 2; offset++) {
+          const idx = targetIdx + offset
+          if (idx >= metas.length) break
+          if (frameQueueRef.current.size >= 180) break
+          await decodeSegmentBatch(idx, 60)
         }
-        // await 后再次检查代数，seek 期间不更新标记
-        if (decodeGenerationRef.current !== gen) return
+      } finally {
+        isFeedingRef.current = false
       }
     },
-    [decodeSegment],
+    [decodeSegmentBatch],
   )
 
   // ── 标题绘制 ──
@@ -696,7 +667,6 @@ export function useCanvasPlayer(
 
       // 按 "/" 分割为多行（"/" 作为手动换行符）
       const lines = title.text.split(/[/／⁄∕]/)
-      console.log("[drawTitle] 原始标题:", JSON.stringify(title.text), "分割后:", lines)
       const lineHeight = fontSize * 1.3
       const totalHeight = lines.length * lineHeight
 
@@ -808,8 +778,10 @@ export function useCanvasPlayer(
         }
         return s
       })
-      // 按需解码当前 ±1 片段
-      decodeAroundPosition(currentTime)
+      // 后台补充帧：队列不足时自动续解码
+      if (frameQueueRef.current.size < 80) {
+        void feedFrames(currentTime)
+      }
     }
 
     if (currentTime >= totalDuration) {
@@ -818,29 +790,46 @@ export function useCanvasPlayer(
     }
 
     rafRef.current = requestAnimationFrame(renderFrame)
-  }, [canvasRef, totalDuration, titleSettings, drawTitle, computeDrawRect, decodeAroundPosition])
+  }, [canvasRef, totalDuration, titleSettings, drawTitle, computeDrawRect, feedFrames])
 
   // ── 播放控制 ──
   const play = useCallback(async () => {
     if (!state.hasSupport || isDestroyedRef.current) return
 
-    // 重播场景：currentTime 已回到起点但 decodedSegmentsRef 仍有旧标记
-    // 此时 FrameQueue 中旧帧已被淘汰，需清空标记让 decodeAroundPosition 重新解码
-    if (state.currentTime <= 0.1 && decodedSegmentsRef.current.size > 0) {
+    // 重播：必须关闭旧解码器、清空队列、重置游标，从头重新解码
+    if (state.currentTime >= totalDuration - 0.1 || state.currentTime <= 0.1) {
       decodeGenerationRef.current++
-      decodedSegmentsRef.current.clear()
-      // 同步清空帧缓冲，避免旧帧残留导致 getCurrentFrame 返回 null
+      // 关闭所有持久化解码器
+      for (const d of segmentDecodersRef.current.values()) {
+        try {
+          if (d.state !== "closed") d.close()
+        } catch {
+          /* noop */
+        }
+      }
+      segmentDecodersRef.current.clear()
+      segmentSampleCursorRef.current.clear()
       frameQueueRef.current.clear()
+      playStartOffsetRef.current = 0
+      setState((s) => ({ ...s, currentTime: 0 }))
+      // 重新初始化解码
+      const metas = segmentMetaRef.current
+      const initialDecodeCount = Math.min(metas.length, 2)
+      for (let i = 0; i < initialDecodeCount; i++) {
+        await decodeSegmentBatch(i, 60)
+      }
     }
 
     setState((s) => ({ ...s, isPlaying: true }))
     playStartRef.current = performance.now()
-    playStartOffsetRef.current = state.currentTime
+    if (state.currentTime < 0.1) {
+      playStartOffsetRef.current = 0
+    } else {
+      playStartOffsetRef.current = state.currentTime
+    }
     lastProgressUpdateRef.current = 0
     rafRef.current = requestAnimationFrame(renderFrame)
-    // 立即触发一次按需解码，不等渲染循环 200ms 节流
-    decodeAroundPosition(state.currentTime)
-  }, [state.hasSupport, state.currentTime, renderFrame, decodeAroundPosition])
+  }, [state.hasSupport, state.currentTime, totalDuration, renderFrame, decodeSegmentBatch])
 
   const pause = useCallback(() => {
     setState((s) => ({ ...s, isPlaying: false }))
@@ -850,33 +839,61 @@ export function useCanvasPlayer(
   const seek = useCallback(
     async (time: number) => {
       const clampedTime = Math.max(0, Math.min(time, totalDuration))
+      decodeGenerationRef.current++
+      // 关闭所有解码器、清空队列、重置游标
+      for (const d of segmentDecodersRef.current.values()) {
+        try {
+          if (d.state !== "closed") d.close()
+        } catch {
+          /* noop */
+        }
+      }
+      segmentDecodersRef.current.clear()
+      segmentSampleCursorRef.current.clear()
+      frameQueueRef.current.clear()
       setState((s) => ({ ...s, currentTime: clampedTime }))
       playStartOffsetRef.current = clampedTime
       playStartRef.current = performance.now()
-      // seek 时递增解码代数，作废正在进行的异步解码
-      decodeGenerationRef.current++
-      // 清空帧队列（clear 内部会 close 所有帧）+ 清空已解码标记
-      frameQueueRef.current.clear()
-      decodedSegmentsRef.current.clear()
-      await decodeAroundPosition(clampedTime)
+      // 找到 seek 目标片段，从该片段开始解码
+      const metas = segmentMetaRef.current
+      let targetIdx = 0,
+        acc = 0
+      for (let i = 0; i < metas.length; i++) {
+        const dur = metas[i].globalEndTime - metas[i].globalStartTime
+        if (clampedTime < acc + dur) {
+          targetIdx = i
+          break
+        }
+        acc += dur
+      }
+      await decodeSegmentBatch(targetIdx, 60)
+      await decodeSegmentBatch(Math.min(targetIdx + 1, metas.length - 1), 60)
     },
-    [totalDuration, decodeAroundPosition],
+    [totalDuration, decodeSegmentBatch],
   )
 
   const destroy = useCallback(() => {
     isDestroyedRef.current = true
     cancelAnimationFrame(rafRef.current)
 
+    // 关闭所有持久化解码器
+    for (const d of segmentDecodersRef.current.values()) {
+      try {
+        if (d.state !== "closed") d.close()
+      } catch {
+        /* noop */
+      }
+    }
+    segmentDecodersRef.current.clear()
+    segmentSampleCursorRef.current.clear()
+
     if (decoderRef.current && decoderRef.current.state !== "closed") {
       decoderRef.current.close()
     }
 
-    // 递增代数中止进行中的异步解码，清空帧队列（clear 内部 close 所有帧）
-    decodeGenerationRef.current++
     frameQueueRef.current.clear()
     segmentDataRef.current.clear()
     segmentMetaRef.current = []
-    decodedSegmentsRef.current.clear()
   }, [])
 
   // ── 预加载下一个片段的数据 ──
@@ -893,6 +910,7 @@ export function useCanvasPlayer(
 
   // ── 初始化：加载并解码所有片段 ──
   useEffect(() => {
+    if (!enabled) return
     if (!state.hasSupport || segments.length === 0) {
       console.log("[useCanvasPlayer] Skip init:", {
         hasSupport: state.hasSupport,
@@ -902,10 +920,24 @@ export function useCanvasPlayer(
     }
 
     let cancelled = false
-    console.log("[useCanvasPlayer] Init start, segments:", segments.length)
 
     const init = async () => {
-      setState((s) => ({ ...s, isBuffering: true }))
+      // ✅ 关键修复：重置销毁标记，允许新的 init 周期正常工作
+      // destroy() 在 useEffect cleanup 中被调用，将 isDestroyedRef 设为 true
+      // 如果不重置，后续的 loadSegment / decodeSegment 会立即 return
+      isDestroyedRef.current = false
+      // ✅ Strict Mode 修复：init 不再递增 generation
+      // seek() 和 play() 仍保留 generation 递增用于中止异步解码
+      // 重置错误状态，避免上一轮的解码错误影响新的 init 周期
+      setState((s) => ({
+        ...s,
+        isBuffering: true,
+        hasDecodeError: false,
+        errorMessage: "",
+        isReady: false,
+      }))
+
+      console.log("[useCanvasPlayer] Init start v2_DIAG, segments:", segments.length)
 
       // 1. 加载所有片段数据
       for (const seg of segments) {
@@ -947,32 +979,44 @@ export function useCanvasPlayer(
 
       segmentMetaRef.current = metas
 
-      // 3. 按需解码：初始只解码前 3 个片段，后续通过 decodeAroundPosition 动态加载
-      //    避免一次性全量解码导致 frameQueue 环形缓冲区旧帧被丢弃引发黑屏
-      decodedSegmentsRef.current.clear()
-      const initGen = decodeGenerationRef.current
-      const initialDecodeCount = Math.min(metas.length, 3)
-      for (let i = 0; i < initialDecodeCount; i++) {
-        if (cancelled) break
-        // seek 或 destroy 已作废当前初始化
-        if (decodeGenerationRef.current !== initGen) break
-        const meta = metas[i]
-        const buffer = segmentDataRef.current.get(meta.assetId)
-        if (!buffer) continue
-        // 先标记为解码中，防止重复解码
-        decodedSegmentsRef.current.add(i)
+      // 3. 初始化解码：关闭旧解码器，前 2 个片段各解 60 帧
+      //    后续由 feedFrames 后台补充
+      for (const d of segmentDecodersRef.current.values()) {
         try {
-          await decodeSegment(buffer, meta, MAX_INIT_FRAMES)
-        } catch (e) {
-          // 解码失败则移除标记，允许后续重试
-          decodedSegmentsRef.current.delete(i)
-          console.warn(`[useCanvasPlayer] 初始化解码片段 ${i} 失败:`, e)
+          if (d.state !== "closed") d.close()
+        } catch {
+          /* noop */
         }
       }
+      segmentDecodersRef.current.clear()
+      segmentSampleCursorRef.current.clear()
+      frameQueueRef.current.clear()
+
+      const initGen = decodeGenerationRef.current
+      const initialDecodeCount = Math.min(metas.length, 2)
+      console.log(
+        `[useCanvasPlayer] Starting init decode: ${initialDecodeCount} segments, metas: ${metas.length}`,
+      )
+      for (let i = 0; i < initialDecodeCount; i++) {
+        if (cancelled) break
+        if (decodeGenerationRef.current !== initGen) break
+        console.log(`[DIAG_v2] Init decode segment ${i}...`)
+        try {
+          await decodeSegmentBatch(i, 60)
+          console.log(`[DIAG_v2] Init decode segment ${i} done`)
+        } catch (e) {
+          console.warn(`[useCanvasPlayer] 初始化解码片段 ${i} 失败:`, e)
+        }
+        if (cancelled) break
+      }
+
+      console.log(`[useCanvasPlayer] Init decode finished, cancelled:`, cancelled)
 
       if (!cancelled) {
-        console.log("[useCanvasPlayer] Init complete, isReady = true")
+        console.log("[useCanvasPlayer] Init complete, isReady = true, duration:", totalDuration)
         setState((s) => ({ ...s, duration: totalDuration, isReady: true, isBuffering: false }))
+      } else {
+        console.warn("[useCanvasPlayer] Init was cancelled before completion")
       }
     }
 

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

@@ -0,0 +1,124 @@
+/**
+ * 草稿自动保存工具 Hook
+ *
+ * 背景：后端 PUT /templates/{id}/editor 的 config 是「整体替换」语义，
+ * 直接发送 { config: { asset_ids } } 会把 title 等其他字段覆盖掉。
+ * 本 Hook 统一执行「GET 当前 config → 浅合并新字段 → PUT 回去」，
+ * 并用串行队列 + AbortController 保证：
+ *   - 同一时刻只有一个保存请求在飞
+ *   - 快速连续变化时只提交最后一次
+ *   - 组件卸载时取消未完成请求
+ *   - 保存失败时保留补丁，自动重试（指数退避，最多 5 次）
+ *
+ * 保存失败只 console.warn，不弹窗、不阻塞。
+ */
+import { useCallback, useEffect, useRef } from "react"
+import { getEditPlan, updateEditPlan } from "@/api/template-editor"
+
+type ConfigPatch = Record<string, unknown>
+
+/** 最大自动重试次数 */
+const MAX_RETRIES = 5
+/** 初始重试延迟（ms），每次翻倍 */
+const BASE_RETRY_DELAY = 1000
+
+export function useDraftAutoSave(templateId?: string) {
+  const timerRef = useRef<ReturnType<typeof setTimeout>>()
+  const abortRef = useRef<AbortController | null>(null)
+  // 待合并的补丁队列（解决「保存进行中又来了新变化」）
+  const pendingPatchRef = useRef<ConfigPatch | null>(null)
+  const savingRef = useRef(false)
+  const templateIdRef = useRef(templateId)
+  templateIdRef.current = templateId
+
+  const flush = useCallback(async (retryCount = 0) => {
+    const tid = templateIdRef.current
+    if (!tid) return
+    // 已有保存在飞：把新补丁暂存，等当前请求结束后再合并一次
+    if (savingRef.current) return
+
+    // 快照当前补丁，但先不清空 —— 成功后才清除，失败时保留以便重试
+    const patchToSave = pendingPatchRef.current
+    if (!patchToSave) {
+      savingRef.current = false
+      return
+    }
+    savingRef.current = true
+
+    const controller = new AbortController()
+    abortRef.current = controller
+    try {
+      // 1. 读当前 config（拿最新，避免覆盖别人/别的步骤写入的字段）
+      const current = await getEditPlan(tid)
+      if (controller.signal.aborted) return
+      const merged = { ...(current.config || {}), ...patchToSave }
+      // 2. 写回完整合并后的 config
+      await updateEditPlan(tid, { config: merged }, controller.signal)
+      // 3. 保存成功才清除已保存的补丁
+      //    （保存期间可能有新补丁进来，只清除我们已经保存的部分）
+      pendingPatchRef.current = null
+    } catch (err) {
+      const name = (err as { name?: string })?.name
+      if (name === "CanceledError" || name === "AbortError") {
+        // 组件卸载或新请求取消，不重试
+        return
+      }
+      console.warn("[useDraftAutoSave] 自动保存草稿失败:", err)
+
+      // 保存失败：把本次尝试保存的补丁合并回 pendingPatchRef
+      // （保存期间可能有新补丁，新补丁优先）
+      pendingPatchRef.current = {
+        ...patchToSave,
+        ...(pendingPatchRef.current || {}),
+      }
+
+      // 指数退避重试
+      if (retryCount < MAX_RETRIES && !controller.signal.aborted) {
+        const delay = BASE_RETRY_DELAY * Math.pow(2, retryCount)
+        timerRef.current = setTimeout(() => {
+          void flush(retryCount + 1)
+        }, delay)
+      }
+      // 超过最大重试次数后，补丁仍保留在 pendingPatchRef 中，
+      // 下次 scheduleSave 触发时会一起带上
+    } finally {
+      savingRef.current = false
+      // 保存期间又积累了新变化（且不是在重试路径中），再触发一次
+      if (pendingPatchRef.current && !controller.signal.aborted && retryCount === 0) {
+        timerRef.current = setTimeout(() => {
+          void flush()
+        }, 50)
+      }
+    }
+  }, [])
+
+  /**
+   * 调度一次自动保存（防抖）
+   * @param patch 要合并进 config 的局部字段
+   * @param delay 防抖毫秒数
+   */
+  const scheduleSave = useCallback(
+    (patch: ConfigPatch, delay = 500) => {
+      const tid = templateIdRef.current
+      if (!tid) return
+      if (timerRef.current) clearTimeout(timerRef.current)
+      // 累计补丁（同一周期内多次变化合并成一次写入）
+      pendingPatchRef.current = { ...(pendingPatchRef.current || {}), ...patch }
+      timerRef.current = setTimeout(() => {
+        void flush()
+      }, delay)
+    },
+    [flush],
+  )
+
+  useEffect(() => {
+    return () => {
+      if (timerRef.current) clearTimeout(timerRef.current)
+      if (abortRef.current) abortRef.current.abort()
+    }
+  }, [])
+
+  return { scheduleSave }
+}
+
+export default useDraftAutoSave

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

@@ -4,8 +4,8 @@
  */
 import { useState } from "react"
 import { useSearchParams } from "react-router-dom"
-import type { GeneratedVideo } from "@/api/template-editor"
 import type { EditingTemplate } from "@/api/editing-planner"
+import type { EditPlanClip } from "@/api/template-editor"
 import type { CoverConfig } from "../../types/cover"
 import type { PresetVoiceItem } from "@/api/voices"
 import { DEFAULT_COVER_SETTINGS } from "../../constants"
@@ -14,11 +14,12 @@ import { useTemplateSelection } from "./useTemplateSelection"
 import { useTitleCoverSync } from "./useTitleCoverSync"
 import { useVoiceState } from "./useVoiceState"
 import { usePlanConfigLoader } from "./usePlanConfigLoader"
+import { usePersistedState } from "../usePersistedState"
 
 const DEFAULT_TITLE_SETTINGS: TitleSettings = {
   aiAutoSelect: false,
   title: "",
-  position: "bottom",
+  position: "top",
   font: "思源黑体",
   size: 28,
   bold: true,
@@ -46,6 +47,10 @@ export interface GenerateFormState {
   smartSelectedIds: string[]
   setSmartSelectedIds: (ids: string[]) => void
 
+  /* 服务端片段（/clips/from-assets 创建后获取） */
+  serverClips: EditPlanClip[]
+  setServerClips: (clips: EditPlanClip[]) => void
+
   /* 标题 */
   titleSettings: TitleSettings
   setTitleSettings: (settings: TitleSettings | ((prev: TitleSettings) => TitleSettings)) => void
@@ -67,10 +72,6 @@ export interface GenerateFormState {
   cloneModalOpen: boolean
   setCloneModalOpen: (open: boolean) => void
 
-  /* 生成数量 */
-  generateCount: number
-  setGenerateCount: (n: number) => void
-
   /* 高级设置 */
   videoRatio: string
   duration: number
@@ -82,11 +83,21 @@ export interface GenerateFormState {
   editPlanId: string | null
   planConfigStr: string | null
 
-  /* 预览弹窗 */
-  previewVideo: GeneratedVideo | null
-  setPreviewVideo: (v: GeneratedVideo | null) => void
-  previewModalOpen: boolean
-  setPreviewModalOpen: (open: boolean) => void
+  /**
+   * 传给 Worker 的 source_edit_plan_id。
+   * 仅使用 URL 中的 edit_plan_id（从剪辑模板编辑器跳转时携带）。
+   * URL 没有时传 null，后端正式生成 API 会通过 template_id+user_id 兜底查找正确的 plan。
+   * 注意：selectedTemplate 是模板 ID，不是 edit_plan_id，不能作为此值传递。
+   */
+  sourceEditPlanId: string | null
+
+  /** 预览任务 ID（由 useStep6Cover 创建后写入，供 useGenerateVideo 复用） */
+  previewTaskId: string | null
+  setPreviewTaskId: (id: string | null) => void
+
+  /** 从预览响应中提取的 source_edit_plan_id（供 fallback 路径使用） */
+  storedSourceEditPlanId: string | null
+  setStoredSourceEditPlanId: (planId: string | null) => void
 }
 
 export const useGenerateFormState = (): GenerateFormState => {
@@ -100,11 +111,19 @@ export const useGenerateFormState = (): GenerateFormState => {
   /* ── 模板选择 ── */
   const { selectedTemplate, setSelectedTemplate, userTemplates } = useTemplateSelection()
 
+  /* ── source_edit_plan_id：仅取 URL 参数，无则 null 让后端兜底 ── */
+  // selectedTemplate 是模板 ID 而非 edit_plan_id，不能混淆；
+  // 后端正式生成 API 会在 source_edit_plan_id 为空时通过 template_id+user_id 自动关联。
+  const sourceEditPlanId = editPlanId || null
+
   /* ── 素材 ── */
   const [selectedMaterials, setSelectedMaterials] = useState<string[]>([])
   const [materialMode, setMaterialMode] = useState<"manual" | "auto">("manual")
   const [smartSelectedIds, setSmartSelectedIds] = useState<string[]>([])
 
+  /* ── 服务端片段（供预览播放器使用）── */
+  const [serverClips, setServerClips] = useState<EditPlanClip[]>([])
+
   /* ── 标题设置 ── */
   const [titleSettings, setTitleSettings] = useState<TitleSettings>(DEFAULT_TITLE_SETTINGS)
 
@@ -133,9 +152,6 @@ export const useGenerateFormState = (): GenerateFormState => {
   /* ── 克隆声音弹窗 ── */
   const [cloneModalOpen, setCloneModalOpen] = useState(false)
 
-  /* ── 生成数量 ── */
-  const [generateCount, setGenerateCount] = useState(1)
-
   /* ── 高级设置（隐藏但保留） ── */
   const [videoRatio] = useState("9:16")
   const [duration] = useState(30)
@@ -143,9 +159,29 @@ export const useGenerateFormState = (): GenerateFormState => {
   const [autoSubtitles] = useState(true)
   const [bgm] = useState(true)
 
-  /* ── 预览弹窗 ── */
-  const [previewVideo, setPreviewVideo] = useState<GeneratedVideo | null>(null)
-  const [previewModalOpen, setPreviewModalOpen] = useState(false)
+  /* ── 预览任务 ID（useStep6Cover 创建预览时写入，useGenerateVideo 复用） ── */
+  // 持久化到 localStorage，key 按 editPlanId/templateId 区分，刷新页面后可恢复
+  const previewStorageKey = editPlanId
+    ? `preview_task_id_${editPlanId}`
+    : selectedTemplate
+      ? `preview_task_id_tpl_${selectedTemplate}`
+      : null
+  const [previewTaskId, setPreviewTaskId] = usePersistedState<string | null>(
+    previewStorageKey,
+    null,
+  )
+
+  /* ── 从预览响应中提取的 source_edit_plan_id（供 fallback 路径使用） ── */
+  // 持久化到 localStorage，刷新页面后 fallback 路径仍能正确传递 source_edit_plan_id
+  const planIdStorageKey = editPlanId
+    ? `source_edit_plan_id_${editPlanId}`
+    : selectedTemplate
+      ? `source_edit_plan_id_tpl_${selectedTemplate}`
+      : null
+  const [storedSourceEditPlanId, setStoredSourceEditPlanId] = usePersistedState<string | null>(
+    planIdStorageKey,
+    null,
+  )
 
   /* ── 从 URL / 编辑计划加载配置 ── */
   usePlanConfigLoader({
@@ -168,6 +204,8 @@ export const useGenerateFormState = (): GenerateFormState => {
     setMaterialMode,
     smartSelectedIds,
     setSmartSelectedIds,
+    serverClips,
+    setServerClips,
     titleSettings,
     setTitleSettings,
     coverSettings,
@@ -181,18 +219,17 @@ export const useGenerateFormState = (): GenerateFormState => {
     presetVoices,
     cloneModalOpen,
     setCloneModalOpen,
-    generateCount,
-    setGenerateCount,
     videoRatio,
     duration,
     style,
     autoSubtitles,
     bgm,
     editPlanId,
+    sourceEditPlanId,
     planConfigStr,
-    previewVideo,
-    setPreviewVideo,
-    previewModalOpen,
-    setPreviewModalOpen,
+    previewTaskId,
+    setPreviewTaskId,
+    storedSourceEditPlanId,
+    setStoredSourceEditPlanId,
   }
 }

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

@@ -4,16 +4,17 @@
  */
 import { useState, useCallback } from "react"
 import { message } from "antd"
-import type { GeneratedVideo } from "@/api/template-editor"
+import { type GeneratedVideo, getEditPlanClips, createClipsFromAssets } from "@/api/template-editor"
 import { createGenerationTask } from "@/api/tasks/tasks"
 import type { UseGenerateVideoProps } from "./generate-video/types"
 import { getGenerationPhase } from "./generate-video/phase"
 import { useGenerationPolling } from "./generate-video/useGenerationPolling"
 import { validateGenerateInputs } from "./generate-video/buildPayload"
+import { calculateResolution } from "../utils/calculateResolution"
 import { extractBackendError, translateError } from "./generate-video/errorUtils"
 
 export function useGenerateVideo(props: UseGenerateVideoProps) {
-  const { selectedTemplate } = props
+  const { selectedTemplate, onGenerationSuccess } = props
 
   /* ── 生成状态 ── */
   const [generating, setGenerating] = useState(false)
@@ -23,18 +24,21 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
   const [generatedVideos, setGeneratedVideos] = useState<GeneratedVideo[]>([])
 
   const handleProgress = useCallback((p: number) => setProgress(p), [])
-  const handleComplete = useCallback((videos: unknown[]) => {
-    setGenerating(false)
-    setGenerated(true)
-    setGeneratedVideos(videos as GeneratedVideo[])
-  }, [])
+  const handleComplete = useCallback(
+    (videos: unknown[]) => {
+      setGenerating(false)
+      setGenerated(true)
+      setGeneratedVideos(videos as GeneratedVideo[])
+      onGenerationSuccess?.()
+    },
+    [onGenerationSuccess],
+  )
   const handleFailed = useCallback((errorMsg: string) => {
     setGenerating(false)
     setGenerateError(errorMsg)
   }, [])
 
   const { startPolling, clearTimer } = useGenerationPolling({
-    templateId: selectedTemplate,
     onProgress: handleProgress,
     onComplete: handleComplete,
     onFailed: handleFailed,
@@ -55,68 +59,81 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
     clearTimer()
 
     try {
-      // 解析分辨率
-      const ratio = props.videoRatio || "9:16"
-      let outputWidth: number
-      let outputHeight: number
-
-      if (ratio.includes(":")) {
-        const [rw, rh] = ratio.split(":").map(Number)
-        if (rw > 0 && rh > 0) {
-          const [longSide, shortSide] = rw < rh ? [rh, rw] : [rw, rh]
-          const baseLong = 1920
-          const baseShort = Math.round((baseLong * shortSide) / longSide)
-          const evenShort = baseShort - (baseShort % 2)
-          if (rw < rh) {
-            outputWidth = evenShort
-            outputHeight = baseLong
-          } else {
-            outputWidth = baseLong
-            outputHeight = evenShort
-          }
-        } else {
-          outputWidth = 1080
-          outputHeight = 1920
-        }
-      } else if (ratio.includes("x")) {
-        const [wStr, hStr] = ratio.split("x")
-        outputWidth = parseInt(wStr, 10) || 1080
-        outputHeight = parseInt(hStr, 10) || 1920
-      } else {
-        outputWidth = 1080
-        outputHeight = 1920
-      }
+      const { width: outputWidth, height: outputHeight } = calculateResolution(
+        props.videoRatio || "9:16",
+      )
 
       const assetIds =
         props.materialMode === "auto" ? props.smartSelectedIds : props.selectedMaterials
 
-      // 直接创建正式生成任务
-      await createGenerationTask({
-        template_id: selectedTemplate,
-        asset_ids: assetIds,
-        output_width: outputWidth,
-        output_height: outputHeight,
-        cover_url: props.coverSettings?.upload_url || "",
-        custom_title: props.titleSettings?.title || "",
-        duration: props.duration || undefined,
-        video_ratio: props.videoRatio,
-        ...(props.titleSettings?.title
-          ? {
-              title_config: {
-                text: props.titleSettings.title,
-                font: props.titleSettings.font,
-                font_size: props.titleSettings.size,
-                font_color: props.titleSettings.color,
-                position: props.titleSettings.position,
-                bold: props.titleSettings.bold,
-                stroke: props.titleSettings.stroke,
-                shadow: props.titleSettings.shadow,
-              },
-            }
-          : {}),
-      })
+      // from-assets 已由 useStep2Materials 在用户选素材时（debounce 800ms）调用，
+      // 后端已改为异步秒级返回，这里做一次轻量兜底：
+      // 单次查 clips，已有则直接放行；没有则再调一次 from-assets。
+      if (assetIds.length > 0 && selectedTemplate) {
+        try {
+          const clipList = await getEditPlanClips(selectedTemplate, { limit: 500 })
+          if (clipList.items.length === 0) {
+            // 片段不存在（极端情况：useStep2Materials 的 debounce 还没触发）
+            // 手动补一次 from-assets（后端秒级返回）
+            await createClipsFromAssets(selectedTemplate, assetIds, "main")
+          }
+        } catch {
+          // 查询失败不阻塞，继续生成
+        }
+      }
 
-      startPolling()
+      const hide = message.loading("正在生成预览视频...", 0)
+
+      const coverUrl = props.coverSettings?.thumbnail_url || props.coverSettings?.upload_url || ""
+
+      const voiceLibraryId =
+        props.voiceMode === "clone"
+          ? props.selectedClonedVoice || props.selectedVoice || ""
+          : props.selectedVoice || ""
+
+      try {
+        const taskResp = await createGenerationTask({
+          template_id: selectedTemplate,
+          asset_ids: assetIds,
+          output_width: outputWidth,
+          output_height: outputHeight,
+          cover_url: coverUrl,
+          custom_title: props.titleSettings?.title || "",
+          duration: props.duration || undefined,
+          video_ratio: props.videoRatio,
+          voice_library_id: voiceLibraryId,
+          ...(props.selectedVoice && !voiceLibraryId ? { voice_ids: [props.selectedVoice] } : {}),
+          bgm_config: {
+            enabled: props.bgm !== false,
+            ...(props.bgmConfig?.music_id ? { preset_id: props.bgmConfig.music_id } : {}),
+          },
+          ...(props.sourceEditPlanId ? { source_edit_plan_id: props.sourceEditPlanId } : {}),
+          ...(props.titleSettings?.title
+            ? {
+                title_config: {
+                  text: props.titleSettings.title,
+                  font: props.titleSettings.font,
+                  font_size: props.titleSettings.size,
+                  font_color: props.titleSettings.color,
+                  position: props.titleSettings.position,
+                  bold: props.titleSettings.bold,
+                  stroke: props.titleSettings.stroke,
+                  shadow: props.titleSettings.shadow,
+                },
+              }
+            : {}),
+        })
+        hide()
+        const taskId = taskResp.items?.[0]?.id
+
+        if (!taskId) {
+          throw new Error("创建任务成功但未返回任务 ID，请稍后在任务列表查看")
+        }
+        startPolling(taskId)
+      } catch (err) {
+        hide()
+        throw err
+      }
     } catch (err: unknown) {
       console.error("[handleGenerate] 生成失败:", err)
       setGenerating(false)
@@ -128,18 +145,15 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
     }
   }, [props, clearTimer, startPolling, selectedTemplate])
 
-  /* 重新生成（失败后重试） */
   const retry = useCallback(() => {
     setGenerateError(null)
     generate()
   }, [generate])
 
-  /* 清除错误 */
   const dismissError = useCallback(() => {
     setGenerateError(null)
   }, [])
 
-  /* ── 下载视频 ── */
   const download = useCallback(async () => {
     if (!generatedVideos.length) return
     const video = generatedVideos[0]
@@ -160,7 +174,6 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
     }
   }, [generatedVideos])
 
-  /* ── 分享视频 ── */
   const share = useCallback(async () => {
     if (!generatedVideos.length) return
     const video = generatedVideos[0]
@@ -174,19 +187,16 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
   }, [generatedVideos])
 
   return {
-    // 状态
     generating,
     progress,
     generated,
     generateError,
     generatedVideos,
-    // 操作
     generate,
     retry,
     dismissError,
     download,
     share,
-    // 工具
     getGenerationPhase,
   }
 }

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

@@ -0,0 +1,63 @@
+import { useState, useEffect, useCallback, useRef } from "react"
+
+/**
+ * 持久化到 localStorage 的 state hook
+ *
+ * 用于在页面刷新后恢复 previewTaskId / sourceEditPlanId 等关键状态。
+ * 当 localStorage 不可用（SSR、隐私模式等）时自动降级为普通 useState。
+ * 当 key 变化时（例如切换模板），自动从新 key 重新读取并更新 state。
+ */
+export function usePersistedState<T>(
+  key: string | null | undefined,
+  defaultValue: T,
+): [T, (value: T | ((prev: T) => T)) => void] {
+  const storageKey = key ? `xiaoxia_${key}` : null
+  // defaultValue 用 ref 持有，避免作为 useEffect 依赖导致频繁重跑
+  const defaultValueRef = useRef(defaultValue)
+  defaultValueRef.current = defaultValue
+
+  const readFromStorage = useCallback((k: string | null): T => {
+    if (!k) return defaultValueRef.current
+    try {
+      const stored = localStorage.getItem(k)
+      if (stored !== null) {
+        return JSON.parse(stored) as T
+      }
+    } catch (e) {
+      // localStorage 不可用或 JSON 解析失败，使用默认值
+      console.warn("[usePersistedState] 读取 localStorage 失败:", e)
+    }
+    return defaultValueRef.current
+  }, [])
+
+  const [state, setState] = useState<T>(() => readFromStorage(storageKey))
+
+  // key 变化时（如切换模板/草稿），从新 key 重新读取，避免状态与存储不同步
+  useEffect(() => {
+    setState(readFromStorage(storageKey))
+  }, [storageKey, readFromStorage])
+
+  const setPersistedState = useCallback(
+    (value: T | ((prev: T) => T)) => {
+      setState((prev) => {
+        const nextValue = typeof value === "function" ? (value as (prev: T) => T)(prev) : value
+        if (storageKey) {
+          try {
+            if (nextValue === null || nextValue === undefined || nextValue === "") {
+              localStorage.removeItem(storageKey)
+            } else {
+              localStorage.setItem(storageKey, JSON.stringify(nextValue))
+            }
+          } catch (e) {
+            // localStorage 写入失败（存储空间满/隐私模式），静默忽略
+            console.warn("[usePersistedState] 写入 localStorage 失败:", e)
+          }
+        }
+        return nextValue
+      })
+    },
+    [storageKey],
+  )
+
+  return [state, setPersistedState]
+}

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

@@ -1,55 +1,37 @@
 /**
- * 素材片段调度器 Hook（多 video 元素方案 v2）
- * 每个片段对应一个独立 <video> 元素，全部预加载，通过 display 切换实现无缝播放
- * 替代单 video + 切 src 方案，消除片段切换延迟
+ * 素材片段调度器 Hook（多 video 元素方案 v3）
+ *
+ * v3 修复：
+ * - 所有动态状态存入 ref，tick 为稳定函数，彻底消除 RAF 闭包陷阱
+ * - 片段切换时先启动下一个 video 再切可见性，消除冻屏间隔
+ * - 进度更新 200ms 节流
  */
+
 import { useState, useRef, useCallback, useEffect, useMemo } from "react"
 
-/** 单个播放片段 */
 export interface PlaybackSegment {
-  /** 素材 ID */
   assetId: string
-  /** 素材视频 URL */
   videoUrl: string
-  /** 片段在素材中的入点（秒） */
   startTime: number
-  /** 片段在素材中的出点（秒） */
   endTime: number
-  /** 片段在时间线中的顺序 */
   order: number
 }
 
-/** 调度器返回 */
 export interface SegmentSchedulerState {
-  /** 是否正在播放 */
   isPlaying: boolean
-  /** 当前播放的全局时间（秒） */
   currentTime: number
-  /** 总时长（秒） */
   totalDuration: number
-  /** 当前片段索引 */
   currentSegmentIndex: number
-  /** 当前片段的本地播放时间 */
   segmentLocalTime: number
-  /** 是否已播完 */
   isEnded: boolean
-  /** 是否可以播放（至少有 1 个片段） */
   canPlay: boolean
-  /** 播放 */
   play: () => void
-  /** 暂停 */
   pause: () => void
-  /** 切换播放/暂停 */
   togglePlayPause: () => void
-  /** 跳转到全局时间 */
   seekTo: (time: number) => void
-  /** 每个片段对应的 video 元素 ref 数组 */
   videoRefs: React.MutableRefObject<(HTMLVideoElement | null)[]>
 }
 
-/**
- * 根据全局时间定位对应的片段和本地时间
- */
 function findSegmentAtTime(
   segments: PlaybackSegment[],
   globalTime: number,
@@ -66,9 +48,6 @@ function findSegmentAtTime(
   return { index: segments.length - 1, localTime: segments[segments.length - 1].endTime }
 }
 
-/**
- * 计算每个片段的全局起始时间
- */
 function buildTimeline(segments: PlaybackSegment[]): number[] {
   const starts: number[] = []
   let acc = 0
@@ -79,224 +58,248 @@ function buildTimeline(segments: PlaybackSegment[]): number[] {
   return starts
 }
 
-/**
- * useSegmentScheduler — 多 video 元素版素材片段调度器
- *
- * 核心改变：
- * - 每个片段对应一个独立 <video> 元素（由组件渲染，ref 传入）
- * - 所有 video 在挂载时即设置 src + preload="auto"，浏览器自动预加载
- * - 切换片段仅改 currentSegmentIndex + display，无需重新 load
- * - 实现无缝切换，无加载延迟
- */
 export function useSegmentScheduler(segments: PlaybackSegment[]): SegmentSchedulerState {
-  /** 每个片段对应的 video 元素 ref（由组件 JSX 渲染并绑定） */
   const videoRefs = useRef<(HTMLVideoElement | null)[]>([])
   const [isPlaying, setIsPlaying] = useState(false)
   const [currentTime, setCurrentTime] = useState(0)
   const [currentSegmentIndex, setCurrentSegmentIndex] = useState(0)
   const [isEnded, setIsEnded] = useState(false)
-  const rafRef = useRef<number>(0)
+  const rafRef = useRef(0)
   const isSeekingRef = useRef(false)
+  const lastTimeUpdateRef = useRef(0)
 
-  // 计算时间线
-  const timelineStarts = useMemo(() => buildTimeline(segments), [segments])
-  const totalDuration = useMemo(
+  // 所有动态值存入 ref，tick 始终读取最新值，不依赖闭包
+  const segIdxRef = useRef(0)
+  const segmentsRef = useRef(segments)
+  const timelineStartsData = useMemo(() => buildTimeline(segments), [segments])
+  const totalDurationData = useMemo(
     () => segments.reduce((sum, seg) => sum + (seg.endTime - seg.startTime), 0),
     [segments],
   )
+  const timelineStartsRef = useRef(timelineStartsData)
+  const totalDurationRef = useRef(totalDurationData)
+  const isPlayingRef = useRef(false)
+
+  segmentsRef.current = segments
+  timelineStartsRef.current = timelineStartsData
+  totalDurationRef.current = totalDurationData
 
   const canPlay = segments.length > 0
 
-  // 当前片段信息
-  const currentSegment = segments[currentSegmentIndex] || null
-  const segmentLocalTime = currentSegment
-    ? currentTime - (timelineStarts[currentSegmentIndex] || 0) + currentSegment.startTime
-    : 0
+  useEffect(() => {
+    segIdxRef.current = currentSegmentIndex
+  }, [currentSegmentIndex])
+
+  useEffect(() => {
+    isPlayingRef.current = isPlaying
+  }, [isPlaying])
+
+  const waitForReady = useCallback((video: HTMLVideoElement, timeout = 3000): Promise<void> => {
+    if (video.readyState >= 3) return Promise.resolve()
+    return new Promise((resolve) => {
+      const onCanPlay = () => {
+        video.removeEventListener("canplay", onCanPlay)
+        clearTimeout(timer)
+        resolve()
+      }
+      const timer = setTimeout(() => {
+        video.removeEventListener("canplay", onCanPlay)
+        resolve()
+      }, timeout)
+      video.addEventListener("canplay", onCanPlay)
+    })
+  }, [])
 
-  /**
-   * 切换到指定片段
-   * 不改变 src（video 已在 JSX 中设置），仅 seek + 等待可播
-   */
   const switchToSegment = useCallback(
-    (index: number, seekToLocalTime?: number): Promise<void> => {
-      return new Promise((resolve) => {
-        // 暂停当前视频
-        const prevVideo = videoRefs.current[currentSegmentIndex]
-        if (prevVideo) prevVideo.pause()
+    async (index: number, seekToLocalTime?: number) => {
+      const segs = segmentsRef.current
+      const video = videoRefs.current[index]
+      if (!video || index >= segs.length) return
 
-        const video = videoRefs.current[index]
-        if (!video || index >= segments.length) {
-          resolve()
-          return
-        }
+      const seg = segs[index]
+      const localTime = seekToLocalTime ?? seg.startTime
+      const oldIdx = segIdxRef.current
+      const oldVideo = videoRefs.current[oldIdx]
 
-        const seg = segments[index]
-        const localTime = seekToLocalTime ?? seg.startTime
+      if (oldVideo && oldVideo !== video) oldVideo.pause()
 
-        // 设置播放位置
+      if (!video.src && seg.videoUrl) {
+        video.src = seg.videoUrl
+        video.load()
+      }
+
+      if (Math.abs(video.currentTime - localTime) > 0.05) {
         video.currentTime = localTime
+      }
 
-        // 如果已有足够帧数据，直接 resolve
-        if (video.readyState >= 2) {
-          setCurrentSegmentIndex(index)
-          resolve()
-          return
-        }
+      segIdxRef.current = index
+      setCurrentSegmentIndex(index)
 
-        // 等待 canplay 事件
-        const onCanPlay = () => {
-          video.removeEventListener("canplay", onCanPlay)
-          clearTimeout(timeoutId)
-          setCurrentSegmentIndex(index)
-          resolve()
-        }
-
-        // 10 秒超时保护
-        const timeoutId = setTimeout(() => {
-          video.removeEventListener("canplay", onCanPlay)
-          console.warn(
-            `[useSegmentScheduler] 片段 ${index} 预加载超时 (10s), readyState=${video.readyState}`,
-          )
-          setCurrentSegmentIndex(index)
-          resolve()
-        }, 10000)
-
-        video.addEventListener("canplay", onCanPlay)
-      })
+      await waitForReady(video)
     },
-    [segments, currentSegmentIndex],
+    [waitForReady],
   )
 
-  /** 播放循环 — 检测片段边界并切换 */
+  // 稳定的 tick 函数，空依赖，所有值从 ref 读取
   const tick = useCallback(() => {
-    const video = videoRefs.current[currentSegmentIndex]
+    const segs = segmentsRef.current
+    const idx = segIdxRef.current
+    const video = videoRefs.current[idx]
+
     if (!video || isSeekingRef.current) {
       rafRef.current = requestAnimationFrame(tick)
       return
     }
 
-    const seg = segments[currentSegmentIndex]
+    const seg = segs[idx]
     if (!seg) return
 
-    // 检查是否到达出点（容差 0.15s）
-    if (video.currentTime >= seg.endTime - 0.15) {
-      video.pause()
-      const nextIndex = currentSegmentIndex + 1
-      if (nextIndex < segments.length) {
-        switchToSegment(nextIndex).then(() => {
-          setIsPlaying(true)
-          rafRef.current = requestAnimationFrame(tick)
-          const nextVideo = videoRefs.current[nextIndex]
-          if (nextVideo) {
-            const canPlay = () => {
-              nextVideo
-                .play()
-                .catch((e) =>
-                  console.warn("[useSegmentScheduler] auto-play next segment failed:", e),
-                )
-            }
-            if (nextVideo.readyState >= 3) {
-              canPlay()
-            } else {
-              const timeout = setTimeout(canPlay, 300)
-              nextVideo.addEventListener(
-                "canplay",
-                () => {
-                  clearTimeout(timeout)
-                  canPlay()
-                },
-                { once: true },
-              )
-            }
+    // 预加载下一个片段
+    const nextIndex = idx + 1
+    if (nextIndex < segs.length) {
+      const nextVideo = videoRefs.current[nextIndex]
+      if (nextVideo) {
+        const timeToEnd = seg.endTime - video.currentTime
+        if (timeToEnd <= 2 && nextVideo.readyState < 3) {
+          const nextSeg = segs[nextIndex]
+          if (Math.abs(nextVideo.currentTime - nextSeg.startTime) > 0.5) {
+            nextVideo.currentTime = nextSeg.startTime
           }
-        })
+        }
+      }
+    }
+
+    // 检测片段边界
+    if (video.currentTime >= seg.endTime - 0.1) {
+      if (nextIndex < segs.length) {
+        const nextVideo = videoRefs.current[nextIndex]
+        const nextSeg = segs[nextIndex]
         const accumulatedTime =
-          (timelineStarts[currentSegmentIndex] || 0) + (seg.endTime - seg.startTime)
+          (timelineStartsRef.current[idx] || 0) + (seg.endTime - seg.startTime)
+
+        if (nextVideo) {
+          if (Math.abs(nextVideo.currentTime - nextSeg.startTime) > 0.1) {
+            nextVideo.currentTime = nextSeg.startTime
+          }
+          // 先启动下一个视频（muted，可安全同时播放）
+          nextVideo
+            .play()
+            .catch((e) => console.warn("[useSegmentScheduler] next segment play failed:", e))
+        }
+
+        // 立即切换可见性
+        segIdxRef.current = nextIndex
+        setCurrentSegmentIndex(nextIndex)
         setCurrentTime(accumulatedTime)
+        lastTimeUpdateRef.current = 0
+        setIsPlaying(true)
+
+        // 下一帧暂停旧视频（让新视频先渲染，避免冻屏）
+        const oldVideo = video
+        requestAnimationFrame(() => {
+          oldVideo.pause()
+        })
+
+        rafRef.current = requestAnimationFrame(tick)
+        return
       } else {
+        video.pause()
         setIsPlaying(false)
         setIsEnded(true)
-        setCurrentTime(totalDuration)
+        setCurrentTime(totalDurationRef.current)
         return
       }
-    } else {
-      const globalTime =
-        (timelineStarts[currentSegmentIndex] || 0) + (video.currentTime - seg.startTime)
-      setCurrentTime(Math.max(0, Math.min(globalTime, totalDuration)))
+    }
+
+    const globalTime = (timelineStartsRef.current[idx] || 0) + (video.currentTime - seg.startTime)
+    const now = performance.now()
+    if (now - lastTimeUpdateRef.current >= 200) {
+      lastTimeUpdateRef.current = now
+      setCurrentTime(Math.max(0, Math.min(globalTime, totalDurationRef.current)))
     }
 
     rafRef.current = requestAnimationFrame(tick)
-  }, [segments, currentSegmentIndex, timelineStarts, totalDuration, switchToSegment])
+  }, [])
 
-  /** 播放 */
   const play = useCallback(async () => {
     if (!canPlay) return
-
     setIsEnded(false)
-
-    // 确保第一段可播放
-    const firstVideo = videoRefs.current[0]
-    if (firstVideo && currentSegmentIndex === 0 && firstVideo.readyState < 2) {
-      await switchToSegment(0)
-    }
-
-    const video = videoRefs.current[currentSegmentIndex]
+    const idx = segIdxRef.current
+    const video = videoRefs.current[idx]
     if (!video) return
 
-    try {
-      const playPromise = video.play()
-      if (playPromise !== undefined) {
-        await playPromise
+    if (idx === 0 && video.readyState < 2) {
+      if (!video.src && segmentsRef.current[0]?.videoUrl) {
+        video.src = segmentsRef.current[0].videoUrl
+        video.load()
       }
+      await waitForReady(video)
+    }
+
+    // 播放前 seek 到片段起始时间，确保 progress 计算正确
+    const seg = segmentsRef.current[idx]
+    if (seg && Math.abs(video.currentTime - seg.startTime) > 0.1) {
+      video.currentTime = seg.startTime
+    }
+
+    try {
+      await video.play()
       setIsPlaying(true)
+      cancelAnimationFrame(rafRef.current)
       rafRef.current = requestAnimationFrame(tick)
     } catch (err) {
       console.warn("[useSegmentScheduler] 播放失败:", err)
     }
-  }, [canPlay, switchToSegment, tick, currentSegmentIndex])
+  }, [canPlay, waitForReady, tick])
 
-  /** 暂停 */
   const pause = useCallback(() => {
-    const video = videoRefs.current[currentSegmentIndex]
+    const video = videoRefs.current[segIdxRef.current]
     if (video) video.pause()
     setIsPlaying(false)
     cancelAnimationFrame(rafRef.current)
-  }, [currentSegmentIndex])
+  }, [])
 
-  /** 切换播放/暂停 */
   const togglePlayPause = useCallback(() => {
-    if (isPlaying) {
+    if (isPlayingRef.current) {
       pause()
     } else {
       if (isEnded) {
-        // 播放结束后再次播放，从头开始
         setIsEnded(false)
-        switchToSegment(0, segments[0]?.startTime).then(() => {
-          const video = videoRefs.current[0]
-          if (video) {
-            video.play().catch((e) => console.warn("[useSegmentScheduler] restart play failed:", e))
-            setIsPlaying(true)
-            setCurrentTime(0)
-            rafRef.current = requestAnimationFrame(tick)
-          }
-        })
+        lastTimeUpdateRef.current = 0
+        const firstVideo = videoRefs.current[0]
+        if (firstVideo) {
+          videoRefs.current.forEach((v, i) => {
+            if (v && i !== 0) v.pause()
+          })
+          firstVideo.currentTime = segmentsRef.current[0]?.startTime || 0
+          segIdxRef.current = 0
+          setCurrentSegmentIndex(0)
+          setCurrentTime(0)
+          firstVideo
+            .play()
+            .then(() => {
+              setIsPlaying(true)
+              cancelAnimationFrame(rafRef.current)
+              rafRef.current = requestAnimationFrame(tick)
+            })
+            .catch((e) => console.warn("[useSegmentScheduler] restart failed:", e))
+        }
       } else {
         play()
       }
     }
-  }, [isPlaying, isEnded, pause, play, switchToSegment, segments, tick])
+  }, [isEnded, pause, play, tick])
 
-  /** 跳转到指定全局时间 */
   const seekTo = useCallback(
     async (time: number) => {
       if (!canPlay) return
-
-      const clampedTime = Math.max(0, Math.min(time, totalDuration))
-      const { index, localTime } = findSegmentAtTime(segments, clampedTime)
+      const clampedTime = Math.max(0, Math.min(time, totalDurationRef.current))
+      const { index, localTime } = findSegmentAtTime(segmentsRef.current, clampedTime)
 
       isSeekingRef.current = true
+      cancelAnimationFrame(rafRef.current)
 
-      if (index !== currentSegmentIndex) {
+      if (index !== segIdxRef.current) {
         await switchToSegment(index, localTime)
       } else {
         const video = videoRefs.current[index]
@@ -305,48 +308,54 @@ export function useSegmentScheduler(segments: PlaybackSegment[]): SegmentSchedul
 
       setCurrentTime(clampedTime)
       setIsEnded(false)
+      lastTimeUpdateRef.current = 0
+
+      if (isPlayingRef.current) {
+        const video = videoRefs.current[index]
+        if (video) {
+          video.play().catch(() => {})
+        }
+        rafRef.current = requestAnimationFrame(tick)
+      }
 
       setTimeout(() => {
         isSeekingRef.current = false
       }, 200)
     },
-    [canPlay, totalDuration, segments, currentSegmentIndex, switchToSegment],
+    [canPlay, switchToSegment, tick],
   )
 
-  // 确保 videoRefs 数组长度与 segments 一致 + 强制预加载
   useEffect(() => {
     videoRefs.current = videoRefs.current.slice(0, segments.length)
     while (videoRefs.current.length < segments.length) {
       videoRefs.current.push(null)
     }
-    // 强制预加载：所有 video 元素挂载后，调用 load() 确保浏览器真正开始加载数据
-    videoRefs.current.forEach((video) => {
-      if (video) {
-        video.load()
-      }
-    })
   }, [segments])
 
-  // 组件卸载时清理
   useEffect(() => {
     return () => {
       cancelAnimationFrame(rafRef.current)
     }
   }, [])
 
-  // 片段列表变化时重置
   useEffect(() => {
     cancelAnimationFrame(rafRef.current)
+    segIdxRef.current = 0
     setIsPlaying(false)
     setCurrentTime(0)
     setCurrentSegmentIndex(0)
     setIsEnded(false)
   }, [segments])
 
+  const currentSegment = segments[currentSegmentIndex] || null
+  const segmentLocalTime = currentSegment
+    ? currentTime - (timelineStartsRef.current[currentSegmentIndex] || 0) + currentSegment.startTime
+    : 0
+
   return {
     isPlaying,
     currentTime,
-    totalDuration,
+    totalDuration: totalDurationData,
     currentSegmentIndex,
     segmentLocalTime,
     isEnded,

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

@@ -0,0 +1,294 @@
+/**
+ * 服务器渲染预览 Hook
+ *
+ * 核心职责：
+ * 1. 调用 POST /generation/preview 创建服务器预览渲染任务
+ * 2. 轮询 GET /generation/preview/{task_id} 直到完成
+ * 3. 返回服务器渲染的真实视频 URL（供 <video> 标签播放）
+ * 4. 检测配置变更，标记预览失效（stale）或自动重新渲染
+ * 5. 网络错误自动重试 2 次
+ *
+ * 状态机：
+ *   idle → loading → ready → stale (config changed)
+ *                   ↘ failed → idle (retry)
+ */
+import { useState, useCallback, useRef, useEffect } from "react"
+import { createPreview, getPreviewStatus } from "@/api/generation/preview"
+import type { CreatePreviewRequest } from "@/api/generation/types"
+
+export type ServerPreviewStatus = "idle" | "loading" | "ready" | "stale" | "failed"
+
+interface UseServerPreviewOptions {
+  /** 是否启用预览（Step4+ 且有素材和模板时） */
+  enabled: boolean
+  /** 构建预览请求参数（每次 render 调用，获取最新配置） */
+  buildRequest: () => CreatePreviewRequest
+  /** 预览任务创建成功回调 */
+  onPreviewTaskCreated?: (taskId: string, sourceEditPlanId?: string) => void
+}
+
+interface UseServerPreviewReturn {
+  status: ServerPreviewStatus
+  videoUrl: string | null
+  error: string | null
+  /** 进度 0-100 */
+  progress: number
+  /** 手动触发预览创建（"重新预览"按钮或标题变更后手动刷新） */
+  triggerPreview: () => void
+  /** 当前预览任务 ID */
+  taskId: string | null
+}
+
+const POLL_INTERVAL = 2000
+const POLL_TIMEOUT = 120_000
+const MAX_NETWORK_RETRIES = 2
+
+/**
+ * 对配置参数做指纹，用于检测配置是否变化
+ */
+function buildFingerprint(req: CreatePreviewRequest): string {
+  return JSON.stringify({
+    t: req.template_id,
+    a: [...req.asset_ids].sort(),
+    d: req.duration,
+    r: req.video_ratio,
+    v: req.voice_library_id,
+    b: req.bgm_config,
+    title: req.title_config,
+  })
+}
+
+export function useServerPreview({
+  enabled,
+  buildRequest,
+  onPreviewTaskCreated,
+}: UseServerPreviewOptions): UseServerPreviewReturn {
+  const [status, setStatus] = useState<ServerPreviewStatus>("idle")
+  const [videoUrl, setVideoUrl] = useState<string | null>(null)
+  const [error, setError] = useState<string | null>(null)
+  const [progress, setProgress] = useState(0)
+  const [taskId, setTaskId] = useState<string | null>(null)
+
+  const pollTimerRef = useRef<ReturnType<typeof setTimeout> | null>(null)
+  const timeoutTimerRef = useRef<ReturnType<typeof setTimeout> | null>(null)
+  const requestSeqRef = useRef(0)
+  const renderedFingerprintRef = useRef<string>("")
+  const mountedRef = useRef(true)
+  const networkRetriesRef = useRef(0)
+
+  // 始终持有最新的 buildRequest 和回调
+  const buildRequestRef = useRef(buildRequest)
+  buildRequestRef.current = buildRequest
+  const onCreatedRef = useRef(onPreviewTaskCreated)
+  onCreatedRef.current = onPreviewTaskCreated
+
+  /* ── 清理 ── */
+  const clearTimers = useCallback(() => {
+    if (pollTimerRef.current) {
+      clearTimeout(pollTimerRef.current)
+      pollTimerRef.current = null
+    }
+    if (timeoutTimerRef.current) {
+      clearTimeout(timeoutTimerRef.current)
+      timeoutTimerRef.current = null
+    }
+  }, [])
+
+  useEffect(() => {
+    mountedRef.current = true
+    return () => {
+      mountedRef.current = false
+      clearTimers()
+    }
+  }, [clearTimers])
+
+  /* ── 创建预览 + 轮询 ── */
+  const createAndPoll = useCallback(
+    async (request: CreatePreviewRequest, seq: number) => {
+      setStatus("loading")
+      setProgress(0)
+      setError(null)
+      networkRetriesRef.current = 0
+
+      try {
+        const resp = await createPreview(request)
+        if (seq !== requestSeqRef.current || !mountedRef.current) return
+
+        setTaskId(resp.task_id)
+        onCreatedRef.current?.(resp.task_id, resp.source_edit_plan_id)
+
+        let completed = false
+
+        // 超时保护
+        timeoutTimerRef.current = setTimeout(() => {
+          if (completed || seq !== requestSeqRef.current || !mountedRef.current) return
+          completed = true
+          clearTimers()
+          setStatus("failed")
+          setError("预览渲染超时（120秒），请重试")
+        }, POLL_TIMEOUT)
+
+        const poll = async () => {
+          if (completed || seq !== requestSeqRef.current || !mountedRef.current) return
+
+          try {
+            const st = await getPreviewStatus(resp.task_id)
+            if (completed || seq !== requestSeqRef.current || !mountedRef.current) return
+
+            if (st.status === "completed" && st.video_url) {
+              completed = true
+              clearTimers()
+              renderedFingerprintRef.current = buildFingerprint(request)
+              setVideoUrl(st.video_url)
+              setProgress(100)
+              setStatus("ready")
+              setError(null)
+              return
+            }
+
+            if (st.status === "failed" || st.status === "cancelled") {
+              completed = true
+              clearTimers()
+              setStatus("failed")
+              setError(
+                st.status === "cancelled"
+                  ? "预览任务已取消"
+                  : st.error_message || "预览渲染失败，请重试",
+              )
+              return
+            }
+
+            // pending / generating
+            if (typeof st.progress === "number") setProgress(st.progress)
+            pollTimerRef.current = setTimeout(poll, POLL_INTERVAL)
+          } catch (pollErr) {
+            if (completed || seq !== requestSeqRef.current || !mountedRef.current) return
+            if (networkRetriesRef.current < MAX_NETWORK_RETRIES) {
+              networkRetriesRef.current += 1
+              console.warn(
+                `[ServerPreview] 轮询网络错误，第 ${networkRetriesRef.current} 次重试`,
+                pollErr,
+              )
+              pollTimerRef.current = setTimeout(poll, POLL_INTERVAL * 2)
+            } else {
+              completed = true
+              clearTimers()
+              setStatus("failed")
+              setError("网络错误，无法获取预览状态，请重试")
+            }
+          }
+        }
+
+        poll()
+      } catch (createErr) {
+        if (seq !== requestSeqRef.current || !mountedRef.current) return
+        console.error("[ServerPreview] 创建预览任务失败:", createErr)
+
+        const isNetworkError =
+          !!(createErr as { request?: unknown })?.request ||
+          (createErr as { code?: string })?.code === "ERR_NETWORK"
+
+        if (isNetworkError && networkRetriesRef.current < MAX_NETWORK_RETRIES) {
+          networkRetriesRef.current += 1
+          console.warn(`[ServerPreview] 创建任务网络错误，第 ${networkRetriesRef.current} 次重试`)
+          setTimeout(() => {
+            if (seq === requestSeqRef.current && mountedRef.current) {
+              createAndPoll(request, seq)
+            }
+          }, POLL_INTERVAL * 2)
+          return
+        }
+
+        const errData = (
+          createErr as { response?: { data?: { detail?: string; message?: string } } }
+        )?.response?.data
+        setStatus("failed")
+        setError(errData?.detail || errData?.message || "预览任务创建失败，请重试")
+      }
+    },
+    [clearTimers],
+  )
+
+  /* ── 手动触发预览 ── */
+  const triggerPreview = useCallback(() => {
+    if (!enabled) return
+    const request = buildRequestRef.current()
+    if (!request.template_id || request.asset_ids.length === 0) return
+
+    clearTimers()
+    const seq = ++requestSeqRef.current
+    setVideoUrl(null)
+    setTaskId(null)
+    createAndPoll(request, seq)
+  }, [enabled, clearTimers, createAndPoll])
+
+  /* ── 自动触发 + 配置变更检测 ── */
+  // 每次 render 都检查最新配置 fingerprint，与已渲染的 fingerprint 比较
+  const request = enabled ? buildRequest() : null
+  const currentFingerprint = request
+    ? request.template_id && request.asset_ids.length > 0
+      ? buildFingerprint(request)
+      : ""
+    : ""
+
+  // 首次进入自动触发
+  const didInitRef = useRef(false)
+  useEffect(() => {
+    if (!enabled || !currentFingerprint) {
+      didInitRef.current = false
+      // 禁用时取消进行中的轮询，避免回到前序步骤后仍在后台轮询
+      requestSeqRef.current += 1
+      clearTimers()
+      return
+    }
+    if (!didInitRef.current) {
+      didInitRef.current = true
+      renderedFingerprintRef.current = currentFingerprint
+      triggerPreview()
+    }
+  }, [enabled, currentFingerprint, triggerPreview, clearTimers])
+
+  // 配置变更检测：素材/配音/BGM 等变化 → 自动重渲染；标题样式变化 → 标记 stale
+  const prevFingerprintRef = useRef(currentFingerprint)
+  useEffect(() => {
+    if (!enabled || !currentFingerprint) return
+    const prev = prevFingerprintRef.current
+    prevFingerprintRef.current = currentFingerprint
+
+    if (!prev || prev === currentFingerprint) return
+    if (currentFingerprint === renderedFingerprintRef.current) return
+
+    // 配置已变更
+    // 判断是标题样式变更还是素材/配音/BGM 变更
+    const prevParsed = JSON.parse(prev) as Record<string, unknown>
+    const currParsed = JSON.parse(currentFingerprint) as Record<string, unknown>
+    const nonTitleChanged =
+      prevParsed.t !== currParsed.t ||
+      prevParsed.a !== currParsed.a ||
+      prevParsed.d !== currParsed.d ||
+      prevParsed.r !== currParsed.r ||
+      prevParsed.v !== currParsed.v ||
+      JSON.stringify(prevParsed.b) !== JSON.stringify(currParsed.b)
+
+    if (nonTitleChanged) {
+      // 素材/配音/BGM/模板等变化 → 自动重新渲染
+      renderedFingerprintRef.current = currentFingerprint
+      triggerPreview()
+    } else {
+      // 仅标题文字/样式变化 → 标记 stale，不自动重渲染（避免频繁请求）
+      // 实时预览由 CSS TitleOverlay 提供
+      setStatus((s) => (s === "ready" ? "stale" : s))
+    }
+  }, [enabled, currentFingerprint, triggerPreview])
+
+  return {
+    status,
+    videoUrl,
+    error,
+    progress,
+    triggerPreview,
+    taskId,
+  }
+}
+
+export default useServerPreview

apps/web/src/pages/generate/hooks/useStep2Materials.ts

@@ -3,9 +3,14 @@
  * 组合素材库加载 + 智能匹配两个子 Hook
  */
 import { useCallback, useEffect, useRef } from "react"
+import { message } from "antd"
+import type { TemplateSegment } from "@/api/templates/types"
+import type { EditPlanClip } from "@/api/template-editor"
+import { updateEditPlanClips, createClipsFromAssets, getEditPlanClips } from "@/api/template-editor"
 import { formatDuration } from "../utils/formatDuration"
 import { useMaterialLibrary } from "./step2-materials/useMaterialLibrary"
 import { useSmartMatch } from "./step2-materials/useSmartMatch"
+import { useDraftAutoSave } from "./useDraftAutoSave"
 
 interface UseStep2MaterialsProps {
   materialMode: "manual" | "auto"
@@ -14,6 +19,12 @@ interface UseStep2MaterialsProps {
   onSelectedMaterialsChange: (ids: string[]) => void
   smartSelectedIds: string[]
   onSmartSelectedIdsChange: (ids: string[]) => void
+  /** 当前选中的模板/草稿 ID，用于自动保存 */
+  selectedTemplate?: string
+  /** 当前模板的 segments（用于构建 clips duration） */
+  templateSegments?: TemplateSegment[]
+  /** 服务端 clips 创建成功后的回调，用于通知预览播放器 */
+  onServerClipsChange?: (clips: EditPlanClip[]) => void
 }
 
 export function useStep2Materials({
@@ -23,6 +34,9 @@ export function useStep2Materials({
   onSelectedMaterialsChange,
   smartSelectedIds,
   onSmartSelectedIdsChange,
+  selectedTemplate,
+  templateSegments,
+  onServerClipsChange,
 }: UseStep2MaterialsProps) {
   const { libraries, selectedLibraryId, setSelectedLibraryId, materials, materialsLoading } =
     useMaterialLibrary()
@@ -53,6 +67,90 @@ export function useStep2Materials({
     handleSmartMatch()
   }, [selectedLibraryId, materialMode, materialsLoading, materials.items, handleSmartMatch])
 
+  /* ── Step2 选择素材后自动保存草稿 asset_ids（防抖 500ms，失败静默） ── */
+  const { scheduleSave } = useDraftAutoSave(selectedTemplate)
+  useEffect(() => {
+    if (!selectedTemplate) return
+    const ids = materialMode === "auto" ? smartSelectedIds : selectedMaterials
+    scheduleSave({ asset_ids: ids }, 500)
+  }, [selectedTemplate, materialMode, selectedMaterials, smartSelectedIds, scheduleSave])
+
+  /* ── Step2 选择素材后同步写入 edit_plan_clips（防抖 800ms，失败静默） ──
+   * 调用后端 POST /clips/from-assets，由后端处理：
+   *   - 素材不够时同一素材切多个片段
+   *   - 随机 start_time，不重复
+   *   - required_clips_count 保证片段数与模板 segments 一致
+   * 先 PUT /clips（空数组）清空旧片段，再调用 from-assets 创建新片段
+   */
+  const clipsTimerRef = useRef<ReturnType<typeof setTimeout>>()
+  const clipsAbortRef = useRef<AbortController | null>(null)
+  const templateSegmentsRef = useRef(templateSegments)
+  templateSegmentsRef.current = templateSegments
+  const selectedTemplateRef = useRef(selectedTemplate)
+  selectedTemplateRef.current = selectedTemplate
+  const onServerClipsChangeRef = useRef(onServerClipsChange)
+  onServerClipsChangeRef.current = onServerClipsChange
+
+  useEffect(() => {
+    const tid = selectedTemplateRef.current
+    if (!tid) return
+    const ids = materialMode === "auto" ? smartSelectedIds : selectedMaterials
+    if (!ids.length) {
+      onServerClipsChangeRef.current?.([])
+      return
+    }
+
+    if (clipsTimerRef.current) clearTimeout(clipsTimerRef.current)
+    clipsTimerRef.current = setTimeout(async () => {
+      // 取消上一次未完成的请求
+      if (clipsAbortRef.current) clipsAbortRef.current.abort()
+      const controller = new AbortController()
+      clipsAbortRef.current = controller
+
+      const segs = templateSegmentsRef.current || []
+      const requiredClipsCount = segs.length > 0 ? segs.length : undefined
+
+      try {
+        // 1. 清空旧片段
+        await updateEditPlanClips(tid, [], controller.signal)
+        // 2. 调用后端 from-assets 接口创建片段（异步秒级返回，60s 超时仅为兜底）
+        await createClipsFromAssets(tid, ids, "main", requiredClipsCount, {
+          signal: controller.signal,
+        })
+        // 3. 获取服务端生成的 clips（含 start_time/duration），供预览播放器使用
+        const clipList = await getEditPlanClips(tid, { limit: 500 })
+        const readyClips = clipList.items
+          .filter((c) => c.status === "ready")
+          .sort((a, b) => a.order - b.order)
+        onServerClipsChangeRef.current?.(readyClips)
+      } catch (err) {
+        const name = (err as { name?: string })?.name
+        // 用户切换素材导致的主动取消，静默
+        if (name === "CanceledError" || name === "AbortError") return
+        // from-assets 60s 超时（MediaKit 智能选片未完成）
+        const code = (err as { code?: string })?.code
+        if (code === "ECONNABORTED" || /timeout/i.test((err as Error)?.message || "")) {
+          console.warn("[useStep2Materials] 智能选片超时:", err)
+          message.error("智能选片失败，请重试")
+          return
+        }
+        console.warn("[useStep2Materials] 写入 clips 失败:", err)
+      }
+    }, 800)
+
+    return () => {
+      if (clipsTimerRef.current) clearTimeout(clipsTimerRef.current)
+    }
+  }, [selectedTemplate, materialMode, selectedMaterials, smartSelectedIds, templateSegments])
+
+  // 组件卸载时取消未完成请求
+  useEffect(() => {
+    return () => {
+      if (clipsTimerRef.current) clearTimeout(clipsTimerRef.current)
+      if (clipsAbortRef.current) clipsAbortRef.current.abort()
+    }
+  }, [])
+
   /* ── 手动选择素材 ── */
   const handleToggleMaterial = useCallback(
     (materialId: string) => {

apps/web/src/pages/generate/hooks/useStep4Title/index.ts

@@ -4,17 +4,24 @@ import { getTitles } from "@/api/titles"
 import type { TitleSettings } from "../../types"
 import { useAiTitleGenerator } from "./useAiTitleGenerator"
 import { useTitleStyleUpdaters } from "./useTitleStyleUpdaters"
+import { useDraftAutoSave } from "../useDraftAutoSave"
 
 interface UseStep4TitleProps {
   titleSettings: TitleSettings
   onTitleSettingsChange: (settings: TitleSettings) => void
+  /** 当前选中的模板/草稿 ID，用于自动保存 */
+  selectedTemplate?: string
 }
 
 /**
  * Step 4 标题设置 Hook
  * 封装 AI 标题生成、标题样式设置等逻辑
  */
-export function useStep4Title({ titleSettings, onTitleSettingsChange }: UseStep4TitleProps) {
+export function useStep4Title({
+  titleSettings,
+  onTitleSettingsChange,
+  selectedTemplate,
+}: UseStep4TitleProps) {
   // 标题库数据
   const { data: userTitles = [] } = useQuery({
     queryKey: ["titles"],
@@ -42,6 +49,38 @@ export function useStep4Title({ titleSettings, onTitleSettingsChange }: UseStep4
   const prevAiAutoSelect = useRef(titleSettings.aiAutoSelect)
   const isFirstMount = useRef(true)
 
+  /* ── Step4 标题内容/样式变化后自动保存草稿（防抖 800ms，失败静默） ── */
+  const { scheduleSave: scheduleTitleSave } = useDraftAutoSave(selectedTemplate)
+  useEffect(() => {
+    if (!selectedTemplate) return
+    scheduleTitleSave(
+      {
+        title: {
+          text: titleSettings.title,
+          font: titleSettings.font,
+          font_size: titleSettings.size,
+          color: titleSettings.color,
+          position: titleSettings.position,
+          bold: titleSettings.bold,
+          stroke: titleSettings.stroke,
+          shadow: titleSettings.shadow,
+        },
+      },
+      800,
+    )
+  }, [
+    selectedTemplate,
+    titleSettings.title,
+    titleSettings.font,
+    titleSettings.size,
+    titleSettings.color,
+    titleSettings.position,
+    titleSettings.bold,
+    titleSettings.stroke,
+    titleSettings.shadow,
+    scheduleTitleSave,
+  ])
+
   // 当 AI 自动选择开关打开时，自动生成/选择一个标题填入
   // 首次挂载时如果开关已经是 true 且无标题，也需要触发
   useEffect(() => {

apps/web/src/pages/generate/hooks/useStep6Cover.ts

@@ -2,11 +2,13 @@
  * Step 6 封面设置 Hook
  * 封装封面设置的交互逻辑，对接后端封面模板 CRUD API
  */
-import { useCallback, useEffect, useState } from "react"
+import { useCallback, useEffect, useRef, useState } from "react"
 import { message } from "antd"
 import type { CoverConfig, CoverTemplate } from "../types/cover"
 import { generateCover } from "@/api/generation"
 import { createPreview, getPreviewStatus } from "@/api/generation/preview"
+import { updateEditPlan } from "@/api/template-editor"
+import type { TitleSettings } from "../types"
 import {
   fetchCoverTemplates,
   createCoverTemplate,
@@ -22,6 +24,22 @@ interface UseStep6CoverProps {
   assetIds?: string[]
   /** 当前选中的模板 ID */
   selectedTemplate?: string
+  /** Step4 标题设置，用于预览视频烧录标题 & 封面叠加标题 */
+  titleSettings?: TitleSettings
+  /** 预览任务创建回调——将 task_id 暴露给父组件供 confirmGeneration 复用 */
+  onPreviewTaskCreated?: (taskId: string) => void
+  /** 从预览响应中提取到 source_edit_plan_id 时的回调 */
+  onSourceEditPlanIdExtracted?: (planId: string) => void
+  /** 配音模式 */
+  voiceMode?: "preset" | "custom" | "clone"
+  /** 选中的配音素材 ID（配音素材库 asset ID） */
+  selectedVoice?: string
+  /** 选中的克隆音色 ID */
+  selectedClonedVoice?: string
+  /** BGM 开关 */
+  bgm?: boolean
+  /** BGM 配置（来自模板） */
+  bgmConfig?: { enabled: boolean; music_id?: string }
 }
 
 export function useStep6Cover({
@@ -30,8 +48,18 @@ export function useStep6Cover({
   duration,
   assetIds = [],
   selectedTemplate = "",
+  titleSettings,
+  onPreviewTaskCreated,
+  onSourceEditPlanIdExtracted,
+  voiceMode,
+  selectedVoice,
+  selectedClonedVoice,
+  bgm,
+  bgmConfig,
 }: UseStep6CoverProps) {
   const [generating, setGenerating] = useState(false)
+  // 防竞态：记录当前预览生成的参数指纹，任务完成时校验一致性
+  const previewParamsRef = useRef<string>("")
 
   // ── 封面设置弹窗状态 ──
   const [showCoverSettings, setShowCoverSettings] = useState(false)
@@ -92,6 +120,20 @@ export function useStep6Cover({
       const response = await generateCover(selectedTemplate, {
         asset_ids: assetIds,
         cover_type: "ai_frame",
+        ...(titleSettings?.title
+          ? {
+              title_config: {
+                text: titleSettings.title,
+                font: titleSettings.font,
+                font_size: titleSettings.size,
+                font_color: titleSettings.color,
+                position: titleSettings.position,
+                bold: titleSettings.bold,
+                stroke: titleSettings.stroke,
+                shadow: titleSettings.shadow,
+              },
+            }
+          : {}),
       })
       clearTimeout(timeoutId)
       const thumbnailUrl = response.cover?.image_url || ""
@@ -132,11 +174,48 @@ export function useStep6Cover({
         console.log("[Step6] 检测到预览缺失，尝试自动创建预览渲染任务...")
         message.info("正在准备预览视频，请稍候...")
         try {
+          // 记录当前参数指纹，用于任务完成时校验一致性（防竞态）
+          previewParamsRef.current = JSON.stringify({ selectedTemplate, assetIds, titleSettings })
+          // 解析配音参数：voiceMode=clone 时用 selectedClonedVoice，否则用 selectedVoice
+          const previewVoiceLibraryId =
+            voiceMode === "clone" ? selectedClonedVoice || selectedVoice || "" : selectedVoice || ""
           const previewResp = await createPreview({
             template_id: selectedTemplate,
             asset_ids: assetIds,
             duration: duration || 30,
+            // 配音：始终传递 voice_library_id，确保后端能正确接收
+            voice_library_id: previewVoiceLibraryId,
+            // 兜底：如果 voice_library_id 为空但 selectedVoice 有值，也传 voice_ids
+            ...(selectedVoice && !previewVoiceLibraryId ? { voice_ids: [selectedVoice] } : {}),
+            // BGM 配置：受 bgm 开关控制
+            bgm_config: {
+              enabled: bgm !== false,
+              ...(bgmConfig?.music_id ? { preset_id: bgmConfig.music_id } : {}),
+            },
+            ...(titleSettings?.title
+              ? {
+                  title_config: {
+                    text: titleSettings.title,
+                    font: titleSettings.font,
+                    font_size: titleSettings.size,
+                    font_color: titleSettings.color,
+                    position: titleSettings.position,
+                    bold: titleSettings.bold,
+                    stroke: titleSettings.stroke,
+                    shadow: titleSettings.shadow,
+                  },
+                }
+              : {}),
           })
+          // 将预览任务 ID 暴露给父组件，供 Step7 确认生成时复用（confirmGeneration）
+          const currentFingerprint = JSON.stringify({ selectedTemplate, assetIds, titleSettings })
+          if (previewResp.task_id && previewParamsRef.current === currentFingerprint) {
+            onPreviewTaskCreated?.(previewResp.task_id)
+            // 提取后端自动关联的 source_edit_plan_id，供 fallback 路径使用
+            if (previewResp.source_edit_plan_id) {
+              onSourceEditPlanIdExtracted?.(previewResp.source_edit_plan_id)
+            }
+          }
           // 轮询等待预览渲染完成：递归 setTimeout 避免请求重叠 + 120s 超时兜底
           await new Promise<void>((resolve, reject) => {
             let finished = false
@@ -154,6 +233,21 @@ export function useStep6Cover({
               try {
                 const status = await getPreviewStatus(previewResp.task_id)
                 if (status.status === "completed") {
+                  // 保存预览视频地址到 plan.config.rendered_storage_key，
+                  // 供封面 API 的 E1 兜底路径定位渲染后的视频（含标题烧录）。
+                  // video_url 可能是完整 http(s) URL 或 OSS storage_key，两种格式后端都能处理。
+                  if (status.video_url) {
+                    try {
+                      await updateEditPlan(selectedTemplate, {
+                        config: { rendered_storage_key: status.video_url },
+                      })
+                    } catch (saveErr) {
+                      console.warn(
+                        "[Step6] 保存 rendered_storage_key 失败（不阻塞封面重试）:",
+                        saveErr,
+                      )
+                    }
+                  }
                   done(() => resolve())
                 } else if (status.status === "failed") {
                   done(() => reject(new Error(status.error_message || "预览渲染失败")))
@@ -171,6 +265,20 @@ export function useStep6Cover({
           const retryResp = await generateCover(selectedTemplate, {
             asset_ids: assetIds,
             cover_type: "ai_frame",
+            ...(titleSettings?.title
+              ? {
+                  title_config: {
+                    text: titleSettings.title,
+                    font: titleSettings.font,
+                    font_size: titleSettings.size,
+                    font_color: titleSettings.color,
+                    position: titleSettings.position,
+                    bold: titleSettings.bold,
+                    stroke: titleSettings.stroke,
+                    shadow: titleSettings.shadow,
+                  },
+                }
+              : {}),
           })
           const retryUrl = retryResp.cover?.image_url || ""
           if (retryUrl) {
@@ -212,7 +320,22 @@ export function useStep6Cover({
       clearTimeout(timeoutId)
       setGenerating(false)
     }
-  }, [selectedTemplate, assetIds, coverSettings, onCoverSettingsChange, generating, duration])
+  }, [
+    selectedTemplate,
+    assetIds,
+    coverSettings,
+    onCoverSettingsChange,
+    generating,
+    duration,
+    titleSettings,
+    onPreviewTaskCreated,
+    onSourceEditPlanIdExtracted,
+    voiceMode,
+    selectedVoice,
+    selectedClonedVoice,
+    bgm,
+    bgmConfig,
+  ])
 
   // ── 模板操作方法 ──
   const handleSelectTemplate = useCallback((id: string) => {

apps/web/src/pages/generate/hooks/useStep7Generate.ts

@@ -25,8 +25,6 @@ interface UseStep7GenerateProps {
   presetVoices: PresetVoiceItem[]
   clonedVoices: VoiceClone[]
   coverSettings: CoverConfig
-  generateCount: number
-  onGenerateCountChange: (count: number) => void
   generating: boolean
   generated: boolean
   generateError: string | null
@@ -47,8 +45,6 @@ export function useStep7Generate({
   presetVoices: _presetVoices,
   clonedVoices: _clonedVoices,
   coverSettings,
-  generateCount,
-  onGenerateCountChange,
   generating,
   generated,
   generateError,
@@ -90,14 +86,6 @@ export function useStep7Generate({
     return { label: "即将完成", icon: "✨" }
   }
 
-  const handleDecrement = () => {
-    onGenerateCountChange(Math.max(1, generateCount - 1))
-  }
-
-  const handleIncrement = () => {
-    onGenerateCountChange(Math.min(10, generateCount + 1))
-  }
-
   const handleScrollToPreview = () => {
     const el = document.querySelector(".xx-preview-section")
     el?.scrollIntoView({ behavior: "smooth", block: "start" })
@@ -109,9 +97,6 @@ export function useStep7Generate({
     title,
     voiceName,
     coverSummary,
-    generateCount,
-    handleDecrement,
-    handleIncrement,
     generating,
     generated,
     generateError,

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

@@ -1,9 +1,9 @@
 /**
  * GeneratePage 步骤导航
  * 管理步骤切换与各步骤的前置校验
- * 步骤顺序：模板(1) → 素材(2) → 配音(3) → 标题(4) → 预览(5) → 封面(6) → 确认(7)
+ * 步骤顺序（6步）：模板(1) → 素材(2) → 配音(3) → 标题(4) → 封面(5) → 确认(6)
  *
- * V24: previewReady 改为前端素材加载状态
+ * 原 Step4(标题) + Step5(预览) 已合并为 Step4
  */
 import { message } from "antd"
 import type { TitleSettings } from "../types"
@@ -16,7 +16,7 @@ export interface UseStepNavigationOptions {
   selectedMaterials: string[]
   smartSelectedIds: string[]
   titleSettings: TitleSettings
-  /** 预览是否就绪（前端素材已加载） */
+  /** 预览是否已就绪（素材已加载，可播放） */
   previewReady: boolean
 }
 
@@ -50,16 +50,18 @@ export const useStepNavigation = (options: UseStepNavigationOptions): UseStepNav
       message.warning("请先进行智能匹配并选择素材")
       return
     }
-    // Step3 配音：配音为可选项，不强制校验，用户可跳过
-    if (currentStep === 4 && !titleSettings.title.trim()) {
-      message.warning("请选择或输入标题")
-      return
+    // Step4（合并后的标题+预览）：标题必填 + 预览必须已生成
+    if (currentStep === 4) {
+      if (!titleSettings.title.trim()) {
+        message.warning("请选择或输入标题")
+        return
+      }
+      if (!previewReady) {
+        message.warning("预览视频正在加载，请稍候")
+        return
+      }
     }
-    if (currentStep === 5 && !previewReady) {
-      message.warning("请先选择素材以预览效果")
-      return
-    }
-    if (currentStep < 7) {
+    if (currentStep < 6) {
       setCurrentStep((s) => s + 1)
     }
   }

apps/web/src/pages/generate/utils/calculateResolution.ts

@@ -0,0 +1,46 @@
+/**
+ * 根据视频比例计算输出分辨率
+ *
+ * 规则：
+ * - 长边固定 1920
+ * - 短边按 1920 × (短/长) 计算，对齐到偶数
+ * - 9:16 → 1080×1920（竖屏）
+ * - 16:9 → 1920×1080（横屏）
+ * - 1:1   → 1920×1920
+ * - 无法解析时 fallback 到 1080×1920
+ */
+export interface Resolution {
+  width: number
+  height: number
+}
+
+export function calculateResolution(ratio: string): Resolution {
+  if (!ratio) return { width: 1080, height: 1920 }
+
+  if (ratio.includes(":")) {
+    const [rw, rh] = ratio.split(":").map(Number)
+    if (rw > 0 && rh > 0) {
+      const [longSide, shortSide] = rw < rh ? [rh, rw] : [rw, rh]
+      const baseLong = 1920
+      const baseShort = Math.round((baseLong * shortSide) / longSide)
+      const evenShort = baseShort - (baseShort % 2)
+      if (rw < rh) {
+        // 竖屏：短边是宽，长边是高
+        return { width: evenShort, height: baseLong }
+      } else {
+        // 横屏：长边是宽，短边是高
+        return { width: baseLong, height: evenShort }
+      }
+    }
+    return { width: 1080, height: 1920 }
+  }
+
+  if (ratio.includes("x")) {
+    const [wStr, hStr] = ratio.split("x")
+    const w = parseInt(wStr, 10) || 1080
+    const h = parseInt(hStr, 10) || 1920
+    return { width: w, height: h }
+  }
+
+  return { width: 1080, height: 1920 }
+}

apps/web/src/pages/templates/hooks/useTemplateLibrary.ts

@@ -87,7 +87,7 @@ export const useTemplateLibrary = () => {
     [copyMutation],
   )
 
-  /* 操作：使用模板 → 跳转剪辑编辑器 */
+  /* 操作：使用模板 → 跳转剪辑模板 */
   const handleUse = useCallback(
     (template: TemplateItem) => {
       navigate(`/app/editing-planner?templateId=${template.id}`)

apps/web/src/test/api/bgm.test.ts

@@ -32,7 +32,7 @@ describe("bgm API", () => {
 
   describe("getBgmPresets", () => {
     it("should resolve successfully", async () => {
-      await expect(getBgmPresets("test-params?")).resolves.not.toThrow()
+      await expect(getBgmPresets("test-template", { category: "test" })).resolves.not.toThrow()
     })
 
     it("should reject on API error", async () => {
@@ -42,7 +42,7 @@ describe("bgm API", () => {
       mockDelete.mockRejectedValue(new Error("Network error"))
       mockPatch.mockRejectedValue(new Error("Network error"))
 
-      await expect(getBgmPresets("test-params?")).rejects.toThrow()
+      await expect(getBgmPresets("test-template", { category: "test" })).rejects.toThrow()
     })
   })
 })

apps/web/src/test/api/edit-plans.test.ts

@@ -1,16 +1,9 @@
 import { describe, expect, it, vi, beforeEach } from "vitest"
 import {
-  getEditPlans,
   getEditPlan,
-  createEditPlan,
   updateEditPlan,
-  deleteEditPlan,
-  generateEditPlan,
-  getGenerationStatus,
   aiRecommendClips,
-  getEditPlanGenerations,
   getGenerationTaskResults,
-  cancelGeneration,
   getEditPlanClips,
   getEditPlanClip,
   createEditPlanClip,
@@ -19,7 +12,6 @@ import {
   reorderEditPlanClips,
   batchDeleteEditPlanClips,
   createClipsFromAssets,
-  copyEditPlan,
   getMediaAssets,
   getMediaAsset,
 } from "@/api/template-editor"
@@ -53,22 +45,6 @@ describe("editPlans API", () => {
     mockPatch.mockResolvedValue({ data: { success: true, items: [] } })
   })
 
-  describe("getEditPlans", () => {
-    it("should resolve successfully", async () => {
-      await expect(getEditPlans("test-params?")).resolves.not.toThrow()
-    })
-
-    it("should reject on API error", async () => {
-      mockGet.mockRejectedValue(new Error("Network error"))
-      mockPost.mockRejectedValue(new Error("Network error"))
-      mockPut.mockRejectedValue(new Error("Network error"))
-      mockDelete.mockRejectedValue(new Error("Network error"))
-      mockPatch.mockRejectedValue(new Error("Network error"))
-
-      await expect(getEditPlans("test-params?")).rejects.toThrow()
-    })
-  })
-
   describe("getEditPlan", () => {
     it("should resolve successfully", async () => {
       await expect(getEditPlan("test-planId")).resolves.not.toThrow()
@@ -85,22 +61,6 @@ describe("editPlans API", () => {
     })
   })
 
-  describe("createEditPlan", () => {
-    it("should resolve successfully", async () => {
-      await expect(createEditPlan({ name: "test-item" })).resolves.not.toThrow()
-    })
-
-    it("should reject on API error", async () => {
-      mockGet.mockRejectedValue(new Error("Network error"))
-      mockPost.mockRejectedValue(new Error("Network error"))
-      mockPut.mockRejectedValue(new Error("Network error"))
-      mockDelete.mockRejectedValue(new Error("Network error"))
-      mockPatch.mockRejectedValue(new Error("Network error"))
-
-      await expect(createEditPlan({ name: "test-item" })).rejects.toThrow()
-    })
-  })
-
   describe("updateEditPlan", () => {
     it("should resolve successfully", async () => {
       await expect(updateEditPlan("test-planId")).resolves.not.toThrow()
@@ -117,54 +77,6 @@ describe("editPlans API", () => {
     })
   })
 
-  describe("deleteEditPlan", () => {
-    it("should resolve successfully", async () => {
-      await expect(deleteEditPlan("test-planId")).resolves.not.toThrow()
-    })
-
-    it("should reject on API error", async () => {
-      mockGet.mockRejectedValue(new Error("Network error"))
-      mockPost.mockRejectedValue(new Error("Network error"))
-      mockPut.mockRejectedValue(new Error("Network error"))
-      mockDelete.mockRejectedValue(new Error("Network error"))
-      mockPatch.mockRejectedValue(new Error("Network error"))
-
-      await expect(deleteEditPlan("test-planId")).rejects.toThrow()
-    })
-  })
-
-  describe("generateEditPlan", () => {
-    it("should resolve successfully", async () => {
-      await expect(generateEditPlan("test-planId")).resolves.not.toThrow()
-    })
-
-    it("should reject on API error", async () => {
-      mockGet.mockRejectedValue(new Error("Network error"))
-      mockPost.mockRejectedValue(new Error("Network error"))
-      mockPut.mockRejectedValue(new Error("Network error"))
-      mockDelete.mockRejectedValue(new Error("Network error"))
-      mockPatch.mockRejectedValue(new Error("Network error"))
-
-      await expect(generateEditPlan("test-planId")).rejects.toThrow()
-    })
-  })
-
-  describe("getGenerationStatus", () => {
-    it("should resolve successfully", async () => {
-      await expect(getGenerationStatus("test-planId")).resolves.not.toThrow()
-    })
-
-    it("should reject on API error", async () => {
-      mockGet.mockRejectedValue(new Error("Network error"))
-      mockPost.mockRejectedValue(new Error("Network error"))
-      mockPut.mockRejectedValue(new Error("Network error"))
-      mockDelete.mockRejectedValue(new Error("Network error"))
-      mockPatch.mockRejectedValue(new Error("Network error"))
-
-      await expect(getGenerationStatus("test-planId")).rejects.toThrow()
-    })
-  })
-
   describe("aiRecommendClips", () => {
     it("should resolve successfully", async () => {
       await expect(aiRecommendClips("test-planId")).resolves.not.toThrow()
@@ -181,22 +93,6 @@ describe("editPlans API", () => {
     })
   })
 
-  describe("getEditPlanGenerations", () => {
-    it("should resolve successfully", async () => {
-      await expect(getEditPlanGenerations("test-planId")).resolves.not.toThrow()
-    })
-
-    it("should reject on API error", async () => {
-      mockGet.mockRejectedValue(new Error("Network error"))
-      mockPost.mockRejectedValue(new Error("Network error"))
-      mockPut.mockRejectedValue(new Error("Network error"))
-      mockDelete.mockRejectedValue(new Error("Network error"))
-      mockPatch.mockRejectedValue(new Error("Network error"))
-
-      await expect(getEditPlanGenerations("test-planId")).rejects.toThrow()
-    })
-  })
-
   describe("getGenerationTaskResults", () => {
     it("should resolve successfully", async () => {
       await expect(getGenerationTaskResults("test-taskId")).resolves.not.toThrow()
@@ -213,22 +109,6 @@ describe("editPlans API", () => {
     })
   })
 
-  describe("cancelGeneration", () => {
-    it("should resolve successfully", async () => {
-      await expect(cancelGeneration("test-planId")).resolves.not.toThrow()
-    })
-
-    it("should reject on API error", async () => {
-      mockGet.mockRejectedValue(new Error("Network error"))
-      mockPost.mockRejectedValue(new Error("Network error"))
-      mockPut.mockRejectedValue(new Error("Network error"))
-      mockDelete.mockRejectedValue(new Error("Network error"))
-      mockPatch.mockRejectedValue(new Error("Network error"))
-
-      await expect(cancelGeneration("test-planId")).rejects.toThrow()
-    })
-  })
-
   describe("getEditPlanClips", () => {
     it("should resolve successfully", async () => {
       await expect(getEditPlanClips("test-planId")).resolves.not.toThrow()
@@ -357,22 +237,6 @@ describe("editPlans API", () => {
     })
   })
 
-  describe("copyEditPlan", () => {
-    it("should resolve successfully", async () => {
-      await expect(copyEditPlan("test-planId")).resolves.not.toThrow()
-    })
-
-    it("should reject on API error", async () => {
-      mockGet.mockRejectedValue(new Error("Network error"))
-      mockPost.mockRejectedValue(new Error("Network error"))
-      mockPut.mockRejectedValue(new Error("Network error"))
-      mockDelete.mockRejectedValue(new Error("Network error"))
-      mockPatch.mockRejectedValue(new Error("Network error"))
-
-      await expect(copyEditPlan("test-planId")).rejects.toThrow()
-    })
-  })
-
   describe("getMediaAssets", () => {
     it("should resolve successfully", async () => {
       await expect(getMediaAssets("test-libraryId?")).resolves.not.toThrow()

apps/web/src/test/api/templates.test.ts

@@ -5,7 +5,6 @@ import {
   getTemplate,
   toggleFavoriteTemplate,
   copyTemplate,
-  generateFromTemplate,
 } from "@/api/templates"
 
 const mockGet = vi.fn()
@@ -116,20 +115,4 @@ describe("templates API", () => {
       await expect(copyTemplate("test-templateId")).rejects.toThrow()
     })
   })
-
-  describe("generateFromTemplate", () => {
-    it("should resolve successfully", async () => {
-      await expect(generateFromTemplate("test-templateId")).resolves.not.toThrow()
-    })
-
-    it("should reject on API error", async () => {
-      mockGet.mockRejectedValue(new Error("Network error"))
-      mockPost.mockRejectedValue(new Error("Network error"))
-      mockPut.mockRejectedValue(new Error("Network error"))
-      mockDelete.mockRejectedValue(new Error("Network error"))
-      mockPatch.mockRejectedValue(new Error("Network error"))
-
-      await expect(generateFromTemplate("test-templateId")).rejects.toThrow()
-    })
-  })
 })

apps/web/src/test/pages/EditingPlanner.test.tsx

@@ -148,14 +148,9 @@ vi.mock("@/api/editing-planner", () => ({
 
 vi.mock("@/api/template-editor", () => ({
   getMediaAssets: vi.fn().mockResolvedValue({ items: [] }),
-  getEditPlanGenerations: vi.fn().mockResolvedValue({ items: [] }),
   getEditPlan: vi.fn().mockResolvedValue({}),
-  createEditPlan: vi.fn().mockResolvedValue({ id: "test-plan" }),
   updateEditPlan: vi.fn().mockResolvedValue({}),
-  generateEditPlan: vi.fn().mockResolvedValue({ task_id: "test-task" }),
-  getGenerationStatus: vi.fn().mockResolvedValue({ status: "completed" }),
   getGenerationTaskResults: vi.fn().mockResolvedValue({ items: [] }),
-  cancelGeneration: vi.fn().mockResolvedValue({}),
   getEditPlanClips: vi.fn().mockResolvedValue({ items: [] }),
   createEditPlanClip: vi.fn().mockResolvedValue({}),
   batchDeleteEditPlanClips: vi.fn().mockResolvedValue({}),
@@ -225,12 +220,6 @@ vi.mock("@/pages/editing-planner/components/StickerPanel", () => ({
 vi.mock("@/pages/editing-planner/components/SaveModal", () => ({
   default: () => React.createElement("div", { "data-testid": "SaveModal" }),
 }))
-vi.mock("@/pages/editing-planner/components/GenerationHistoryModal", () => ({
-  default: () => React.createElement("div", { "data-testid": "GenerationHistoryModal" }),
-}))
-vi.mock("@/pages/editing-planner/components/GenerationProgressModal", () => ({
-  default: () => React.createElement("div", { "data-testid": "GenerationProgressModal" }),
-}))
 
 // === useUndoRedo hook mock ===
 vi.mock("@/pages/editing-planner/hooks/useUndoRedo", () => ({

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

@@ -214,12 +214,6 @@ vi.mock("@/api/titles", () => ({
 }))
 
 vi.mock("@/api/template-editor", () => ({
-  generateEditPlan: vi.fn().mockResolvedValue({
-    plan_id: "test-plan",
-    generation_task_id: "test-task",
-    plan_status: "processing",
-    clip_count: 5,
-  }),
   updateEditPlan: vi.fn().mockResolvedValue({ plan_id: "test-plan", template_id: "test-template" }),
   getEditPlan: vi.fn().mockResolvedValue({
     plan_id: "test-plan",
@@ -228,9 +222,6 @@ vi.mock("@/api/template-editor", () => ({
     config: {},
     status: "draft",
   }),
-  getGenerationStatus: vi
-    .fn()
-    .mockResolvedValue({ plan_status: "completed", generation_task_id: "test-task", clips: [] }),
   getGenerationTaskResults: vi.fn().mockResolvedValue({ items: [] }),
 }))
 

apps/web/src/test/pages/TemplateLibrary.test.tsx

@@ -109,7 +109,6 @@ vi.mock("@/api/templates", () => ({
   getTemplate: vi.fn().mockResolvedValue({ items: [], total: 0, success: true }),
   toggleFavoriteTemplate: vi.fn().mockResolvedValue({ items: [], total: 0, success: true }),
   copyTemplate: vi.fn().mockResolvedValue({ items: [], total: 0, success: true }),
-  generateFromTemplate: vi.fn().mockResolvedValue({ items: [], total: 0, success: true }),
 }))
 
 vi.mock("@/pages/templates/TemplateLibrary.css", () => ({}))

apps/web/src/test/pages/editing-planner/components/GenerationHistoryModal.test.tsx

@@ -1,39 +0,0 @@
-import { describe, expect, it, vi } from "vitest"
-import { render } from "@testing-library/react"
-import GenerationHistoryModal from "@/pages/editing-planner/components/GenerationHistoryModal"
-
-const defaultProps = {
-  open: true,
-  loading: false,
-  history: [],
-  onClose: vi.fn(),
-  onCancel: vi.fn(),
-  cancelLoading: false,
-}
-
-describe("GenerationHistoryModal", () => {
-  it("should render without crashing", () => {
-    const { container } = render(<GenerationHistoryModal {...defaultProps} />)
-    expect(container).toBeTruthy()
-  })
-
-  it("should not render when open is false", () => {
-    const { container } = render(<GenerationHistoryModal {...defaultProps} open={false} />)
-    expect(container.firstChild).toBeNull()
-  })
-
-  it("should render with history items", () => {
-    const history = [
-      {
-        id: "1",
-        status: "completed",
-        created_at: "2024-01-01T00:00:00Z",
-        duration: 60,
-      },
-    ]
-    const { container } = render(
-      <GenerationHistoryModal {...defaultProps} history={history as any} />,
-    )
-    expect(container).toBeTruthy()
-  })
-})

apps/web/src/test/pages/editing-planner/components/GenerationProgressModal.test.tsx

@@ -1,102 +0,0 @@
-import { describe, expect, it, vi } from "vitest"
-import { render } from "@testing-library/react"
-import GenerationProgressModal from "@/pages/editing-planner/components/GenerationProgressModal"
-
-const baseProps = {
-  open: true,
-  voiceoverDuration: null,
-  estimatedDuration: 60,
-  onDurationChange: vi.fn(),
-  onGenerate: vi.fn(),
-  task: null,
-  submitting: false,
-  onCancel: vi.fn(),
-  onRetry: vi.fn(),
-  onClose: vi.fn(),
-}
-
-describe("GenerationProgressModal", () => {
-  it("should render setup phase", () => {
-    const { container } = render(<GenerationProgressModal {...baseProps} phase="setup" />)
-    expect(container).toBeTruthy()
-  })
-
-  it("should render progress phase without task", () => {
-    const { container } = render(<GenerationProgressModal {...baseProps} phase="progress" />)
-    expect(container).toBeTruthy()
-  })
-
-  it("should render progress phase with task data", () => {
-    const { container } = render(
-      <GenerationProgressModal
-        {...baseProps}
-        phase="progress"
-        task={
-          {
-            id: "task-123",
-            status: "generating_video",
-            progress: 50,
-            current_step: "generating_video",
-            user_message: "正在生成视频",
-          } as any
-        }
-      />,
-    )
-    expect(container).toBeTruthy()
-  })
-
-  it("should render completed phase", () => {
-    const { container } = render(
-      <GenerationProgressModal
-        {...baseProps}
-        phase="completed"
-        task={{ id: "task-1", status: "completed", progress: 100 } as any}
-      />,
-    )
-    expect(container).toBeTruthy()
-  })
-
-  it("should render failed phase with retry", () => {
-    const { container } = render(
-      <GenerationProgressModal
-        {...baseProps}
-        phase="failed"
-        task={
-          {
-            id: "task-1",
-            status: "failed",
-            progress: 30,
-            error_message: "生成失败",
-            retryable: true,
-          } as any
-        }
-      />,
-    )
-    expect(container).toBeTruthy()
-  })
-
-  it("should render failed phase without retry", () => {
-    const { container } = render(
-      <GenerationProgressModal
-        {...baseProps}
-        phase="failed"
-        task={
-          {
-            id: "task-1",
-            status: "failed",
-            progress: 30,
-            retryable: false,
-          } as any
-        }
-      />,
-    )
-    expect(container).toBeTruthy()
-  })
-
-  it("should not render when closed", () => {
-    const { container } = render(
-      <GenerationProgressModal {...baseProps} phase="setup" open={false} />,
-    )
-    expect(container).toBeTruthy()
-  })
-})

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

@@ -31,8 +31,6 @@ import "@/pages/editing-planner/components/ClipPropertiesPanel"
 import "@/pages/editing-planner/components/EditorClipList"
 import "@/pages/editing-planner/components/EditingDrawers"
 import "@/pages/editing-planner/components/FilterPanel"
-import "@/pages/editing-planner/components/GenerationHistoryModal"
-import "@/pages/editing-planner/components/GenerationProgressModal"
 import "@/pages/editing-planner/components/GreenScreenPanel"
 import "@/pages/editing-planner/components/IntroOutroPanel"
 import "@/pages/editing-planner/components/MediaPanel"

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

@@ -17,7 +17,6 @@ import "@/api/generation/types"
 // 直接引入所有 Step 组件，建立完整依赖链
 import "@/pages/generate/GeneratePage"
 import "@/pages/generate/components/Step2MaterialSelect"
-import "@/pages/generate/components/Step5GeneratePreview"
 import "@/pages/generate/components/Step4TitleSettings"
 import "@/pages/generate/components/Step5VoiceSelect"
 import "@/pages/generate/components/PreviewVideoPanel"

apps/web/src/test/pages/generate/step5-preview.smoke.test.tsx

@@ -1,10 +1,9 @@
 /**
- * Step5GeneratePreview smoke test
+ * Step4+5 merged preview smoke test
  * 确保 vitest related 模式能匹配到第5步预览生成相关文件的改动
  */
 import { describe, it, expect } from "vitest"
 
-import "@/pages/generate/components/Step5GeneratePreview"
 import "@/pages/generate/hooks/usePreviewAssets"
 import "@/pages/generate/hooks/useSegmentScheduler"
 import "@/pages/generate/components/FrontendPreviewPlayer"
@@ -12,8 +11,8 @@ import "@/pages/generate/hooks/useStepNavigation"
 import "@/pages/generate/components/GenerateStepContent"
 import "@/pages/generate/GeneratePage"
 
-describe("Step5GeneratePreview module smoke test", () => {
-  it("should load all step5 preview modules", () => {
+describe("Merged Step4+5 preview module smoke test", () => {
+  it("should load all merged step4+5 preview modules", () => {
     expect(true).toBe(true)
   })
 })

apps/worker/video_processing/__init__.py

@@ -7,7 +7,7 @@ VideoProcessor 等）按需从子模块导入，避免 __init__ 阶段引入
 packages / DB 等重依赖。
 """
 
-# 共享工具模块（零外部依赖，供 editing_modes / generation / edit_plan_generation 等复用）
+# 共享工具模块（零外部依赖，供 editing_modes / generation 等复用）
 from . import dedup_helpers, ffmpeg_utils, oss_helpers, url_security
 
 __all__ = [

apps/worker/video_processing/dedup_helpers.py

@@ -1,6 +1,6 @@
 """查重辅助函数 — 从 generation.py 提取的 GeneratedVideo 记录 + 查重逻辑.
 
-供 render_edit_plan 和 generate_video 共同复用，
+供 generate_video 共同复用，
 创建 GeneratedVideo 记录后计算指纹并执行项目级 + 批次内查重。
 """
 

apps/worker/video_processing/oss_helpers.py

@@ -1,4 +1,4 @@
-"""OSS 工具函数 — 从 generation.py / edit_plan_generation.py 提取的共享 OSS 操作.
+"""OSS 工具函数 — 从 generation.py 提取的共享 OSS 操作.
 
 提供 OSS 配置读取、Bucket 创建、素材上传/下载、asset_id → 本地路径解析
 等能力，供 render_edit_plan 和 generate_video 共同复用。

apps/worker/video_processing/render_adapter.py

@@ -84,6 +84,7 @@ class RenderAdapterResult:
     cover_candidates: list[dict] | None = (
         None  # 封面候选帧 [{"image_url": "...", "frame_time": 5.0, "storage_key": "..."}]
     )
+    temp_dir: str | None = None  # 渲染临时目录，成功时由调用方清理，失败时由 finally 清理
 
     def __post_init__(self):
         if self.rendered_clip_ids is None:
@@ -128,6 +129,7 @@ class RenderAdapter:
         job_id: str = "",
         work_dir: Path | None = None,
         progress_cb: ProgressCallback | None = None,
+        voiceover_audio_path: str | None = None,
     ) -> RenderAdapterResult:
         """渲染一个 EditPlan。
 
@@ -142,6 +144,7 @@ class RenderAdapter:
             job_id: 关联的 Job ID（用于结果存储路径）
             work_dir: 工作目录，不传则使用临时目录
             progress_cb: 进度回调函数
+            voiceover_audio_path: 配音音频本地路径（一键生成场景使用）
 
         Returns:
             RenderAdapterResult
@@ -198,7 +201,7 @@ class RenderAdapter:
             self._report_progress(progress_cb, 35.0, "准备 BGM 音频")
 
             # 3~6. 统一渲染核心流程（BGM + ASR + 渲染 + 缩略图 + 上传）
-            return self._do_render(
+            result = self._do_render(
                 plan=plan,
                 clips=ready_clips,
                 asset_path_map=asset_path_map,
@@ -208,7 +211,13 @@ class RenderAdapter:
                 progress_cb=progress_cb,
                 rendered_clip_ids=rendered_clip_ids,
                 failed_clip_ids=failed_clip_ids,
+                voiceover_audio_path=voiceover_audio_path,
             )
+            # 成功时将临时目录所有权转移给调用方，阻止 finally 清理
+            if result.success and temp_dir:
+                result.temp_dir = temp_dir
+                temp_dir = None  # 阻止 finally 块清理
+            return result
 
         except subprocess.CalledProcessError as exc:
             stderr_text = (exc.stderr or "").strip()
@@ -566,12 +575,13 @@ class RenderAdapter:
 
         self._report_progress(progress_cb, 90.0, "生成封面缩略图")
 
-        # 6. 生成封面缩略图
+        # 6. 生成封面缩略图（结果通过 RenderAdapterResult.thumbnail_url 返回给调用方）
         thumbnail_url = ""
         try:
             from video_processing.thumbnail_generator import generate_and_upload_thumbnail
 
-            thumb_storage_key = f"rendered/{plan_id}/thumbnail.jpg"
+            # job_id 是 _do_render 方法的参数（参见方法签名）
+            thumb_storage_key = f"rendered/{plan_id}/thumbnails/{job_id}.jpg"
             thumbnail_url = generate_and_upload_thumbnail(str(result.output_path), thumb_storage_key)
         except Exception as thumb_err:
             logger.warning(
@@ -585,14 +595,11 @@ class RenderAdapter:
         try:
             from video_processing.thumbnail_generator import extract_and_upload_cover_frames
 
-            # 从 plan config 提取标题文字，叠加到封面候选帧上
-            _title_cfg = (plan_config or {}).get("title", {}) or {}
-            if not isinstance(_title_cfg, dict):
-                _title_cfg = {}
-            _title_text = (_title_cfg.get("text", "") or "").strip() if _title_cfg.get("enabled", True) else ""
-
+            # 已渲染视频在统一渲染阶段已通过 ASS 字幕把标题烧录进画面，
+            # 抽帧天然带标题，因此这里传空字符串，避免 Pillow 二次叠加导致重影。
+            # Pillow 叠加仅用于 API 从源素材抽帧（源素材本身无标题）的兜底场景。
             cover_candidates = extract_and_upload_cover_frames(
-                str(result.output_path), plan_id, num_frames=3, title_text=_title_text
+                str(result.output_path), plan_id, task_id=job_id, num_frames=3, title_text=""
             )
             if cover_candidates:
                 logger.info(

apps/worker/video_processing/render_audio.py

@@ -67,6 +67,16 @@ def clip_has_audio(ctx: RenderContext, clip: ResolvedClip) -> bool:
     return ctx._audio_cache[key]
 
 
+def _clip_volume(clip: ResolvedClip) -> float:
+    """读取 clip 的音量配置（0.0~1.0，>1 放大）。缺省 1.0 原声。"""
+    cfg = getattr(clip, "config", None) or {}
+    try:
+        vol = float(cfg.get("volume", 1.0))
+    except (TypeError, ValueError):
+        return 1.0
+    return max(0.0, vol)
+
+
 # ── 音频混音 ──────────────────────────────────────────────────────────────────
 
 
@@ -82,12 +92,13 @@ def mix_audio(
     """音频后处理混音.
 
     处理逻辑：
-    1. 丢弃主图层（main/broll/overlay/corner_voice）的原始音频，避免录入源视频杂音
-    2. 仅使用独立音频轨（audio role，TTS/配音）作为主音频
-    3. 如果提供了 bgm_path，则额外混入 BGM（支持淡入淡出、循环、人声闪避）
-    4. 如果配置了 audio_tracks，则混入多轨道音频（配音、音效等）
-    5. 输出时长截断到 video_duration
-    6. 如果配置了降噪，最后应用降噪
+    1. 保留主图层（main/broll/overlay/corner_voice）视频素材的原声，按顺序 concat 拼接
+    2. 每个 clip 按 config.volume 应用音量（volume=0 静音，=1 原声）
+    3. 独立音频轨（audio role，TTS/配音）通过 amix 混入
+    4. 如果提供了 bgm_path，则额外混入 BGM（支持淡入淡出、循环、人声闪避）
+    5. 如果配置了 audio_tracks，则混入多轨道音频（配音、音效等）
+    6. 输出时长截断到 video_duration
+    7. 如果配置了降噪，最后应用降噪
 
     Args:
         ctx: 渲染上下文
@@ -123,10 +134,12 @@ def mix_audio(
     if "audio" in layer_map:
         audio_clips = layer_map["audio"].clips
 
-    # ── 丢弃源视频的原始音频（避免录入杂音），成片仅保留 TTS 配音 + BGM ──
-    main_clips = []
+    # ── 保留源视频原声：过滤掉无音频流的 main clip（图片/无声素材） ──
+    # 注意：volume=0 的 clip 不能移除——移除会导致后续 clip 音频时间轴前移、音画不同步。
+    # volume=0 通过滤镜链生成静音流，保持时间轴对齐。
+    main_clips = [c for c in main_clips if clip_has_audio(ctx, c)]
 
-    # ── 防御：过滤掉无音频流的 clip ──
+    # ── 防御：过滤掉无音频流的独立音频轨 ──
     audio_clips = [c for c in audio_clips if clip_has_audio(ctx, c)]
 
     if not main_clips and not audio_clips:
@@ -144,7 +157,7 @@ def mix_audio(
     # 构建音频处理命令
     output_path = ctx.work_dir / f"audio_{ctx.plan_id}.aac"
 
-    # 源视频原始音频已被丢弃（main_clips = []），最终音频完全由独立音频轨 + BGM + 多轨配置组成。
+    # 主音频为视频素材原声 concat；独立音频轨（TTS/配音）通过 amix 混入。
     # 当无 main_clips 时，将独立音频轨作为主音频走 concat 拼接；当二者均有则走 amix 混音。
     if main_clips:
         effective_main = main_clips
@@ -266,28 +279,67 @@ def concat_main_audio(
         has_speed = abs(speed - 1.0) >= 1e-6
 
         if not has_speed and not has_reverse:
-            # 无调速无倒放：简单命令行，-ss 裁剪更高效
-            command = [
-                FFMPEG_BIN,
-                "-y",
-                "-i",
-                str(clip.local_path),
-                "-vn",
-                "-acodec",
-                "aac",
-                "-b:a",
-                "128k",
-                "-ar",
-                "48000",
-                "-ac",
-                "2",
-            ]
-            if trim_start > 0:
-                command.extend(["-ss", f"{trim_start:.3f}"])
-            if final_duration > 0:
-                command.extend(["-t", f"{final_duration:.3f}"])
-            command.append(str(output_path))
-            run_ffmpeg(command)
+            # 无调速无倒放：根据是否需要裁剪/音量选择最高效的路径。
+            vol = _clip_volume(clip)
+            need_trim = trim_start > 0 or (effective_duration > 0 and final_duration < adjusted_duration)
+            need_volume = abs(vol - 1.0) >= 1e-6
+
+            if need_trim:
+                # 需要裁剪：用 atrim 滤镜在滤镜链中精确裁剪（采样点级精度，不浪费解码）。
+                # 滤镜顺序：atrim → asetpts → volume（先裁剪再调音量，避免处理被丢弃的数据）。
+                af_parts: list[str] = []
+                if trim_start > 0 and effective_duration > 0:
+                    af_parts.append(f"atrim=start={trim_start:.3f}:duration={final_duration:.3f}")
+                elif trim_start > 0:
+                    af_parts.append(f"atrim=start={trim_start:.3f}")
+                elif final_duration > 0:
+                    af_parts.append(f"atrim=duration={final_duration:.3f}")
+                af_parts.append("asetpts=PTS-STARTPTS")
+                if need_volume:
+                    af_parts.append(f"volume={vol:.4f}")
+                command = [
+                    FFMPEG_BIN,
+                    "-y",
+                    "-i",
+                    str(clip.local_path),
+                    "-vn",
+                    "-af",
+                    ",".join(af_parts),
+                    "-acodec",
+                    "aac",
+                    "-b:a",
+                    "128k",
+                    "-ar",
+                    "48000",
+                    "-ac",
+                    "2",
+                ]
+                # atrim 已精确控制时长，无需额外 -t
+                command.append(str(output_path))
+                run_ffmpeg(command)
+            else:
+                # 无需裁剪：直接提取，最高效。音量用单个 -af（如有）。
+                command = [
+                    FFMPEG_BIN,
+                    "-y",
+                    "-i",
+                    str(clip.local_path),
+                    "-vn",
+                    "-acodec",
+                    "aac",
+                    "-b:a",
+                    "128k",
+                    "-ar",
+                    "48000",
+                    "-ac",
+                    "2",
+                ]
+                if need_volume:
+                    command.extend(["-af", f"volume={vol:.4f}"])
+                if final_duration > 0:
+                    command.extend(["-t", f"{final_duration:.3f}"])
+                command.append(str(output_path))
+                run_ffmpeg(command)
         else:
             # 有调速或倒放：用 filter_complex
             speed_engine = SpeedEngine()
@@ -312,6 +364,11 @@ def concat_main_audio(
                 if reverse_filter:
                     audio_filters.append(reverse_filter)
 
+            # 音量
+            vol = _clip_volume(clip)
+            if abs(vol - 1.0) >= 1e-6:
+                audio_filters.append(f"volume={vol:.4f}")
+
             # aformat 归一化：统一输出格式为 48000Hz + stereo + fltp
             audio_filters.append("aformat=sample_rates=48000:channel_layouts=stereo:sample_fmts=fltp")
 
@@ -378,6 +435,11 @@ def concat_main_audio(
             if reverse_filter:
                 audio_filters.append(reverse_filter)
 
+        # 音量（0=静音，1=原声）
+        vol = _clip_volume(clip)
+        if abs(vol - 1.0) >= 1e-6:
+            audio_filters.append(f"volume={vol:.4f}")
+
         # aformat 归一化：统一采样率48000Hz + 双声道stereo + fltp采样格式
         # concat filter 要求所有输入音频参数完全一致，否则 exit=234 失败
         audio_filters.append("aformat=sample_rates=48000:channel_layouts=stereo:sample_fmts=fltp")
@@ -450,6 +512,9 @@ def mix_with_independent_audio(
                 clip_filters.append("asetpts=PTS-STARTPTS")
             else:
                 clip_filters.append("asetpts=PTS-STARTPTS")
+            vol = _clip_volume(clip)
+            if abs(vol - 1.0) >= 1e-6:
+                clip_filters.append(f"volume={vol:.4f}")
             # aformat 归一化：concat/amix 前统一音频参数，否则不同采样率/声道会失败
             clip_filters.append(AFORMAT)
             filter_parts.append(f"[{input_idx}:a]{','.join(clip_filters)}[ma{input_idx}]")

apps/worker/video_processing/thumbnail_generator.py

@@ -1,7 +1,8 @@
-"""视频封面抽帧工具 — 从已渲染视频中抽取帧作为封面。
+"""视频封面抽帧工具 — 从视频中抽取帧作为封面，支持标题文字叠加。
 
-统一封面管道：视频渲染时标题已通过 ASS 字幕烧进视频，
-渲染完成后直接从此视频抽帧，封面天然带标题，无需额外叠加逻辑。
+统一封面管道：
+- 从已渲染视频抽帧：标题已通过 ASS 字幕烧进视频，帧天然带标题，无需再叠加。
+- 从源素材抽帧（API E2 兜底）：源素材无标题，通过 Pillow 在帧上绘制标题文字。
 """
 
 from __future__ import annotations
@@ -12,6 +13,40 @@ from pathlib import Path
 
 logger = logging.getLogger(__name__)
 
+# ── 标题叠加（Pillow）──────────────────────────────────────────────────────
+# 实现统一放在 packages/shared/title_overlay.py，API 和 Worker 共用。
+
+
+def apply_title_overlay(
+    image_path: str,
+    title_text: str,
+    *,
+    color: str = "#ffffff",
+    position: str = "bottom",
+    font_size: int | None = None,
+    margin_ratio: float = 0.06,
+    stroke_width_ratio: float = 0.04,
+) -> str:
+    """在图片上绘制标题文字（指定颜色 + 黑色描边/阴影）。
+
+    委托给 packages.shared.title_overlay.apply_title_to_image，
+    保持 Worker 内调用方式不变。title_text 为空时直接返回原路径。
+    """
+    from packages.shared.title_overlay import apply_title_to_image
+
+    if not title_text or not title_text.strip():
+        return image_path
+    result = apply_title_to_image(
+        image_path,
+        title_text,
+        color=color,
+        position=position,
+        font_size=font_size,
+        margin_ratio=margin_ratio,
+        stroke_width_ratio=stroke_width_ratio,
+    )
+    return result or image_path
+
 
 def extract_first_frame(
     video_path: str,
@@ -169,24 +204,106 @@ def generate_and_upload_thumbnail(
         Path(tmp.name).unlink(missing_ok=True)
 
 
+def _extract_frames_via_mediakit(
+    video_path: str,
+    plan_id: str,
+    num_frames: int,
+) -> list[dict] | None:
+    """使用 MediaKit 智能抽帧 API 提取封面帧。
+
+    Args:
+        video_path: 本地视频文件路径
+        plan_id: 编辑计划 ID
+        num_frames: 需要的帧数
+
+    Returns:
+        帧列表 [{"image_url": str, "timestamp": float}, ...]，失败返回 None
+    """
+    import uuid
+
+    from video_processing.oss_helpers import upload_to_oss
+
+    from packages.shared.mediakit_client import get_mediakit_client
+
+    client = get_mediakit_client()
+    if not client.is_available:
+        logger.info("[thumbnail] MediaKit 未配置，跳过智能抽帧")
+        return None
+
+    # 1. 上传视频到 OSS 获取 URL
+    try:
+        video_storage_key = f"temp/{plan_id}/{uuid.uuid4().hex[:8]}_{Path(video_path).name}"
+        video_url = upload_to_oss(video_path, video_storage_key)
+        if not video_url:
+            logger.warning("[thumbnail] 视频上传 OSS 失败，无法使用 MediaKit")
+            return None
+        logger.info("[thumbnail] 视频已上传 OSS: %s", video_url[:80])
+    except Exception as e:
+        logger.warning("[thumbnail] 视频上传 OSS 异常: %s，降级到 ffmpeg", e)
+        return None
+
+    # 2. 调用 MediaKit 智能抽帧
+    try:
+        frames = client.extract_frames(
+            video_url=video_url,
+            strategy="SceneChange",
+            max_frames=num_frames * 2,  # 多取一些帧供选择
+        )
+        if not frames:
+            logger.warning("[thumbnail] MediaKit 抽帧返回空，降级到 ffmpeg")
+            return None
+
+        # 选取最均匀的 num_frames 个帧
+        if len(frames) > num_frames:
+            step = len(frames) // num_frames
+            frames = [frames[i * step] for i in range(num_frames)]
+
+        logger.info("[thumbnail] MediaKit 抽帧成功: %d 帧", len(frames))
+        return frames
+
+    except Exception as e:
+        logger.warning("[thumbnail] MediaKit 抽帧异常: %s，降级到 ffmpeg", e)
+        return None
+    finally:
+        # 清理临时视频文件
+        try:
+            from video_processing.oss_helpers import delete_from_oss
+
+            delete_from_oss(video_storage_key)
+        except Exception:
+            pass
+
+
 def extract_and_upload_cover_frames(
     video_path: str,
     plan_id: str,
     *,
+    task_id: str = "",
     num_frames: int = 3,
     title_text: str = "",
+    title_color: str = "#ffffff",
+    title_position: str = "bottom",
+    title_font_size: int | None = None,
 ) -> list[dict]:
     """从视频中抽取多帧作为封面候选，上传到 OSS。
 
+    优先使用 MediaKit 智能抽帧，失败时降级到 ffmpeg 直接抽帧。
+
     Args:
         video_path: 视频文件路径
         plan_id: 编辑计划 ID（用于生成 storage key）
+        task_id: 任务 ID（用于生成独立的 storage key，避免标题变更时封面冲突）
         num_frames: 抽取帧数（默认 3）
-        title_text: 标题文字（当前版本未叠加，预留参数）
+        title_text: 标题文字；非空时用 Pillow 叠加到每帧。
+            从已渲染视频抽帧时通常传空（标题已烧录）；从源素材抽帧时传标题。
+        title_color: 标题字体颜色（#RRGGBB）
+        title_position: 标题位置 top/center/bottom
+        title_font_size: 标题字号，None 时自动计算
 
     Returns:
         封面候选列表，每项包含 {"url": str, "position": float}
     """
+    import httpx
     from video_processing.ffmpeg_utils import probe_duration
     from video_processing.oss_helpers import upload_to_oss
 
@@ -196,6 +313,51 @@ def extract_and_upload_cover_frames(
         duration = 0.0
 
     candidates: list[dict] = []
+
+    # 优先尝试 MediaKit 智能抽帧
+    mediakit_frames = _extract_frames_via_mediakit(video_path, plan_id, num_frames)
+    if mediakit_frames:
+        for i, frame in enumerate(mediakit_frames):
+            frame_url = frame.get("image_url")
+            if not frame_url:
+                continue
+            tmp = tempfile.NamedTemporaryFile(suffix=".jpg", delete=False)
+            tmp.close()
+            try:
+                # 下载 MediaKit 返回的帧图
+                resp = httpx.get(frame_url, timeout=30, follow_redirects=True)
+                resp.raise_for_status()
+                with open(tmp.name, "wb") as f:
+                    f.write(resp.content)
+
+                # 叠加标题文字（如需要）
+                if title_text and title_text.strip():
+                    apply_title_overlay(
+                        tmp.name,
+                        title_text,
+                        color=title_color,
+                        position=title_position,
+                        font_size=title_font_size,
+                    )
+
+                storage_key = f"covers/{plan_id}/{task_id}/mediakit_frame_{i}.jpg"
+                url = upload_to_oss(tmp.name, storage_key)
+                if url:
+                    seek_time = frame.get("timestamp", 0.0)
+                    candidates.append({"url": url, "position": round(seek_time, 2)})
+            except Exception as e:
+                logger.warning("[thumbnail] MediaKit 帧 %d 处理失败: %s", i, e)
+            finally:
+                Path(tmp.name).unlink(missing_ok=True)
+
+        if len(candidates) >= num_frames:
+            logger.info("[thumbnail] MediaKit 智能抽帧完成: %d 帧", len(candidates))
+            return candidates[:num_frames]
+
+        logger.warning("[thumbnail] MediaKit 抽帧不足 %d 帧，降级到 ffmpeg", num_frames)
+
+    # Fallback: ffmpeg 直接抽帧
+    logger.info("[thumbnail] 使用 ffmpeg 抽帧")
     # 均匀分布抽帧点：从 10% 到 90%
     for i in range(num_frames):
         ratio = 0.1 + 0.8 * i / max(num_frames - 1, 1)
@@ -208,7 +370,16 @@ def extract_and_upload_cover_frames(
                 seek_ratio=ratio,
                 min_seek_seconds=0.5,
             )
-            storage_key = f"covers/{plan_id}/frame_{i}.jpg"
+            # 从源素材抽帧时叠加标题文字；已渲染视频标题已烧录时传空字符串跳过
+            if title_text and title_text.strip():
+                apply_title_overlay(
+                    frame_path,
+                    title_text,
+                    color=title_color,
+                    position=title_position,
+                    font_size=title_font_size,
+                )
+            storage_key = f"covers/{plan_id}/{task_id}/frame_{i}.jpg"
             url = upload_to_oss(frame_path, storage_key)
             if url:
                 seek_time = max(0.5, duration * ratio) if duration > 0 else 0.0

apps/worker/video_processing/unified_render_service.py

@@ -36,6 +36,7 @@ from video_processing.ffmpeg_utils import (
     DEFAULT_TRANSITION_DURATION,
     FFMPEG_BIN,
     probe_duration,
+    probe_has_audio,
     probe_video_info,
     run_ffmpeg,
 )
@@ -823,6 +824,17 @@ class UnifiedRenderService:
                 )
                 audio_layer.clips.append(vo_clip)
 
+            # replace 模式：静音原视频音轨（main + broll 图层）
+            if tts_config.overlap_mode == "replace":
+                for layer in layers:
+                    if layer.role in ("main", "broll"):
+                        for clip in layer.clips:
+                            clip.config["volume"] = 0
+                logger.info(
+                    "TTS replace 模式：已静音原视频音轨: plan_id=%s",
+                    self.plan.id,
+                )
+
             logger.info(
                 "TTS 配音已添加: plan_id=%s voice_id=%s segments=%d total_%.2fs",
                 self.plan.id,
@@ -888,6 +900,16 @@ class UnifiedRenderService:
             )
             audio_layer.clips.append(vo_clip)
 
+            # 配音素材库默认替换原音：静音原视频音轨（main + broll 图层）
+            for layer in layers:
+                if layer.role in ("main", "broll"):
+                    for clip in layer.clips:
+                        clip.config["volume"] = 0
+            logger.info(
+                "配音素材库：已静音原视频音轨: plan_id=%s",
+                self.plan.id,
+            )
+
             logger.info(
                 "配音素材库音频已添加到 audio 图层: plan_id=%s duration=%.2fs",
                 self.plan.id,
@@ -1000,6 +1022,11 @@ class UnifiedRenderService:
         if abs(speed - 1.0) >= 1e-6:
             return False, f"有调速: speed={speed:.2f}x"
 
+        # 音量非默认（静音/放大）→ 需要音频滤镜重编码 → 不能 copy
+        vol = UnifiedRenderService._clip_volume(clip)
+        if abs(vol - 1.0) >= 1e-6:
+            return False, f"音量非默认: volume={vol:.2f}"
+
         # 有倒放 → 需要重编码 → 不能 copy
         reverse_config = ReverseConfig.from_dict(clip.config.get("reverse"))
         if reverse_config.enabled and (reverse_config.reverse_video or reverse_config.reverse_audio):
@@ -1270,13 +1297,23 @@ class UnifiedRenderService:
             "+faststart",
         ]
 
-        # 音频处理：background 通常是图片无音频，跳过；其他编码为 aac
-        # background 以外的视频素材，默认带音频
-        has_audio = role != "background"
+        # 音频处理：background 通常是图片无音频，跳过；其他角色先探测是否真有音频流。
+        # volume=0 不丢弃音频流，而是保留后通过 volume=0 滤镜静音，保持时间轴对齐。
+        clip_volume = UnifiedRenderService._clip_volume(clip)
+        if role != "background":
+            try:
+                has_audio = probe_has_audio(clip.local_path)
+            except Exception as e:
+                # probe_has_audio 内部已保守返回 True；只有极端错误才会到这里。
+                # 此时不静默丢音频，记录 error 并向上抛出，让任务失败而不是产出无声视频。
+                logger.error("[unified-render] 探测音频流发生致命错误，终止渲染: %s: %s", clip.local_path, e)
+                raise
+        else:
+            has_audio = False
         if has_audio:
             af_parts: list[str] = []
 
-            # 音频降噪
+            # 音频降噪（最先处理：在原始信号上降噪效果最好）
             try:
                 from video_processing.noise_reduction_engine import NoiseReductionConfig, NoiseReductionEngine
 
@@ -1290,13 +1327,16 @@ class UnifiedRenderService:
             except Exception as e:
                 logger.warning("[unified-render] 直通模式音频降噪应用失败，跳过: %s", e)
 
-            # 音频调速（与视频setpts对应，保持音画同步）
+            # 音频调速（在降噪之后、音量之前，与 render_audio.py concat 路径保持一致）
+            # SpeedEngine.build_audio_filter 内部已实现多级 atempo 串联，
+            # 自动处理超出 [0.5, 2.0] 范围的速度（如 0.25x → atempo=0.5,atempo=0.5）。
             speed = UnifiedRenderService._clip_speed(clip)
             if abs(speed - 1.0) >= 1e-6:
                 try:
                     from video_processing.speed_engine import SpeedConfig, SpeedEngine
 
                     speed_cfg = SpeedConfig(speed=speed)
+                    speed_cfg.clamp()
                     speed_engine = SpeedEngine()
                     af_parts.append(speed_engine.build_audio_filter(speed_cfg))
                 except Exception as e:
@@ -1309,6 +1349,10 @@ class UnifiedRenderService:
                 if af_filter:
                     af_parts.append(af_filter)
 
+            # 片段音量（最后应用：确保调速/倒放后的最终输出音量准确，与 concat 路径一致）
+            if abs(clip_volume - 1.0) >= 1e-6:
+                af_parts.append(f"volume={clip_volume:.4f}")
+
             if af_parts:
                 command.extend(["-af", ",".join(af_parts)])
 
@@ -1976,6 +2020,16 @@ class UnifiedRenderService:
         """
         return _clip_playback_speed_pure(getattr(clip, "playback_speed", 1.0))
 
+    @staticmethod
+    def _clip_volume(clip: ResolvedClip) -> float:
+        """获取 clip 的音量（config.volume）。缺省 1.0 原声，0.0 静音。"""
+        cfg = getattr(clip, "config", None) or {}
+        try:
+            vol = float(cfg.get("volume", 1.0))
+        except (TypeError, ValueError):
+            return 1.0
+        return max(0.0, vol)
+
     @staticmethod
     def _clip_adjusted_duration(clip: ResolvedClip) -> float:
         """计算调速后的 clip 实际时长（用于拼接计算）。

apps/worker/worker_app/celery_app.py

@@ -14,9 +14,17 @@ celery_app.conf.imports = (
     "worker_app.tasks.voice_extraction",
     "worker_app.tasks.voice_clone",
     "worker_app.tasks.tts_synthesis",
-    "worker_app.tasks.edit_plan_generation",
-    "worker_app.tasks.compose_video",
     "worker_app.tasks.batch_download",
     "worker_app.tasks._startup",
     "apps.worker.video_processing.dedup",
+    "worker_app.tasks.cleanup",
 )
+
+# Celery Beat 定时任务调度
+celery_app.conf.beat_schedule = {
+    "cleanup-stale-pending-tasks": {
+        "task": "worker.cleanup_stale_pending_tasks",
+        "schedule": 600.0,  # 每 10 分钟（秒）
+        "options": {"expires": 300},  # 5 分钟过期，避免堆积
+    },
+}

apps/worker/worker_app/tasks/__init__.py

@@ -25,10 +25,6 @@ def __getattr__(name: str):
         from .voice_extraction import extract_voice_task
 
         return extract_voice_task
-    elif name == "compose_video":
-        from .compose_video import compose_video
-
-        return compose_video
     elif name == "extract_background_task":
         from .voice_extraction import extract_background_task
 
@@ -53,12 +49,16 @@ def __getattr__(name: str):
         from .ai_tasks import run_generate_cover
 
         return run_generate_cover
+    elif name == "batch_generate_thumbnails":
+        from .batch_thumbnail import batch_generate_thumbnails
+
+        return batch_generate_thumbnails
     raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
 
 
 __all__ = [
+    "batch_generate_thumbnails",
     "classify_asset",
-    "compose_video",
     "generate_video",
     "healthcheck",
     "ingest_asset",

apps/worker/worker_app/tasks/_startup.py

@@ -10,6 +10,9 @@ logger = logging.getLogger(__name__)
 # 孤儿任务超时阈值：渲染任务超过此时间未更新则视为卡死
 ORPHAN_TASK_TIMEOUT_MINUTES = 10
 
+# Pending 任务超时阈值：pending 任务在队列中等待超过此时间则自动清理
+PENDING_TASK_TIMEOUT_MINUTES = 30
+
 
 def cleanup_orphan_tasks(timeout_minutes: int = ORPHAN_TASK_TIMEOUT_MINUTES) -> int:  # pragma: no cover
     """清理数据库中超时未更新的 running GenerationTask（孤儿任务）。
@@ -83,6 +86,38 @@ def cleanup_stale_jobs(timeout_minutes: int = ORPHAN_TASK_TIMEOUT_MINUTES) -> in
         return 0
 
 
+def cleanup_stale_pending_tasks(timeout_minutes: int = PENDING_TASK_TIMEOUT_MINUTES) -> int:  # pragma: no cover
+    """清理数据库中卡在 pending 状态超时的 GenerationTask。
+
+    全局任务队列有 pending 数量上限，长期卡在 pending 的任务会占满队列，
+    导致新用户无法创建任务。通过 created_at 超时判断并标记为 failed。
+
+    Args:
+        timeout_minutes: 超时时间（分钟），默认 PENDING_TASK_TIMEOUT_MINUTES
+
+    Returns:
+        清理的任务数量
+    """
+    from packages.adapters.sqlalchemy_impl.generation_task_repository import (
+        SQLAlchemyGenerationTaskRepository,
+    )
+
+    session = SessionLocal()
+    try:
+        repo = SQLAlchemyGenerationTaskRepository(session)
+        count = repo.cleanup_stale_pending(timeout_minutes)
+        if count > 0:
+            logger.warning("清理了 %d 个超时的 pending GenerationTask（超过 %d 分钟未处理）", count, timeout_minutes)
+        else:
+            logger.info("无超时 pending GenerationTask 需要清理")
+        return count
+    except Exception as e:
+        logger.error("清理超时 pending GenerationTask 失败: %s", e, exc_info=True)
+        return 0
+    finally:
+        session.close()
+
+
 def cleanup_all_stale_tasks(timeout_minutes: int = ORPHAN_TASK_TIMEOUT_MINUTES) -> dict:  # pragma: no cover
     """统一清理所有超时的孤儿任务。
 
@@ -93,15 +128,17 @@ def cleanup_all_stale_tasks(timeout_minutes: int = ORPHAN_TASK_TIMEOUT_MINUTES)
     """
     gen_count = cleanup_orphan_tasks(timeout_minutes)
     job_count = cleanup_stale_jobs(timeout_minutes)
-    total = gen_count + job_count
+    pending_count = cleanup_stale_pending_tasks(PENDING_TASK_TIMEOUT_MINUTES)
+    total = gen_count + job_count + pending_count
     if total > 0:
         logger.warning(
-            "孤儿任务清理完成: GenerationTask=%d, Job=%d, 总计=%d",
+            "任务清理完成: 孤儿 GenerationTask=%d, 孤儿 Job=%d, 超时 pending=%d, 总计=%d",
             gen_count,
             job_count,
+            pending_count,
             total,
         )
-    return {"generation_tasks": gen_count, "jobs": job_count}
+    return {"generation_tasks": gen_count, "jobs": job_count, "pending": pending_count}
 
 
 @worker_ready.connect