chore: remove temporary CI trigger file

新增 9 个单元测试覆盖 generation_tasks.py 增量代码：
- 空 plan_id 直接返回
- plan 不存在 warning + 不 commit
- 正常写入（有/无 title_config）
- config 非 dict 兜底
- DB 异常 rollback + rollback 失败静默
- 空 title_config 不写入

.gitea/workflows/api-base-image.yml

@@ -0,0 +1,84 @@
+name: API Base Image Build
+
+on:
+  push:
+    branches:
+      - develop
+      - main
+    paths:
+      - 'requirements-base.txt'
+      - 'requirements.txt'
+      - 'infra/docker/api-base.Dockerfile'
+  workflow_dispatch:
+
+jobs:
+  build-api-base:
+    name: Build API Base Image
+    runs-on: runtime-builder
+    timeout-minutes: 45
+    steps:
+    - name: Checkout code
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+      run: |
+        curl -sH "Authorization: token $GITHUB_TOKEN" \
+          "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" \
+          | bash
+
+    - name: Docker login to Registry
+      shell: sh
+      env:
+        ACR_USERNAME: ${{ secrets.ACR_USERNAME }}
+        ACR_PASSWORD: ${{ secrets.ACR_PASSWORD }}
+        GITEA_REGISTRY_USER: xiaoxia
+        GITEA_REGISTRY_TOKEN: ${{ secrets.REGISTRY_TOKEN }}
+      run: |
+        set -eu
+        for i in 1 2 3; do
+          echo "=== Docker login 尝试 $i/3 ==="
+          if printf '%s' "${ACR_PASSWORD}" | docker login xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com -u "${ACR_USERNAME}" --password-stdin \
+              && docker login git.xiaoxiajianji.com -u "${GITEA_REGISTRY_USER}" -p "${GITEA_REGISTRY_TOKEN}"; then
+            echo "✅ Docker login successful"
+            break
+          fi
+          echo "❌ Docker login 失败（尝试 $i/3），5s 后重试..."
+          sleep 5
+        done
+
+    - name: Build and push API base image
+      shell: sh
+      run: |
+        set -eu
+        ACR_IMAGE="xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com/xiaoxiakeji/saas-api-base:latest"
+        GITEA_IMAGE="git.xiaoxiajianji.com/xiaoxia-saas/saas-api-base:latest"
+
+        echo "=== Building API base image ==="
+
+        # 使用普通 docker build（单平台不需要 buildx）
+        docker build \
+          -f infra/docker/api-base.Dockerfile \
+          -t "${ACR_IMAGE}" \
+          .
+
+        echo ""
+        echo "✅ Image built successfully"
+
+        # 推送到 ACR
+        echo "=== Pushing to ACR ==="
+        docker push "${ACR_IMAGE}"
+        echo "✅ Pushed to ACR"
+
+        # 打标签并推送到 Gitea Packages 作为备份
+        echo "=== Pushing to Gitea Packages ==="
+        docker tag "${ACR_IMAGE}" "${GITEA_IMAGE}"
+        docker push "${GITEA_IMAGE}" || echo "⚠️ Gitea Packages push failed (non-fatal)"
+        echo "✅ Gitea backup push completed"
+
+    - name: Cleanup
+      if: always()
+      shell: sh
+      run: |
+        ACR_IMAGE="xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com/xiaoxiakeji/saas-api-base:latest"
+        docker rmi "${ACR_IMAGE}" 2>/dev/null || true
+        echo "Cleanup done"

.gitea/workflows/ci-pipeline.yml

@@ -633,58 +633,25 @@ jobs:
           echo "Docker login failed ($i/3), retrying in 5s..."
           sleep 5
         done
-    - name: Pre-build worker base images (fallback if not exist)
+    - name: Pre-build worker base image (fallback if not exist)
       if: matrix.service == 'worker'
       id: prebuild
       shell: sh
       run: |
         set -eu
-        REGISTRY="git.xiaoxiajianji.com/xiaoxia-saas"
-        BASE_BUILDER="${REGISTRY}/worker-base-builder:latest"
-        BASE_RUNTIME="${REGISTRY}/worker-base-runtime:latest"
+        REGISTRY="xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com/xiaoxiakeji"
+        BASE_IMAGE="${REGISTRY}/saas-worker-base:latest"
         
         # 尝试拉取基础镜像
-        echo "检查基础镜像..."
-        if docker pull "$BASE_BUILDER" 2>/dev/null && docker pull "$BASE_RUNTIME" 2>/dev/null; then
-          echo "基础镜像已存在，使用远程镜像"
+        echo "检查 Worker 基础镜像..."
+        if docker pull "$BASE_IMAGE" 2>/dev/null; then
+          echo "✅ 基础镜像已存在"
           echo "fallback=false" >> $GITHUB_OUTPUT
         else
-          echo "基础镜像不存在，本地构建（fallback模式）..."
-          
-          # 尝试用buildx构建，失败则回退到普通docker build（DooD模式下buildx builder偶发崩溃）
-          BUILDER_NAME="ci-pr-builder-${GITHUB_RUN_ID:-local}"
-          BUILDX_AVAILABLE=true
-          if ! docker buildx create --use --name "$BUILDER_NAME" --driver docker-container > /dev/null 2>&1; then
-            BUILDX_AVAILABLE=false
-          fi
-          if [ "$BUILDX_AVAILABLE" = true ] && ! docker buildx inspect --bootstrap > /dev/null 2>&1; then
-            BUILDX_AVAILABLE=false
-            docker buildx rm "$BUILDER_NAME" > /dev/null 2>&1 || true
-          fi
-          
-          build_base() {
-            local df="$1"
-            local tag="$2"
-            local name="$3"
-            if [ "$BUILDX_AVAILABLE" = true ]; then
-              echo "构建 $name（buildx）..."
-              if docker buildx build --load -f "$df" -t "$tag" . > /dev/null 2>&1; then
-                echo "$name 构建成功"
-                return 0
-              fi
-              echo "buildx失败，回退到普通docker build"
-              BUILDX_AVAILABLE=false
-              docker buildx rm "$BUILDER_NAME" > /dev/null 2>&1 || true
-            fi
-            echo "构建 $name（docker build）..."
-            docker build -f "$df" -t "$tag" .
-          }
-          
-          build_base infra/docker/worker-base-builder.Dockerfile "$BASE_BUILDER" "worker-base-builder"
-          build_base infra/docker/worker-base-runtime.Dockerfile "$BASE_RUNTIME" "worker-base-runtime"
-          
+          echo "⚠️ 基础镜像不存在，本地构建（fallback模式）..."
+          docker build -f infra/docker/worker-base.Dockerfile -t "$BASE_IMAGE" .
           echo "fallback=true" >> $GITHUB_OUTPUT
-          echo "基础镜像本地构建完成"
+          echo "✅ Worker 基础镜像本地构建完成"
         fi
 
     - name: Build PR image (verify only, no push)
@@ -700,15 +667,15 @@ jobs:
           EXTRA_BUILD_ARGS="$EXTRA_BUILD_ARGS NGINX_CONF=infra/docker/nginx-staging.conf"
         fi
 
-        # Worker fallback模式：基础镜像本地已构建，用普通docker build绕过buildx
-        if [ "${{ matrix.service }}" = "worker" ] && [ "${{ steps.prebuild.outputs.fallback }}" = "true" ]; then
-          echo "Fallback模式：用普通docker build（基础镜像本地已构建）"
+        # Worker: 始终用普通docker build（基础镜像已预装全部依赖，无需buildx）
+        if [ "${{ matrix.service }}" = "worker" ]; then
+          echo "Worker: 使用普通docker build"
           BUILD_ARG_STR=""
           for arg in $EXTRA_BUILD_ARGS; do
             BUILD_ARG_STR="$BUILD_ARG_STR --build-arg $arg"
           done
           docker build -f ${{ matrix.dockerfile }} -t "${IMAGE_TAG}" $BUILD_ARG_STR .
-          echo "Fallback PR Build successful"
+          echo "PR Build successful (worker, no buildx)"
           exit 0
         fi
 
@@ -831,6 +798,7 @@ jobs:
         fi
 
     - name: Setup buildx builder
+      if: matrix.service != 'worker'
       shell: sh
       run: |
         set -eu
@@ -843,41 +811,64 @@ jobs:
         fi
         docker buildx inspect --bootstrap
 
-    - name: Build and push ${{ matrix.service_display }} image (with retry)
+    - name: Pre-build worker base image (fallback if not exist)
+      if: matrix.service == 'worker'
+      shell: sh
+      run: |
+        set -eu
+        REGISTRY="xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com/xiaoxiakeji"
+        BASE_IMAGE="${REGISTRY}/saas-worker-base:latest"
+
+        echo "检查 Worker 基础镜像..."
+        if docker pull "$BASE_IMAGE" 2>/dev/null; then
+          echo "✅ 基础镜像已存在"
+        else
+          echo "⚠️ 基础镜像不存在，本地构建（fallback）..."
+          docker build -f infra/docker/worker-base.Dockerfile -t "$BASE_IMAGE" .
+          echo "✅ Worker 基础镜像本地构建完成"
+        fi
+
+    - name: Build and push ${{ matrix.service_display }} image
       shell: sh
       run: |
         set -eu
         REGISTRY="xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com/xiaoxiakeji"
         IMAGE_TAG="${REGISTRY}/${{ matrix.image_name }}:${GITHUB_SHA}"
-        CACHE_REF="${REGISTRY}/${{ matrix.cache_name }}:${GITHUB_REF_NAME}"
 
-        EXTRA_BUILD_ARGS="APP_VERSION=\"${GITHUB_SHA}\""
-        if [ "${{ matrix.service }}" = "web" ]; then
-          EXTRA_BUILD_ARGS="$EXTRA_BUILD_ARGS NGINX_CONF=infra/docker/nginx-staging.conf"
+        if [ "${{ matrix.service }}" = "worker" ]; then
+          # Worker: plain docker build（基础镜像已预装全部依赖，无需 buildx）
+          echo "=== Worker: plain docker build ==="
+          docker build -f ${{ matrix.dockerfile }} -t "${IMAGE_TAG}" --build-arg APP_VERSION="${GITHUB_SHA}" .
+          docker push "${IMAGE_TAG}"
+          echo "✅ Worker image pushed: ${IMAGE_TAG}"
+        else
+          # API/Web: buildx with registry cache
+          CACHE_REF="${REGISTRY}/${{ matrix.cache_name }}:${GITHUB_REF_NAME}"
+          EXTRA_BUILD_ARGS="APP_VERSION=\"${GITHUB_SHA}\""
+          if [ "${{ matrix.service }}" = "web" ]; then
+            EXTRA_BUILD_ARGS="$EXTRA_BUILD_ARGS NGINX_CONF=infra/docker/nginx-staging.conf"
+          fi
+
+          NO_CACHE_FLAG=""
+          for i in 1 2 3; do
+            echo "=== Docker build 尝试 $i/3 ==="
+            if bash scripts/ci/docker_build_push.sh $NO_CACHE_FLAG ${{ matrix.dockerfile }} "${IMAGE_TAG}" "${CACHE_REF}" $EXTRA_BUILD_ARGS; then
+              echo "✅ Docker build 成功"
+              break
+            fi
+            echo "❌ Docker build 失败（尝试 $i/3）"
+            [ $i -eq 3 ] && exit 1
+            sleep 10
+            if [ $i -eq 2 ]; then
+              NO_CACHE_FLAG="--no-cache"
+              echo "下次重试将使用 --no-cache"
+            fi
+          done
+
+          echo "${{ matrix.service_display }} image pushed: ${IMAGE_TAG}"
         fi
-
-        # Docker build 带重试：失败自动重试2次，第2次重试加--no-cache
-        NO_CACHE_FLAG=""
-        for i in 1 2 3; do
-          echo "=== Docker build 尝试 $i/3 ==="
-          if bash scripts/ci/docker_build_push.sh $NO_CACHE_FLAG ${{ matrix.dockerfile }} "${IMAGE_TAG}" "${CACHE_REF}" $EXTRA_BUILD_ARGS; then
-            echo "✅ Docker build 成功"
-            break
-          fi
-          echo "❌ Docker build 失败（尝试 $i/3）"
-          [ $i -eq 3 ] && exit 1
-          sleep 10
-          # 第2次重试使用 --no-cache
-          if [ $i -eq 2 ]; then
-            NO_CACHE_FLAG="--no-cache"
-            echo "下次重试将使用 --no-cache"
-          fi
-        done
-
-        echo
-        echo "${{ matrix.service_display }} image pushed: ${IMAGE_TAG}"
     - name: Cleanup buildx builder
-      if: always()
+      if: matrix.service != 'worker' && always()
       shell: sh
       run: |
         docker buildx rm ci-builder-${GITHUB_RUN_ID}-${GITHUB_JOB}-${{ matrix.cache_name }} 2>/dev/null || true

.gitea/workflows/worker-base-image.yml

@@ -7,35 +7,25 @@ on:
       - main
     paths:
       - 'requirements-base.txt'
+      - 'requirements.txt'
       - 'requirements-worker.txt'
-      - 'infra/docker/worker-base-builder.Dockerfile'
-      - 'infra/docker/worker-base-runtime.Dockerfile'
-  workflow_dispatch:  # 支持手动触发
+      - 'infra/docker/worker-base.Dockerfile'
+  workflow_dispatch:
 
 jobs:
   build-worker-base:
-    name: Build Worker Base Images
+    name: Build Worker Base Image
     runs-on: runtime-builder
-    timeout-minutes: 30
-    strategy:
-      fail-fast: false
-      matrix:
-        include:
-        - name: builder
-          dockerfile: infra/docker/worker-base-builder.Dockerfile
-          image_name: worker-base-builder
-          cache_name: worker-base-builder-cache
-        - name: runtime
-          dockerfile: infra/docker/worker-base-runtime.Dockerfile
-          image_name: worker-base-runtime
-          cache_name: worker-base-runtime-cache
+    timeout-minutes: 45
     steps:
     - name: Checkout code
       shell: sh
       env:
         GITHUB_TOKEN: ${{ github.token }}
       run: |
-        curl -sH "Authorization: token $GITHUB_TOKEN"           "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}"           | bash
+        curl -sH "Authorization: token $GITHUB_TOKEN" \
+          "${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/raw/scripts/ci/step_checkout.sh?ref=${GITHUB_SHA}" \
+          | bash
 
     - name: Docker login to Registry
       shell: sh
@@ -48,7 +38,8 @@ jobs:
         set -eu
         for i in 1 2 3; do
           echo "=== Docker login 尝试 $i/3 ==="
-          if printf '%s' "${ACR_PASSWORD}" | docker login xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com -u "${ACR_USERNAME}" --password-stdin              && docker login git.xiaoxiajianji.com -u "${GITEA_REGISTRY_USER}" -p "${GITEA_REGISTRY_TOKEN}"; then
+          if printf '%s' "${ACR_PASSWORD}" | docker login xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com -u "${ACR_USERNAME}" --password-stdin \
+              && docker login git.xiaoxiajianji.com -u "${GITEA_REGISTRY_USER}" -p "${GITEA_REGISTRY_TOKEN}"; then
             echo "✅ Docker login successful"
             break
           fi
@@ -56,48 +47,40 @@ jobs:
           sleep 5
         done
 
-    - name: Setup buildx builder
+    - name: Build and push Worker base image
       shell: sh
       run: |
         set -eu
-        BUILDER_NAME="ci-builder-${GITHUB_RUN_ID}-${{ matrix.name }}"
-        if ! docker buildx inspect "$BUILDER_NAME" > /dev/null 2>&1; then
-          docker buildx create --use --name "$BUILDER_NAME" --driver docker-container
-          echo "Created $BUILDER_NAME"
-        else
-          docker buildx use "$BUILDER_NAME"
-          echo "Using existing $BUILDER_NAME"
-        fi
-        docker buildx inspect --bootstrap
+        ACR_IMAGE="xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com/xiaoxiakeji/saas-worker-base:latest"
+        GITEA_IMAGE="git.xiaoxiajianji.com/xiaoxia-saas/saas-worker-base:latest"
+
+        echo "=== Building Worker base image ==="
+
+        # 使用普通 docker build（单平台不需要 buildx）
+        docker build \
+          -f infra/docker/worker-base.Dockerfile \
+          -t "${ACR_IMAGE}" \
+          .
 
-    - name: Build and push base image
-      shell: sh
-      run: |
-        set -eu
-        REGISTRY="xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com/xiaoxiakeji"
-        IMAGE_TAG="${REGISTRY}/${{ matrix.image_name }}:latest"
-        SAFE_REF_NAME=$(echo "${GITHUB_REF_NAME}" | tr '/' '-')
-        CACHE_REF="${REGISTRY}/${{ matrix.cache_name }}:${SAFE_REF_NAME}"
-        
-        echo "=== Building ${{ matrix.name }} base image ==="
-        echo "Image: ${IMAGE_TAG}"
-        echo "Cache: ${CACHE_REF}"
-        
-        # 用通用构建脚本
-        bash scripts/ci/docker_build_push.sh           ${{ matrix.dockerfile }}           "${IMAGE_TAG}"           "${CACHE_REF}"
-        
-        # 同时推送到 Gitea Packages 作为备份（可选）
-        GITEA_IMAGE="git.xiaoxiajianji.com/xiaoxia-saas/${{ matrix.image_name }}:latest"
-        docker tag "${IMAGE_TAG}" "${GITEA_IMAGE}"
-        docker push "${GITEA_IMAGE}" || echo "Gitea Packages push failed (non-fatal)"
-        
         echo ""
-        echo "✅ ${{ matrix.name }} base image built and pushed"
+        echo "✅ Image built successfully"
 
-    - name: Cleanup buildx builder
+        # 推送到 ACR
+        echo "=== Pushing to ACR ==="
+        docker push "${ACR_IMAGE}"
+        echo "✅ Pushed to ACR"
+
+        # 打标签并推送到 Gitea Packages 作为备份
+        echo "=== Pushing to Gitea Packages ==="
+        docker tag "${ACR_IMAGE}" "${GITEA_IMAGE}"
+        docker push "${GITEA_IMAGE}" || echo "⚠️ Gitea Packages push failed (non-fatal)"
+        echo "✅ Gitea backup push completed"
+
+    - name: Cleanup
       if: always()
       shell: sh
       run: |
-        docker buildx rm "ci-builder-${GITHUB_RUN_ID}-${{ matrix.name }}" 2>/dev/null || true
-        docker buildx prune -f 2>/dev/null || true
-        echo "Builder cleanup done"
+        ACR_IMAGE="xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com/xiaoxiakeji/saas-worker-base:latest"
+        docker rmi "${ACR_IMAGE}" 2>/dev/null || true
+        docker image prune -f 2>/dev/null || true
+        echo "Cleanup done"

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

@@ -1,5 +1,5 @@
 import logging
-from typing import Any, Optional
+from typing import Any, List, Optional
 
 from app.api.routes._helpers import check_project_access, format_utc_datetime
 from app.auth import AuthenticatedUser, get_current_user
@@ -14,10 +14,10 @@ from app.schemas.asset import (
     AssetResponse,
     BatchClassifyRequest,
     BatchDeleteRequest,
+    BatchGetRequest,
     BatchMarkRequest,
     BatchOperationResponse,
     BatchTagRequest,
-    CreateAssetRequest,
     ListAssetsResponse,
     SmartMatchItem,
     SmartMatchRequest,
@@ -28,11 +28,6 @@ from app.schemas.asset import (
 from app.schemas.tag import TagAssetsRequest
 from fastapi import APIRouter, Depends, HTTPException, Query, Response
 
-from packages.application import (
-    CreateAssetCommand,
-    CreateAssetUseCase,
-)
-from packages.domain import AssetStatus, ClassificationStatus
 from packages.domain.smart_match import smart_select_assets
 
 logger = logging.getLogger(__name__)
@@ -370,6 +365,18 @@ def update_asset_review_status(
     return _to_asset_response(updated)
 
 
+@router.post("/batch", response_model=List[AssetResponse])
+def batch_get_assets(
+    request: BatchGetRequest,
+    authenticated_user: AuthenticatedUser = Depends(get_current_user),
+    asset_repository: Any = Depends(get_asset_repository),
+) -> list[AssetResponse]:
+    """批量获取素材详情（根据 ID 列表）。"""
+    items = asset_repository.find_by_ids(request.ids)
+    storage_service = get_storage_service()
+    return [_to_asset_response(item, storage_service) for item in items]
+
+
 @router.post("/batch-delete", response_model=BatchOperationResponse)
 def batch_delete_assets(
     request: BatchDeleteRequest,
@@ -664,51 +671,12 @@ def untag_asset(
 
 
 @router.post("", response_model=AssetResponse)
-def create_asset(
-    request: CreateAssetRequest,
-    authenticated_user: AuthenticatedUser = Depends(get_current_user),
-    asset_repository: Any = Depends(get_asset_repository),
-    asset_library_repository: Any = Depends(get_asset_library_repository),
-    project_repository: Any = Depends(get_project_repository),
-) -> AssetResponse:
-    # 先获取素材库，用于推导 project_id（前端可能不传）
-    library = asset_library_repository.get(request.library_id)
-    if library is None:
-        raise HTTPException(status_code=404, detail=f"AssetLibrary {request.library_id} not found")
-
-    # project_id 自动推导：优先用请求值，否则从 library 关联的项目获取
-    project_id = request.project_id or library.project_id
-
-    project = project_repository.find_by_id(project_id)
-    if project is None:
-        raise HTTPException(status_code=404, detail=f"Project {project_id} not found")
-    if not project.can_access(authenticated_user.user.id):
-        raise HTTPException(status_code=403, detail="Access denied to project")
-
-    # 确保 library 和 project 归属一致
-    if library.project_id != project_id:
-        raise HTTPException(status_code=400, detail="AssetLibrary does not belong to the specified project")
-
-    use_case = CreateAssetUseCase(asset_repository)
-    item = use_case.execute(
-        CreateAssetCommand(
-            project_id=project_id,
-            library_id=request.library_id,
-            name=request.name,
-            storage_key=request.storage_key,
-            mime_type=request.mime_type,
-            metadata=request.metadata,
-            file_size=request.file_size,
-            thumbnail_url=request.thumbnail_url,
-            duration=request.duration,
-            width=request.width,
-            height=request.height,
-            fps=request.fps,
-            codec=request.codec,
-            status=AssetStatus(request.status),
-            classification_status=ClassificationStatus(request.classification_status),
-            quality_score=request.quality_score,
-            uploaded_by_user_id=authenticated_user.user.id,
-        )
+def create_asset() -> None:
+    """
+    已废弃接口。
+    所有素材上传统一走 uploadAssetDirect → completeDirectUpload → ingest-jobs 流程。
+    """
+    raise HTTPException(
+        status_code=410,
+        detail="此接口已废弃。请使用 uploadAssetDirect 接口上传素材，Worker 会自动处理（视频转码、图片/音频元数据提取）并创建 Asset 记录。",
     )
-    return _to_asset_response(item)

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

@@ -31,8 +31,6 @@ logger = logging.getLogger(__name__)
 router = APIRouter(tags=["Generation"])
 
 
-
-
 # ── Schemas ──────────────────────────────────────────────────────────────
 
 
@@ -49,6 +47,10 @@ class GenerateCoverRequest(BaseModel):
         ge=0.0,
         description="手动选帧时间点（秒），仅 cover_type=manual 时有效",
     )
+    cover_url: Optional[str] = Field(
+        default=None,
+        description="上传的封面图片 URL，仅 cover_type=upload 时有效",
+    )
 
 
 class GenerateCoverResponse(BaseModel):
@@ -61,6 +63,66 @@ class GenerateCoverResponse(BaseModel):
 # ── Route ────────────────────────────────────────────────────────────────
 
 
+def _persist_cover_frame(frame_url: str, plan_id: str, title_text: str = "") -> 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）
+    """
+    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)
+                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,
@@ -80,6 +142,29 @@ def generate_cover(
     _, plan_svc = services
     plan = plan_svc.get_plan_or_raise(plan_id)
 
+    # ── upload 类型：直接保存前端上传的封面图片，不需要预览视频 ──────
+    if body.cover_type == "upload":
+        if not body.cover_url:
+            raise HTTPException(
+                status_code=400,
+                detail="cover_type=upload 时必须提供 cover_url",
+            )
+        cover_data = {
+            "type": "upload",
+            "image_url": body.cover_url,
+        }
+        current_config = dict(plan.config) if plan.config else {}
+        current_config["cover"] = cover_data
+        normalized = normalize_plan_config(current_config)
+        plan_svc.update_plan_config(plan_id, {"cover": normalized["cover"]})
+        logger.info(
+            "封面上传完成: plan_id=%s cover_url=%s by user=%s",
+            plan_id,
+            body.cover_url[:80] if body.cover_url else "",
+            current_user.user.id,
+        )
+        return GenerateCoverResponse(plan_id=plan_id, cover=cover_data)
+
     # ── 3 步查找预览视频 URL ──────────────────────────────────────────
     # 第一步：从 plan.config 读取
     logger.info("[封面生成] 步骤1: 从 plan.config 查找 rendered_storage_key: plan_id=%s", plan_id)
@@ -171,44 +256,31 @@ 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
@@ -283,6 +355,118 @@ 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 读取标题，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 ""
+            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)
+                                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 = {
@@ -298,13 +482,13 @@ 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_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,
@@ -69,6 +71,7 @@ def _to_generation_task_response(task) -> GenerationTaskResponse:
         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 +145,54 @@ 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
+        if title_config:
+            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 +238,7 @@ 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",
@@ -304,6 +356,7 @@ def create_generation_task(
                     output_height=request.output_height,
                     cover_url=request.cover_url,
                     custom_title=request.custom_title,
+                    title_config=request.title_config or {},
                 )
             )
             try:
@@ -315,6 +368,13 @@ def create_generation_task(
                     log_task_status=True,
                 ):
                     created_tasks.append(task)
+                    if request.source_edit_plan_id:
+                        _writeback_edit_plan_config(
+                            plan_id=request.source_edit_plan_id,
+                            task_id=task.id,
+                            title_config=request.title_config,
+                            db=db,
+                        )
                 else:
                     failed_tasks.append(task)
             except UserPendingLimitExceeded as _e:

apps/api/app/api/routes/health.py

@@ -1,6 +1,6 @@
 from datetime import datetime, timezone
 
-import psycopg2
+import psycopg
 import redis
 from app.config import settings
 from fastapi import APIRouter, status
@@ -49,7 +49,7 @@ async def _check_database() -> dict:
             "message": "Using in-memory database",
         }
     try:
-        conn = psycopg2.connect(settings.DATABASE_URL, connect_timeout=3)
+        conn = psycopg.connect(settings.DATABASE_URL, connect_timeout=3)
         with conn.cursor() as cur:
             cur.execute("SELECT 1")
             cur.fetchone()
@@ -124,7 +124,7 @@ async def _check_migrations() -> dict:
             "message": "Using in-memory database, no migrations needed",
         }
     try:
-        conn = psycopg2.connect(settings.DATABASE_URL, connect_timeout=3)
+        conn = psycopg.connect(settings.DATABASE_URL, connect_timeout=3)
         with conn.cursor() as cur:
             cur.execute("""
                 SELECT COUNT(*) FROM information_schema.tables

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

@@ -16,13 +16,16 @@
 from __future__ import annotations
 
 import logging
-from typing import Any
 
 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.services.edit_plan_service import EditPlanService
 from app.services.edit_template_service import EditTemplateService
 from fastapi import APIRouter, Depends, HTTPException, Query, status
 
+from packages.adapters.sqlalchemy_impl.asset_repository import SQLAlchemyAssetRepository
+
 from .dependencies import get_draft_plan_id, get_editor_services
 from .schemas import (
     ClipBatchDeleteRequest,
@@ -43,30 +46,100 @@ logger = logging.getLogger(__name__)
 router = APIRouter(tags=["Template Editor"])
 
 
-def _clip_to_response(clip) -> EditorClipResponse:
-    """统一构造片段响应"""
+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"""
+
+    def _enum_str(val) -> str:
+        return val.value if hasattr(val, "value") else str(val)
+
+    def _fmt_dt(val) -> str:
+        if val is None:
+            return ""
+        if hasattr(val, "isoformat"):
+            return val.isoformat()
+        return str(val)
+
     return EditorClipResponse(
         id=clip.id,
         plan_id=clip.plan_id,
-        clip_type=clip.clip_type.value
-        if hasattr(clip.clip_type, "value")
-        else str(clip.clip_type),
+        clip_type=_enum_str(getattr(clip, "clip_type", "")),
         order=clip.order,
         duration=clip.duration,
+        start_time=getattr(clip, "start_time", 0.0) or 0.0,
         text_content=clip.text_content or "",
-        transition_effect=clip.transition_effect.value
-        if hasattr(clip.transition_effect, "value")
-        else str(clip.transition_effect),
+        transition_effect=_enum_str(getattr(clip, "transition_effect", "cut")),
+        transition_duration=getattr(clip, "transition_duration", 0.0) or 0.0,
         playback_speed=clip.playback_speed or 1.0,
+        asset_id=getattr(clip, "asset_id", "") or "",
+        asset_url=asset_url,
+        status=getattr(clip, "status", "pending") or "pending",
+        template_clip_config_id=getattr(clip, "template_clip_config_id", "") or "",
         config=clip.config or {},
+        created_at=_fmt_dt(getattr(clip, "created_at", None)),
+        updated_at=_fmt_dt(getattr(clip, "updated_at", None)),
     )
 
 
+def _build_asset_url_map(
+    asset_ids: list[str],
+    asset_repo: SQLAlchemyAssetRepository,
+) -> dict[str, str | None]:
+    """批量查询素材并生成签名URL映射.
+
+    Returns:
+        {asset_id: signed_url_or_None}
+    """
+    if not asset_ids:
+        return {}
+
+    # 去重：多个 clip 可能引用同一个素材
+    # 去重并保持顺序
+    seen: set[str] = set()
+    unique_ids = []
+    for aid in asset_ids:
+        if aid and aid not in seen:
+            seen.add(aid)
+            unique_ids.append(aid)
+
+    result: dict[str, str | None] = {}
+    try:
+        storage = get_storage_service()
+    except Exception:
+        logger.warning("获取存储服务失败，跳过asset_url生成")
+        return {aid: None for aid in asset_ids}
+
+    # 批量查询所有 Asset（单次 SQL IN 查询，避免 N+1）
+    try:
+        assets = asset_repo.find_by_ids(unique_ids)
+        asset_map = {a.id: a for a in assets}
+    except Exception:
+        logger.warning("批量查询素材失败: asset_ids=%s", asset_ids, exc_info=True)
+        return {aid: None for aid in asset_ids if aid}
+
+    for aid in unique_ids:
+        try:
+            asset = asset_map.get(aid)
+            if asset is None:
+                result[aid] = None
+                continue
+            storage_key = getattr(asset, "storage_key", None) or ""
+            if not storage_key:
+                result[aid] = None
+                continue
+            result[aid] = storage.get_download_url(storage_key, expires_seconds=3600)
+        except Exception:
+            logger.warning("生成素材签名URL失败: asset_id=%s", aid, exc_info=True)
+            result[aid] = None
+
+    return result
+
+
 @router.get("/clips", response_model=EditorClipListResponse)
 def list_draft_clips(
     template_id: str,
     plan_id: str = Depends(get_draft_plan_id),
     services: tuple[EditTemplateService, EditPlanService] = Depends(get_editor_services),
+    asset_repo: SQLAlchemyAssetRepository = Depends(get_asset_repository),
     skip: int = Query(default=0, ge=0),
     limit: int = Query(default=100, ge=1, le=500),
     _: AuthenticatedUser = Depends(get_current_user),
@@ -75,8 +148,17 @@ def list_draft_clips(
     _, plan_svc = services
     clips = plan_svc.list_clips(plan_id, skip=skip, limit=limit)
     total = plan_svc.count_clips(plan_id)
+
+    # 批量解析素材签名URL
+    asset_ids = [getattr(c, "asset_id", "") or "" for c in clips]
+    asset_ids = [aid for aid in asset_ids if aid]
+    url_map = _build_asset_url_map(asset_ids, asset_repo)
+
     return EditorClipListResponse(
-        items=[_clip_to_response(c) 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,
     )
 
@@ -156,6 +238,7 @@ def get_draft_clip_detail(
     clip_id: str,
     plan_id: str = Depends(get_draft_plan_id),
     services: tuple[EditTemplateService, EditPlanService] = Depends(get_editor_services),
+    asset_repo: SQLAlchemyAssetRepository = Depends(get_asset_repository),
     _: AuthenticatedUser = Depends(get_current_user),
 ):
     """获取草稿中的片段详情"""
@@ -165,16 +248,20 @@ def get_draft_clip_detail(
         raise HTTPException(status_code=404, detail="片段不存在")
     if clip.plan_id != plan_id:
         raise HTTPException(status_code=404, detail="片段不存在")
-    return _clip_to_response(clip)
+
+    asset_id = getattr(clip, "asset_id", "") or ""
+    url_map = _build_asset_url_map([asset_id], asset_repo) if asset_id else {}
+    return _clip_to_response(clip, asset_url=url_map.get(asset_id))
 
 
-@router.post("/clips/{clip_id}/split", response_model=dict[str, Any], status_code=status.HTTP_200_OK)
+@router.post("/clips/{clip_id}/split", status_code=status.HTTP_200_OK)
 def split_draft_clip(
     template_id: str,
     clip_id: str,
     body: SplitClipRequest,
     plan_id: str = Depends(get_draft_plan_id),
     services: tuple[EditTemplateService, EditPlanService] = Depends(get_editor_services),
+    asset_repo: SQLAlchemyAssetRepository = Depends(get_asset_repository),
     _: AuthenticatedUser = Depends(get_current_user),
 ):
     """将一个片段从指定时间点分割为两个片段"""
@@ -190,32 +277,22 @@ def split_draft_clip(
         ) from exc
     left = result["left_clip"]
     right = result["right_clip"]
+    asset_ids = [getattr(left, "asset_id", "") or "", getattr(right, "asset_id", "") or ""]
+    asset_ids = [a for a in asset_ids if a]
+    url_map = _build_asset_url_map(asset_ids, asset_repo)
     return {
-        "left_clip": {
-            "id": left.id,
-            "plan_id": left.plan_id,
-            "clip_type": left.clip_type,
-            "order": left.order,
-            "duration": left.duration,
-            "start_time": left.start_time,
-        },
-        "right_clip": {
-            "id": right.id,
-            "plan_id": right.plan_id,
-            "clip_type": right.clip_type,
-            "order": right.order,
-            "duration": right.duration,
-            "start_time": right.start_time,
-        },
+        "left_clip": _clip_to_response(left, asset_url=url_map.get(getattr(left, "asset_id", "") or "")),
+        "right_clip": _clip_to_response(right, asset_url=url_map.get(getattr(right, "asset_id", "") or "")),
     }
 
 
-@router.post("/clips/merge", response_model=dict[str, Any], status_code=status.HTTP_200_OK)
+@router.post("/clips/merge", status_code=status.HTTP_200_OK)
 def merge_draft_clips(
     template_id: str,
     body: MergeClipsRequest,
     plan_id: str = Depends(get_draft_plan_id),
     services: tuple[EditTemplateService, EditPlanService] = Depends(get_editor_services),
+    asset_repo: SQLAlchemyAssetRepository = Depends(get_asset_repository),
     _: AuthenticatedUser = Depends(get_current_user),
 ):
     """将多个连续的同类型片段合并为一个片段"""
@@ -230,13 +307,11 @@ def merge_draft_clips(
         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 {
-        "id": merged.id,
-        "plan_id": merged.plan_id,
-        "clip_type": merged.clip_type,
-        "order": merged.order,
-        "duration": merged.duration,
-        "text_content": merged.text_content,
+        "merged_clip": _clip_to_response(merged, asset_url=url_map.get(asset_id)),
+        "deleted_clip_ids": body.clip_ids,
     }
 
 

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

@@ -8,8 +8,9 @@
 
 from __future__ import annotations
 
+import json
 import logging
-from typing import Any
+from typing import Any, Optional
 
 from app.auth import AuthenticatedUser, get_current_user
 from app.core.celery_app import celery_app
@@ -45,6 +46,7 @@ from ._fallback import (
 from .dependencies import _check_queue_limits, get_draft_plan_id, get_editor_services
 from .schemas import (
     ClipStatusItem,
+    EditPlanGenerateRequest,
     EditPlanGenerateResponse,
     EditPlanGenerationsResponse,
     EditPlanGenerationStatusResponse,
@@ -54,9 +56,11 @@ logger = logging.getLogger(__name__)
 router = APIRouter(tags=["Template Editor"])
 
 
+# DEPRECATED: 前端已改用 /generation/tasks 体系，此路由保留仅供旧版兼容，计划下线
 @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),
@@ -65,6 +69,7 @@ def generate_editor_draft(
     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)
 
@@ -85,6 +90,36 @@ def generate_editor_draft(
     # 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()
@@ -133,6 +168,21 @@ def generate_editor_draft(
         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 "",
@@ -140,6 +190,8 @@ def generate_editor_draft(
                 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,
             ),
         )
 
@@ -234,6 +286,7 @@ def _get_task_output_url(task, gen_task_repo, db) -> str:
     return ""
 
 
+# DEPRECATED: 前端已改用 /generation/tasks 体系，此路由保留仅供旧版兼容，计划下线
 @router.get("/generation-status", response_model=EditPlanGenerationStatusResponse)
 def get_editor_generation_status(
     template_id: str,

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

@@ -6,7 +6,7 @@
 from __future__ import annotations
 
 import re as _re
-from typing import Any, List, Optional
+from typing import Any, Dict, List, Optional
 
 from app.schemas.generation_task import GenerationTaskResponse
 from pydantic import BaseModel, Field, validator
@@ -44,6 +44,14 @@ class EditPlanGenerationStatusResponse(BaseModel):
     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"""
 
@@ -227,6 +235,7 @@ class ClipsFromAssetsResponse(BaseModel):
 
     success: bool = True
     created_count: int
+    plan_id: str = ""
     message: str = ""
     clip_ids: List[str] = Field(default_factory=list, description="创建的片段ID列表")
 
@@ -439,17 +448,28 @@ class EditorUpdateRequest(BaseModel):
 
 
 class EditorClipResponse(BaseModel):
-    """片段响应"""
+    """片段响应 — 与数据库 edit_plan_clips 表字段对齐"""
 
     id: str
     plan_id: str
     clip_type: str
     order: int
     duration: float
+    start_time: float = 0.0
     text_content: str = ""
     transition_effect: str = "cut"
+    transition_duration: float = 0.0
     playback_speed: float = 1.0
+    asset_id: str = ""
+    asset_url: str | None = Field(
+        default=None,
+        description="素材视频签名URL（1小时有效），用于前端预览播放",
+    )
+    status: str = "pending"
+    template_clip_config_id: str = ""
     config: dict[str, Any] = Field(default_factory=dict)
+    created_at: str = ""
+    updated_at: str = ""
 
 
 class EditorClipListResponse(BaseModel):

apps/api/app/api/routes/upload.py

@@ -206,6 +206,7 @@ async def complete_direct_upload(
                 ingest_job_id="",
                 duplicated=True,
                 asset_id=existing.id,
+                url=storage_service.get_url(normalized_key),
             )
 
     job = _submit_ingest_job(
@@ -215,7 +216,7 @@ async def complete_direct_upload(
         ingest_job_repository=ingest_job_repository,
         file_hash=request.file_hash,
     )
-    return DirectUploadCompleteResponse(storage_key=normalized_key, ingest_job_id=job.id)
+    return DirectUploadCompleteResponse(storage_key=normalized_key, ingest_job_id=job.id, url=storage_service.get_url(normalized_key))
 
 
 @router.post(

apps/api/app/schemas/asset.py

@@ -58,6 +58,12 @@ class AssetResponse(BaseModel):
 MAX_BATCH_SIZE = 200
 
 
+class BatchGetRequest(BaseModel):
+    """批量获取素材详情请求。"""
+
+    ids: list[str] = Field(..., min_length=1, max_length=MAX_BATCH_SIZE, description="素材 ID 列表")
+
+
 class BatchDeleteRequest(BaseModel):
     """批量删除请求（软删除）。"""
 

apps/api/app/schemas/generation_task.py

@@ -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，优先于 custom_title 纯文本）──
+    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 ──
@@ -109,6 +114,7 @@ class GenerationTaskResponse(BaseModel):
     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/upload.py

@@ -39,6 +39,7 @@ class DirectUploadCompleteResponse(BaseModel):
     ingest_job_id: str
     duplicated: bool = Field(default=False, description="是否为重复素材（命中去重）")
     asset_id: str = Field(default="", description="重复素材的 asset_id（duplicated=true 时返回）")
+    url: str = Field(default="", description="Public URL of uploaded file")
 
 
 class UploadAssetResponse(BaseModel):

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

@@ -214,12 +214,8 @@ test.describe("Core generation flow", () => {
     await titleInput.fill(titleText)
     await page.getByRole("button", { name: "下一步" }).click()
 
-    // Step 5: preview — 需要先生成预览视频，才能进入下一步
-    await expect(page.getByRole("heading", { name: /生成预览/ })).toBeVisible({ timeout: 15000 })
-    // 点击"生成预览"按钮触发预览生成
-    await page.locator(".xx-preview-generate-btn").click()
-    // 等待预览生成完成（后端渲染，可能需要较长时间）
-    await expect(page.getByText("预览生成成功")).toBeVisible({ timeout: 300_000 })
+    // Step 5: preview — 前端实时预览架构改造，无需后端生成预览
+    await expect(page.getByRole("heading", { name: /预览设置/ })).toBeVisible({ timeout: 15000 })
     await page.getByRole("button", { name: "下一步" }).click()
 
     // Step 6: cover (默认 AI 智能选帧模式，直接下一步)
@@ -230,16 +226,12 @@ test.describe("Core generation flow", () => {
     await expect(page.getByRole("heading", { name: /确认生成/ })).toBeVisible()
 
     // Wait for generation API to be called
-    // 确认生成走新流程：POST /tasks/{taskId}/confirm（复用预览产物）
-    // 或旧流程：POST /editor/generate（向后兼容）
+    // 前端直接创建生成任务：POST /generation/tasks
     const generatePromise = page.waitForResponse(
       (response) => {
         const url = response.url()
         const path = new URL(url).pathname
-        return (
-          response.request().method() === "POST" &&
-          (path.endsWith("/confirm") || path.endsWith("/editor/generate"))
-        )
+        return response.request().method() === "POST" && path.endsWith("/generation/tasks")
       },
       { timeout: 30_000 },
     )

apps/web/e2e/test_asset.spec.ts

@@ -178,7 +178,7 @@ test.describe("素材库流程", () => {
     expect(kinds).toContain("image")
   })
 
-  test("创建素材记录", async ({ request }) => {
+  test("创建素材记录 — POST /assets 已废弃返回 410", async ({ request }) => {
     const { headers, userId } = await createAuthedUser(request, "asset-create")
     const projectId = await createProject(request, headers, Date.now().toString())
 
@@ -194,7 +194,7 @@ test.describe("素材库流程", () => {
     expect(lib.ok()).toBeTruthy()
     const libData = await lib.json()
 
-    // 创建素材记录
+    // POST /assets 已废弃，应返回 410 Gone
     const response = await request.post(`${apiBase}/assets`, {
       headers,
       data: {
@@ -210,16 +210,9 @@ test.describe("素材库流程", () => {
       },
     })
 
-    expect(
-      response.ok(),
-      `创建素材应返回 2xx，实际: ${response.status()} ${await response.text()}`,
-    ).toBeTruthy()
-
+    expect(response.status()).toBe(410)
     const data = await response.json()
-    expect(data.id, "应返回素材 ID").toBeTruthy()
-    expect(data.name).toContain("test_video")
-    expect(data.mime_type).toBe("video/mp4")
-    expect(data.library_id).toBe(libData.id)
+    expect(data.error?.code).toBe("HTTP_410")
   })
 
   test("列出素材", async ({ request }) => {
@@ -232,51 +225,50 @@ test.describe("素材库流程", () => {
       data: {
         project_id: projectId,
         name: `List Lib ${Date.now()}`,
-        kind: "video",
+        kind: "image",
       },
     })
     expect(lib.ok(), `创建素材库应成功: ${await lib.text()}`).toBeTruthy()
     const libData = await lib.json()
 
-    // 创建 2 个素材
-    await request.post(`${apiBase}/assets`, {
+    // 通过 multipart upload 上传 2 个小图片作为测试素材
+    // 创建一个 1x1 的 PNG buffer
+    const tinyPng = Buffer.from(
+      "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNk+M9QDwADhgGAWjR9awAAAABJRU5ErkJggg==",
+      "base64",
+    )
+
+    await request.post(`${apiBase}/upload`, {
       headers,
-      data: {
+      multipart: {
         project_id: projectId,
         library_id: libData.id,
-        name: `clip_a_${Date.now()}.mp4`,
-        storage_key: `uploads/e2e/clip_a.mp4`,
-        mime_type: "video/mp4",
-        status: "ready",
-        uploaded_by_user_id: userId,
+        file: { name: "clip_a.png", mimeType: "image/png", buffer: tinyPng },
       },
     })
-    await request.post(`${apiBase}/assets`, {
+    await request.post(`${apiBase}/upload`, {
       headers,
-      data: {
+      multipart: {
         project_id: projectId,
         library_id: libData.id,
-        name: `clip_b_${Date.now()}.mp4`,
-        storage_key: `uploads/e2e/clip_b.mp4`,
-        mime_type: "video/mp4",
-        status: "ready",
-        uploaded_by_user_id: userId,
+        file: { name: "clip_b.png", mimeType: "image/png", buffer: tinyPng },
       },
     })
 
-    // 列出素材
-    const response = await request.get(`${apiBase}/assets`, {
-      headers,
-      params: { library_id: libData.id },
-    })
+    // 列出素材（可能需要等待 ingest job 完成）
+    let items: any[] = []
+    for (let i = 0; i < 10; i++) {
+      const response = await request.get(`${apiBase}/assets`, {
+        headers,
+        params: { library_id: libData.id },
+      })
+      expect(response.ok(), `列出素材应返回 2xx`).toBeTruthy()
+      const data = await response.json()
+      items = data.items || []
+      if (items.length >= 2) break
+      await new Promise((r) => setTimeout(r, 2000))
+    }
 
-    expect(
-      response.ok(),
-      `列出素材应返回 2xx，实际: ${response.status()} ${await response.text()}`,
-    ).toBeTruthy()
-
-    const data = await response.json()
-    const items = data.items || []
     expect(items.length, "应至少有 2 个素材").toBeGreaterThanOrEqual(2)
   })
 

apps/web/package-lock.json

@@ -12,6 +12,7 @@
         "@tanstack/react-query": "^5.45.0",
         "antd": "^5.18.0",
         "axios": "^1.7.2",
+        "mp4box": "^2.4.1",
         "react": "^18.3.1",
         "react-dom": "^18.3.1",
         "react-router-dom": "^6.24.0",
@@ -4623,6 +4624,15 @@
       "dev": true,
       "license": "MIT"
     },
+    "node_modules/mp4box": {
+      "version": "2.4.1",
+      "resolved": "https://registry.npmmirror.com/mp4box/-/mp4box-2.4.1.tgz",
+      "integrity": "sha512-0HGX7nXoDIX6FKLVl4a3wtYjBlwqsN3xuQC3GXzNtKp98FXUOhDSq623azsz8DG5ptd9ZXcXodDkgbdMZOjWvw==",
+      "license": "BSD-3-Clause",
+      "engines": {
+        "node": ">=20.8.1"
+      }
+    },
     "node_modules/mrmime": {
       "version": "2.0.1",
       "resolved": "https://registry.npmjs.org/mrmime/-/mrmime-2.0.1.tgz",

apps/web/package.json

@@ -23,6 +23,7 @@
     "@tanstack/react-query": "^5.45.0",
     "antd": "^5.18.0",
     "axios": "^1.7.2",
+    "mp4box": "^2.4.1",
     "react": "^18.3.1",
     "react-dom": "^18.3.1",
     "react-router-dom": "^6.24.0",

apps/web/src/api/assets/assets.ts

@@ -60,18 +60,6 @@ export const smartMatchAssets = async (libraryId: string): Promise<{ items: Asse
   return response.data
 }
 
-/** 创建素材（上传文件后调用，附带 metadata） */
-export const createAsset = async (data: {
-  library_id: string
-  name: string
-  storage_key: string
-  mime_type: string
-  metadata?: AssetMetadata
-}): Promise<AssetItem> => {
-  const response = await apiClient.post("/assets", data)
-  return response.data
-}
-
 /** 更新素材（名称、metadata 等） */
 export const updateAsset = async (
   assetId: string,

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

@@ -13,7 +13,6 @@ export type {
   ClassificationJob,
   AssetDiagnosis,
   BatchOperationResult,
-  UploadResult,
   DirectUploadPrepareResult,
   DirectUploadCompleteResult,
 } from "./types"
@@ -34,14 +33,13 @@ export {
   getAssets,
   getAssetsByKind,
   smartMatchAssets,
-  createAsset,
   updateAsset,
   updateAssetReviewStatus,
   deleteAsset,
 } from "./assets"
 
 // 上传
-export { uploadAsset, prepareDirectUpload, completeDirectUpload, uploadAssetDirect } from "./upload"
+export { prepareDirectUpload, completeDirectUpload, uploadAssetDirect } from "./upload"
 
 // 任务
 export { getIngestJob, submitClassificationJob, getClassificationJob } from "./jobs"

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

@@ -135,4 +135,5 @@ export interface DirectUploadPrepareResult {
 export interface DirectUploadCompleteResult {
   storage_key: string
   ingest_job_id: string
+  url: string
 }

apps/web/src/api/assets/upload.ts

@@ -3,16 +3,7 @@
  */
 import apiClient from "../client"
 import { getOrCreateDefaultProject } from "../projects"
-import type { UploadResult, DirectUploadPrepareResult, DirectUploadCompleteResult } from "./types"
-
-/** 表单上传素材（小文件） */
-export const uploadAsset = async (formData: FormData): Promise<UploadResult> => {
-  const response = await apiClient.post("/upload", formData, {
-    headers: { "Content-Type": "multipart/form-data" },
-    timeout: 30 * 60 * 1000,
-  })
-  return response.data
-}
+import type { DirectUploadPrepareResult, DirectUploadCompleteResult } from "./types"
 
 /** 预签名直传准备 */
 export const prepareDirectUpload = async (data: {

apps/web/src/api/auth/tokenRefresh.ts

@@ -0,0 +1,109 @@
+/**
+ * 主动 Token 刷新模块
+ *
+ * 在 access_token 过期前主动刷新，避免 API 请求触发 401。
+ * JWT payload 是 base64 编码的 JSON，无需第三方库即可解码。
+ */
+import { useAuthStore } from "@/store/authStore"
+import { refreshAccessToken } from "./login"
+
+let refreshTimer: ReturnType<typeof setTimeout> | null = null
+
+/** 正在执行刷新操作的 Promise，防止主动刷新和 401 被动刷新并发竞争 */
+let activeRefreshPromise: Promise<void> | null = null
+
+/** 提前刷新的缓冲时间（秒） */
+const REFRESH_BUFFER_SECONDS = 60
+
+/**
+ * 解码 JWT payload（不验签，仅读取 exp 字段）
+ */
+function decodeJwtPayload(token: string): { exp?: number } | null {
+  try {
+    const parts = token.split(".")
+    if (parts.length !== 3) return null
+    // JWT 使用 base64url 编码，需要转换为标准 base64
+    const payload = parts[1].replace(/-/g, "+").replace(/_/g, "/")
+    const padded = payload + "=".repeat((4 - (payload.length % 4)) % 4)
+    const decoded = atob(padded)
+    return JSON.parse(decoded)
+  } catch {
+    return null
+  }
+}
+
+/**
+ * 取消已调度的主动刷新
+ */
+export function cancelProactiveRefresh(): void {
+  if (refreshTimer) {
+    clearTimeout(refreshTimer)
+    refreshTimer = null
+  }
+}
+
+/**
+ * 执行 token 刷新（带并发锁，供主动刷新和被动 401 共用）
+ * 返回当前刷新操作的 Promise；若已有刷新进行中则复用该 Promise。
+ */
+export function executeTokenRefresh(): Promise<void> | null {
+  // 已有刷新进行中 → 复用
+  if (activeRefreshPromise) {
+    return activeRefreshPromise
+  }
+
+  const { user, refreshToken: refreshTokenValue } = useAuthStore.getState()
+
+  // 安全检查：user 或 refreshToken 为空时跳过刷新
+  if (!user || !refreshTokenValue) {
+    return null
+  }
+
+  activeRefreshPromise = (async () => {
+    try {
+      const data = await refreshAccessToken(refreshTokenValue)
+      const newAccessToken = data.access_token
+      const newRefreshToken = data.refresh_token ?? refreshTokenValue
+
+      // 更新 Zustand store + localStorage
+      useAuthStore.getState().setAuth(user, newAccessToken, newRefreshToken)
+
+      // 递归调度下一次刷新
+      scheduleProactiveRefresh()
+    } catch {
+      // 刷新失败 → 清除认证状态，跳转登录页
+      cancelProactiveRefresh()
+      useAuthStore.getState().clearAuth()
+      window.location.href = "/login"
+    } finally {
+      activeRefreshPromise = null
+    }
+  })()
+
+  return activeRefreshPromise
+}
+
+/**
+ * 调度主动刷新：在 token 过期前 REFRESH_BUFFER_SECONDS 秒自动刷新
+ */
+export function scheduleProactiveRefresh(): void {
+  cancelProactiveRefresh()
+
+  // 统一从 Zustand store 读取（与 setAuth 写入保持一致）
+  const { accessToken, refreshToken: refreshTokenValue } = useAuthStore.getState()
+
+  if (!accessToken || !refreshTokenValue) return
+
+  const payload = decodeJwtPayload(accessToken)
+  if (!payload?.exp) return
+
+  const now = Math.floor(Date.now() / 1000)
+  const secondsUntilExpiry = payload.exp - now
+
+  // 如果 token 已经过期或即将在缓冲时间内过期，立即刷新
+  const delaySeconds = Math.max(secondsUntilExpiry - REFRESH_BUFFER_SECONDS, 0)
+
+  refreshTimer = setTimeout(() => {
+    executeTokenRefresh()
+  }, delaySeconds * 1000)
+}

apps/web/src/api/client.ts

@@ -5,7 +5,8 @@
 import axios, { AxiosError, InternalAxiosRequestConfig } from "axios"
 import { message } from "antd"
 import { useAuthStore } from "@/store/authStore"
-import { refreshAccessToken } from "./auth"
+
+import { cancelProactiveRefresh, executeTokenRefresh } from "./auth/tokenRefresh"
 
 // 创建 Axios 实例
 const apiClient = axios.create({
@@ -97,14 +98,22 @@ apiClient.interceptors.response.use(
       isRefreshing = true
 
       try {
-        const data = await refreshAccessToken(refreshToken)
-        const newAccessToken = data.access_token
-        const newRefreshToken = data.refresh_token ?? refreshToken
+        // 使用共享的刷新函数（带并发锁 + 安全检查）
+        const refreshPromise = executeTokenRefresh()
+        if (!refreshPromise) {
+          // user 或 refreshToken 为空，无法刷新
+          cancelProactiveRefresh()
+          useAuthStore.getState().clearAuth()
+          window.location.href = "/"
+          return Promise.reject(new Error("Unable to refresh: missing user or refresh token"))
+        }
+        await refreshPromise
 
-        // 更新 Zustand + localStorage
-        useAuthStore
-          .getState()
-          .setAuth(useAuthStore.getState().user!, newAccessToken, newRefreshToken)
+        // 获取刷新后的新 token
+        const newAccessToken = useAuthStore.getState().accessToken
+        if (!newAccessToken) {
+          return Promise.reject(new Error("Token refresh failed: no new access token"))
+        }
 
         // 处理排队的请求
         processQueue(null, newAccessToken)
@@ -116,6 +125,7 @@ apiClient.interceptors.response.use(
         return apiClient(originalRequest)
       } catch (refreshError) {
         // 刷新失败 → 登出
+        cancelProactiveRefresh()
         processQueue(refreshError, null)
         useAuthStore.getState().clearAuth()
         window.location.href = "/"

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

@@ -23,13 +23,9 @@ export async function generateCover(
   templateId: string,
   data: GenerateCoverRequest,
 ): Promise<GenerateCoverResponse> {
-  const response = await apiClient.post<GenerateCoverResponse>(
-    "/generation/generate-cover",
-    { ...data, template_id: templateId },
-    {
-      timeout: 300000,
-      params: { template_id: templateId },
-    },
-  )
+  const response = await apiClient.post<GenerateCoverResponse>("/generation/generate-cover", data, {
+    timeout: 300000,
+    params: { template_id: templateId },
+  })
   return response.data
 }

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

@@ -3,13 +3,9 @@ export type {
   CreatePreviewRequest,
   CreatePreviewResponse,
   PreviewTaskResponse,
-  ConfirmGenerationRequest,
-  ConfirmGenerationResponse,
-  ConfirmGenerationTaskItem,
 } from "./types"
 
 export { createPreview, getPreviewStatus } from "./preview"
-export { confirmGeneration } from "./confirm"
 
 export { generateCover } from "./cover"
 export type { GenerateCoverRequest, GenerateCoverResponse } from "./cover"

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

@@ -57,8 +57,31 @@ export interface TaskListResponse {
 export interface CreateGenerationTaskRequest {
   template_id: string
   asset_ids: string[]
-  title_ids: string[]
-  voice_ids: string[]
+  title_ids?: string[]
+  voice_ids?: string[]
+  /** 输出视频宽度 */
+  output_width?: number
+  /** 输出视频高度 */
+  output_height?: number
+  /** 自定义封面图片 URL */
+  cover_url?: string
+  /** 自定义视频标题 */
+  custom_title?: string
+  /** 视频时长（秒） */
+  duration?: number
+  /** 视频宽高比，如 "9:16" */
+  video_ratio?: string
+  /** 标题烧录配置 */
+  title_config?: {
+    text?: string
+    font?: string
+    font_size?: number
+    font_color?: string
+    position?: string
+    bold?: boolean
+    stroke?: boolean
+    shadow?: boolean
+  }
 }
 
 /** 创建生成任务响应（对齐后端 GenerationTaskResponse） */

apps/web/src/components/voice/CloneModal/index.tsx

@@ -1,7 +1,8 @@
 import React, { useState, useCallback, useRef, useEffect } from "react"
 import { Modal, Button } from "@/components/ui"
 import { createVoiceClone, toVoiceClone } from "@/api/voice-clone"
-import { uploadAsset } from "@/api/assets"
+import { uploadAssetDirect, ensureDefaultLibrary } from "@/api/assets"
+import { getOrCreateDefaultProject } from "@/api/projects"
 import { PROGRESS_STEPS, ACCEPTED_MIME } from "./constants"
 import { validateFile } from "./utils"
 import { useAudioRecorder } from "./hooks/useAudioRecorder"
@@ -181,9 +182,15 @@ const CloneModal: React.FC<CloneModalProps> = ({ open, onClose, onSuccess }) =>
         })
       }
 
-      const formData = new FormData()
-      formData.append("file", fileToUpload)
-      const uploadResult = await uploadAsset(formData)
+      // 获取默认项目和素材库
+      const project = await getOrCreateDefaultProject()
+      const library = await ensureDefaultLibrary({ project_id: project.id, kind: "voice" })
+
+      // 直传到 OSS
+      const uploadResult = await uploadAssetDirect({
+        file: fileToUpload,
+        library_id: library.id,
+      })
 
       // 组件已卸载则中止后续操作
       if (!isMountedRef.current) return

apps/web/src/components/voice/hooks/useCloneSubmit.ts

@@ -1,6 +1,6 @@
 import { useRef, useCallback, useEffect } from "react"
 import { createVoiceClone, toVoiceClone } from "@/api/voice-clone"
-import { uploadAsset, ensureDefaultLibrary } from "@/api/assets"
+import { uploadAssetDirect, ensureDefaultLibrary } from "@/api/assets"
 import { getOrCreateDefaultProject } from "@/api/projects"
 import type { VoiceClone } from "@/api/voice-clone"
 
@@ -62,15 +62,15 @@ export function useCloneSubmit({
         })
       }
 
-      // 获取默认项目和素材库（后端 /upload 接口必填）
+      // 获取默认项目和素材库
       const project = await getOrCreateDefaultProject()
       const library = await ensureDefaultLibrary({ project_id: project.id, kind: "voice" })
 
-      const formData = new FormData()
-      formData.append("file", fileToUpload)
-      formData.append("project_id", project.id)
-      formData.append("library_id", library.id)
-      const uploadResult = await uploadAsset(formData)
+      // 直传到 OSS
+      const uploadResult = await uploadAssetDirect({
+        file: fileToUpload,
+        library_id: library.id,
+      })
 
       // 阶段 2：克隆
       setPhase("cloning")

apps/web/src/hooks/useAuth.ts

@@ -4,6 +4,7 @@
 import { useMutation, useQuery, useQueryClient } from "@tanstack/react-query"
 import { useNavigate } from "react-router-dom"
 import * as authApi from "@/api/auth"
+import { scheduleProactiveRefresh, cancelProactiveRefresh } from "@/api/auth/tokenRefresh"
 import { useAuthStore } from "@/store/authStore"
 
 // 登录 Hook
@@ -31,6 +32,9 @@ export const useLogin = () => {
     const user = await authApi.getCurrentUser()
     setAuth(user, data.access_token, refreshToken)
 
+    // 启动主动 token 刷新，避免后续请求触发 401
+    scheduleProactiveRefresh()
+
     // 跳转到登录前页面或仪表盘（与 Login.tsx onFinish 保持一致）
     const redirect = localStorage.getItem("login_redirect") || "/app/dashboard"
     localStorage.removeItem("login_redirect")
@@ -73,6 +77,9 @@ export const useWechatCallback = () => {
     const user = await authApi.getCurrentUser()
     setAuth(user, result.access_token, result.refresh_token)
 
+    // 启动主动 token 刷新
+    scheduleProactiveRefresh()
+
     return { ...result, user }
   }
 
@@ -121,6 +128,7 @@ export const useLogout = () => {
     } catch (error) {
       // 即使登出失败也清除本地状态
     } finally {
+      cancelProactiveRefresh()
       clearAuth()
       queryClient.clear()
       navigate("/")

apps/web/src/main.tsx

@@ -9,6 +9,13 @@ import { QueryClient, QueryClientProvider } from "@tanstack/react-query"
 import { ConfigProvider, App as AntApp } from "antd"
 import zhCN from "antd/locale/zh_CN"
 import router from "./router"
+import { scheduleProactiveRefresh } from "./api/auth/tokenRefresh"
+
+// 应用启动时，如果用户已登录，立即调度主动 token 刷新
+// 这样可以在 token 过期前自动刷新，避免 API 请求触发 401
+if (localStorage.getItem("access_token")) {
+  scheduleProactiveRefresh()
+}
 import "./index.css"
 import "./styles/global.css"
 

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

@@ -1,24 +1,27 @@
 /**
- * 智能剪辑页面 — V22 多预览 + 配音前置
- * 7 步向导：选择模板 → 选择素材 → 选择配音 → 选择标题 → 生成预览 → 选择封面 → 确认生成
+ * 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
  * 左右布局：左侧 generate-form + 右侧 generate-preview
- * 主组件仅保留整体布局与事件编排
- * 状态管理 → hooks/useGenerateFormState
- * 步骤导航 → hooks/useStepNavigation
- * 步骤内容 → components/GenerateStepContent
- * 底部按钮 → components/GenerateStepActions
- * 生成核心逻辑 → hooks/useGenerateVideo
+ *
+ * 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）
  */
-import React, { useState, useMemo } from "react"
+import React, { useMemo } from "react"
 import { Modal, message } from "antd"
 import { useNavigate } from "react-router-dom"
-import { useQuery } from "@tanstack/react-query"
 import type { VoiceClone } from "@/api/voice-clone"
-import { getAssetsByKind } from "@/api/assets/assets"
-import type { AssetItem } from "@/api/assets/types"
+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 GenerateStepsBar from "./components/GenerateStepsBar"
 import GenerateResultPanel from "./components/GenerateResultPanel"
 import PreviewVideoPanel from "./components/PreviewVideoPanel"
@@ -27,7 +30,8 @@ import GenerateStepActions from "./components/GenerateStepActions"
 import { useGenerateFormState } from "./hooks/useGenerateFormState"
 import { useStepNavigation } from "./hooks/useStepNavigation"
 import { useGenerateVideo } from "./hooks/useGenerateVideo"
-import { useStep5Preview } from "./hooks/useStep5Preview"
+import { usePreviewAssets } from "./hooks/usePreviewAssets"
+import { useTitleStyleUpdaters } from "./hooks/useStep4Title/useTitleStyleUpdaters"
 import "./generate.css"
 
 const GeneratePage: React.FC = () => {
@@ -74,20 +78,12 @@ const GeneratePage: React.FC = () => {
     setPreviewModalOpen,
   } = formState
 
-  /* ── 查询视频素材，用于 Step4 标题预览背景 ── */
-  const { data: videoAssets = [] } = useQuery({
-    queryKey: ["generate-video-assets"],
-    queryFn: () => getAssetsByKind("video", { limit: 50 }),
+  /* ── 标题样式回调（Step5 样式面板 + 右侧预览 CSS 层共用） ── */
+  const styleUpdaters = useTitleStyleUpdaters({
+    titleSettings,
+    onTitleSettingsChange: setTitleSettings,
   })
 
-  // 获取第一个选中素材的 URL
-  const sourceVideoUrl = useMemo(() => {
-    const firstId = selectedMaterials[0]
-    if (!firstId) return undefined
-    const asset = videoAssets.find((a: AssetItem) => a.id === firstId)
-    return asset?.file_url
-  }, [selectedMaterials, videoAssets])
-
   /* ── 克隆声音 ── */
   const { clones: clonedVoices, addClone, hasProcessing } = useCloneProgress()
 
@@ -97,37 +93,44 @@ const GeneratePage: React.FC = () => {
     message.success("音色克隆成功！")
   }
 
-  /* ── 预览数量（多预览） ── */
-  const [previewCount, setPreviewCount] = useState(1)
+  /* ── 前端预览：加载选中素材的视频文件信息 ── */
+  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)
 
-  /* ── 根据 voiceMode 构建 voiceIds 传给预览接口 ── */
-  /* selectedVoice / selectedClonedVoice 均为 string 类型（voice ID），
-     见 useGenerateFormState 返回值类型定义 */
-  const previewVoiceIds = useMemo((): string[] => {
-    if (voiceMode === "clone") {
-      const id: string = selectedClonedVoice
-      return id ? [id] : []
-    }
-    // preset / custom 模式
-    const id: string = selectedVoice
-    return id ? [id] : []
-  }, [voiceMode, selectedVoice, selectedClonedVoice])
+  /* ── 当前模板对象（传给前端预览播放器） ── */
+  const currentTemplate = useMemo(
+    () => userTemplates.find((t) => t.id === selectedTemplate) || null,
+    [userTemplates, selectedTemplate],
+  )
 
-  /* ── Step5 预览生成（多预览 + voice_ids） ── */
-  const step5Preview = useStep5Preview({
-    templates: userTemplates,
-    selectedTemplate,
-    materialMode,
-    selectedMaterials,
-    smartSelectedIds,
-    duration,
-    videoRatio,
-    voiceIds: previewVoiceIds,
-    voiceLibraryId: selectedVoice || undefined,
-    previewCount,
-    titleSettings,
+  /* ── 视频总时长计算（用于配音时长校验） ── */
+  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,
@@ -137,7 +140,7 @@ const GeneratePage: React.FC = () => {
     selectedMaterials,
     smartSelectedIds,
     titleSettings,
-    previewReady: step5Preview.canProceed,
+    previewReady: previewAssetsReady,
   })
 
   /* ── 视频生成核心逻辑 ── */
@@ -168,7 +171,6 @@ const GeneratePage: React.FC = () => {
     autoSubtitles,
     bgm,
     generateCount,
-    previewTaskId: step5Preview.selectedTaskId,
   })
 
   /* ================================================================
@@ -200,11 +202,23 @@ 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}
+            onToggleBold={styleUpdaters.toggleBold}
+            onToggleItalic={styleUpdaters.toggleItalic}
+            onToggleStroke={styleUpdaters.toggleStroke}
+            onToggleShadow={styleUpdaters.toggleShadow}
+            onApplyPreset={styleUpdaters.applyPreset}
+            activePreset={styleUpdaters.activePreset}
+            titlePresets={styleUpdaters.titlePresets}
             coverSettings={coverSettings}
             onCoverSettingsChange={setCoverSettings}
             duration={duration}
             selectedVoice={selectedVoice}
             onSelectedVoiceChange={setSelectedVoice}
+            totalVideoDuration={totalVideoDuration}
             voiceMode={voiceMode}
             onVoiceModeChange={setVoiceMode}
             selectedClonedVoice={selectedClonedVoice}
@@ -224,19 +238,6 @@ const GeneratePage: React.FC = () => {
             onRetry={handleRetryGenerate}
             onDismissError={handleDismissError}
             presetVoices={presetVoices}
-            videoRatio={videoRatio}
-            /* Step5 多预览 */
-            previewCount={previewCount}
-            onPreviewCountChange={setPreviewCount}
-            previewItems={step5Preview.items}
-            previewSelectedIndex={step5Preview.selectedIndex}
-            onSelectPreview={step5Preview.setSelectedIndex}
-            previewOverallStatus={step5Preview.previewStatus}
-            previewOverallError={step5Preview.previewError}
-            previewOverallProgress={step5Preview.progress}
-            previewAnyGenerating={step5Preview.anyGenerating}
-            onGeneratePreview={step5Preview.generatePreview}
-            onRegeneratePreview={step5Preview.regeneratePreview}
           />
 
           <GenerateStepActions
@@ -252,23 +253,18 @@ const GeneratePage: React.FC = () => {
 
         {/* ════ 右侧：预览 + 生成结果 ════ */}
         <div className="xx-generate-right-col">
-          {/* 预览视频面板（Step4+ 显示，Step4 显示标题预览叠加，Step5+ 仅视频） */}
+          {/* 预览视频面板（Step4+ 显示，含 CSS 标题实时预览层） */}
           {currentStep >= 4 && (
             <PreviewVideoPanel
-              previewStatus={step5Preview.previewStatus}
-              previewResult={step5Preview.previewResult}
-              previewError={step5Preview.previewError}
-              progress={step5Preview.progress}
+              assets={previewAssets}
+              template={currentTemplate}
               videoRatio={videoRatio}
-              onRegenerate={step5Preview.regeneratePreview}
-              titleText={titleSettings.title}
+              assetsReady={previewAssetsReady}
+              assetsLoading={previewAssetsLoading}
               titleSettings={titleSettings}
-              showTitlePreview={currentStep === 4}
-              sourceVideoUrl={sourceVideoUrl}
+              voiceAudioUrl={voiceAudioUrl}
             />
           )}
-
-          {/* 正式生成结果（Step6+ 才显示） */}
           {currentStep >= 6 && (
             <GenerateResultPanel
               generated={generated}
@@ -290,6 +286,7 @@ const GeneratePage: React.FC = () => {
 
       {/* ── 视频预览弹窗 ── */}
       <Modal
+        className="xx-preview-modal"
         open={previewModalOpen}
         onCancel={() => setPreviewModalOpen(false)}
         footer={null}

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

@@ -0,0 +1,525 @@
+/**
+ * 前端预览播放器 — Canvas + WebCodecs 方案
+ *
+ * 架构：
+ * - 浏览器支持 WebCodecs → Canvas 渲染（帧级精确控制 + 标题合成）
+ * - 浏览器不支持 → fallback 到多 video 元素方案
+ *
+ * 对外 API 不变：assets, template, videoRatio, ready, voiceAudioUrl
+ */
+import React, { useMemo, useCallback, useState, useRef, useEffect } from "react"
+import {
+  PlayCircleOutlined,
+  PauseCircleOutlined,
+  SoundOutlined,
+  LoadingOutlined,
+} from "@ant-design/icons"
+import type { AssetItem } from "@/api/assets"
+import type { EditingTemplate } from "@/api/editing-planner"
+import { useSegmentScheduler, type PlaybackSegment } from "../hooks/useSegmentScheduler"
+import { useCanvasPlayer } from "../hooks/useCanvasPlayer"
+
+interface FrontendPreviewPlayerProps {
+  assets: AssetItem[]
+  template: EditingTemplate | null
+  videoRatio: string
+  ready: boolean
+  voiceAudioUrl?: string
+  titleSettings?: {
+    title: string
+    size: number
+    font: string
+    color: string
+    position: "top" | "center" | "bottom"
+    bold?: boolean
+    italic?: boolean
+    stroke?: boolean
+    shadow?: boolean
+  }
+}
+
+function formatTime(seconds: number): string {
+  const m = Math.floor(seconds / 60)
+  const s = Math.floor(seconds % 60)
+  return `${m}:${s.toString().padStart(2, "0")}`
+}
+
+/**
+ * 将素材映射为播放片段（复用原逻辑）
+ */
+function buildPlaybackSegments(
+  assets: AssetItem[],
+  template: EditingTemplate | null,
+): PlaybackSegment[] {
+  if (!assets.length) return []
+
+  const templateSegments = template?.segments || []
+  const segments: PlaybackSegment[] = []
+
+  assets.forEach((asset, i) => {
+    const assetDuration = asset.duration || asset.metadata?.duration || 30
+    const tplSeg = templateSegments[i] || templateSegments[templateSegments.length - 1]
+    const segDuration = tplSeg
+      ? Math.min(tplSeg.duration_max, Math.max(tplSeg.duration_min, assetDuration))
+      : Math.min(assetDuration, 10)
+
+    const startTime = 0
+    const endTime = Math.min(startTime + segDuration, assetDuration)
+    const videoUrl = asset.file_url || asset.storage_key
+
+    segments.push({ assetId: asset.id, videoUrl, startTime, endTime, order: i })
+  })
+
+  return segments
+}
+
+const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
+  assets,
+  template,
+  videoRatio: _videoRatio,
+  ready,
+  voiceAudioUrl,
+  titleSettings,
+}) => {
+  const segments = useMemo(() => buildPlaybackSegments(assets, template), [assets, template])
+  // 默认走原生 video 播放（浏览器硬件解码，独立线程，不阻塞 UI）
+  // WebCodecs 仅在明确需要时启用（保留代码作为兜底）
+  const useWebCodecs = false
+
+  // ── 两条路径共用同一个 canvas ref（fallback 路径不使用） ──
+  const canvasRef = useRef<HTMLCanvasElement>(null)
+
+  // ── Canvas 播放器（WebCodecs 路径） ──
+  const canvasTitle = titleSettings
+    ? {
+        text: titleSettings.title || "标题预览",
+        fontSize: titleSettings.size,
+        fontFamily: titleSettings.font || "思源黑体",
+        color: titleSettings.color || "#ffffff",
+        position: titleSettings.position || "bottom",
+        bold: titleSettings.bold,
+        stroke: titleSettings.stroke,
+        shadow: titleSettings.shadow,
+      }
+    : undefined
+
+  const canvasSegments = useMemo(
+    () =>
+      segments.map((s) => ({
+        assetId: s.assetId,
+        videoUrl: s.videoUrl,
+        startTime: s.startTime,
+        endTime: s.endTime,
+      })),
+    [segments],
+  )
+
+  // WebCodecs 解码失败后强制走 video fallback
+  const [forceVideoFallback, setForceVideoFallback] = useState(false)
+
+  const handleCanvasError = useCallback((err: Error) => {
+    console.error("[FrontendPreviewPlayer] Canvas decode Error, switching to video fallback:", err)
+    setForceVideoFallback(true)
+  }, [])
+
+  const { state: canvasState, controls: canvasControls } = useCanvasPlayer(
+    canvasRef,
+    useWebCodecs && !forceVideoFallback ? canvasSegments : [],
+    useWebCodecs && !forceVideoFallback ? canvasTitle : undefined,
+    handleCanvasError,
+    useWebCodecs && !forceVideoFallback,
+  )
+
+  // WebCodecs 报告解码失败时自动切换到 video fallback
+  useEffect(() => {
+    if (canvasState.hasDecodeError && !forceVideoFallback) {
+      console.warn("[FrontendPreviewPlayer] hasDecodeError detected, forcing video fallback")
+      setForceVideoFallback(true)
+    }
+  }, [canvasState.hasDecodeError, forceVideoFallback])
+
+  // ── Video 播放器（fallback 路径） ──
+  const {
+    isPlaying: videoIsPlaying,
+    currentTime: videoCurrentTime,
+    totalDuration: videoTotalDuration,
+    currentSegmentIndex: videoCurrentSegIdx,
+    canPlay: videoCanPlay,
+    togglePlayPause: videoTogglePlayPause,
+    seekTo: videoSeekTo,
+    videoRefs,
+  } = useSegmentScheduler(segments)
+
+  // 选择哪条路径的状态（WebCodecs 解码失败时强制走 video fallback）
+  const effectiveUseWebCodecs = useWebCodecs && !forceVideoFallback
+  const isPlaying = effectiveUseWebCodecs ? canvasState.isPlaying : videoIsPlaying
+  const currentTime = effectiveUseWebCodecs ? canvasState.currentTime : videoCurrentTime
+  const totalDuration = effectiveUseWebCodecs ? canvasState.duration : videoTotalDuration
+  const canPlay = effectiveUseWebCodecs ? canvasState.isReady : videoCanPlay
+  const isBuffering = effectiveUseWebCodecs ? canvasState.isBuffering : false
+
+  // ── 配音音频同步 ──
+  const audioRef = useRef<HTMLAudioElement | null>(null)
+  const prevIsPlayingRef = useRef(false)
+
+  useEffect(() => {
+    if (!voiceAudioUrl) {
+      if (audioRef.current) {
+        audioRef.current.pause()
+        audioRef.current.src = ""
+        audioRef.current = null
+      }
+      return
+    }
+    if (!audioRef.current) {
+      audioRef.current = new Audio()
+      audioRef.current.preload = "auto"
+    }
+    if (audioRef.current.src !== voiceAudioUrl) {
+      audioRef.current.src = voiceAudioUrl
+    }
+  }, [voiceAudioUrl])
+
+  useEffect(() => {
+    const audio = audioRef.current
+    if (!audio || !audio.src) return
+    if (isPlaying && !prevIsPlayingRef.current) {
+      audio.currentTime = currentTime
+      audio.play().catch(() => {})
+    } else if (!isPlaying && prevIsPlayingRef.current) {
+      audio.pause()
+    }
+    prevIsPlayingRef.current = isPlaying
+  }, [isPlaying, currentTime])
+
+  // 片段切换时同步音频（仅 fallback 路径需要）
+  const segmentSyncKey = effectiveUseWebCodecs ? -1 : videoCurrentSegIdx
+  useEffect(() => {
+    const audio = audioRef.current
+    if (!audio || !audio.src || !isPlaying) return
+    audio.currentTime = 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) => {
+      if (effectiveUseWebCodecs) {
+        canvasControls.seek(time)
+      } else {
+        videoSeekTo(time)
+      }
+      const audio = audioRef.current
+      if (audio && audio.src) {
+        audio.currentTime = time
+      }
+    },
+    [effectiveUseWebCodecs, canvasControls, videoSeekTo],
+  )
+
+  const handleTogglePlay = useCallback(() => {
+    if (effectiveUseWebCodecs) {
+      if (canvasState.isPlaying) {
+        canvasControls.pause()
+      } else {
+        canvasControls.play()
+      }
+    } else {
+      videoTogglePlayPause()
+    }
+  }, [effectiveUseWebCodecs, canvasState.isPlaying, canvasControls, videoTogglePlayPause])
+
+  // ── 进度条拖拽 ──
+  const [isDragging, setIsDragging] = useState(false)
+  const progressRef = useRef<HTMLDivElement>(null)
+
+  const handleProgressClick = useCallback(
+    (e: React.MouseEvent<HTMLDivElement>) => {
+      if (!progressRef.current || totalDuration <= 0) return
+      const rect = progressRef.current.getBoundingClientRect()
+      const ratio = Math.max(0, Math.min(1, (e.clientX - rect.left) / rect.width))
+      handleSeekTo(ratio * totalDuration)
+    },
+    [totalDuration, handleSeekTo],
+  )
+
+  const handleMouseDown = useCallback(
+    (e: React.MouseEvent<HTMLDivElement>) => {
+      setIsDragging(true)
+      handleProgressClick(e)
+    },
+    [handleProgressClick],
+  )
+
+  useEffect(() => {
+    if (!isDragging) return
+    const handleMouseMove = (e: MouseEvent) => {
+      if (!progressRef.current || totalDuration <= 0) return
+      const rect = progressRef.current.getBoundingClientRect()
+      const ratio = Math.max(0, Math.min(1, (e.clientX - rect.left) / rect.width))
+      handleSeekTo(ratio * totalDuration)
+    }
+    const handleMouseUp = () => setIsDragging(false)
+    window.addEventListener("mousemove", handleMouseMove)
+    window.addEventListener("mouseup", handleMouseUp)
+    return () => {
+      window.removeEventListener("mousemove", handleMouseMove)
+      window.removeEventListener("mouseup", handleMouseUp)
+    }
+  }, [isDragging, totalDuration, handleSeekTo])
+
+  const progressPercent = totalDuration > 0 ? (currentTime / totalDuration) * 100 : 0
+
+  // ── Canvas 容器 ref（保留声明，WebCodecs 兜底路径仍引用） ──
+  const canvasContainerRef = useRef<HTMLDivElement>(null)
+
+  // ── 未就绪 ──
+  if (!ready || !assets.length) {
+    return (
+      <div
+        className="xx-preview-empty"
+        style={{
+          position: "absolute",
+          inset: 0,
+          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>
+      </div>
+    )
+  }
+
+  // ── 无播放片段 ──
+  if (!canPlay) {
+    const showDecodeError = forceVideoFallback && canvasState.hasDecodeError
+    return (
+      <div
+        className="xx-preview-empty"
+        style={{
+          position: "absolute",
+          inset: 0,
+          display: "flex",
+          flexDirection: "column",
+          alignItems: "center",
+          justifyContent: "center",
+          zIndex: 1,
+        }}
+      >
+        {isBuffering ? (
+          <>
+            <LoadingOutlined style={{ fontSize: 48, color: "#fff", marginBottom: 12 }} spin />
+            <p style={{ color: "rgba(255,255,255,0.8)" }}>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)" }}>
+              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" }}
+            >
+              {canvasState.errorMessage || "当前浏览器不支持该视频编码格式，请刷新重试"}
+            </p>
+          </>
+        ) : (
+          <>
+            <PlayCircleOutlined
+              style={{ fontSize: 48, color: "var(--text-tertiary)", 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>
+          </>
+        )}
+      </div>
+    )
+  }
+
+  return (
+    <>
+      {/* ── Canvas 渲染层（WebCodecs 路径） ── */}
+      {effectiveUseWebCodecs && (
+        <div
+          ref={canvasContainerRef}
+          style={{
+            position: "absolute",
+            inset: 0,
+            zIndex: 1,
+            background: "#000",
+          }}
+        >
+          <canvas
+            ref={canvasRef}
+            style={{
+              width: "100%",
+              height: "100%",
+              objectFit: "contain",
+            }}
+          />
+        </div>
+      )}
+
+      {/* ── Video 渲染层（默认路径，浏览器原生硬件解码） ── */}
+      {!effectiveUseWebCodecs &&
+        segments.map((seg, i) => (
+          <video
+            key={seg.assetId}
+            muted
+            ref={(el) => {
+              videoRefs.current[i] = el
+            }}
+            preload="auto"
+            src={seg.videoUrl}
+            style={{
+              position: "absolute",
+              inset: 0,
+              width: "100%",
+              height: "100%",
+              objectFit: "contain",
+              background: "#000",
+              zIndex: 1,
+              opacity: i === videoCurrentSegIdx ? 1 : 0,
+              pointerEvents: i === videoCurrentSegIdx ? "auto" : "none",
+            }}
+            playsInline
+          />
+        ))}
+
+      {/* 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",
+            borderRadius: "50%",
+            width: 56,
+            height: 56,
+            cursor: "pointer",
+            display: "flex",
+            alignItems: "center",
+            justifyContent: "center",
+            color: "#fff",
+            fontSize: 28,
+            zIndex: 10,
+          }}
+        >
+          <PlayCircleOutlined />
+        </button>
+      )}
+
+      {/* 片段指示器 */}
+      <div
+        style={{
+          position: "absolute",
+          top: 8,
+          left: 8,
+          background: "rgba(0,0,0,0.6)",
+          color: "#fff",
+          fontSize: 11,
+          padding: "2px 8px",
+          borderRadius: 4,
+          zIndex: 10,
+        }}
+      >
+        {`片段 ${videoCurrentSegIdx + 1}/${segments.length}`}
+      </div>
+
+      {/* 控制条 */}
+      <div
+        className="xx-preview-controls"
+        style={{
+          position: "absolute",
+          bottom: 0,
+          left: 0,
+          right: 0,
+          display: "flex",
+          alignItems: "center",
+          gap: 8,
+          padding: "8px 12px",
+          background: "linear-gradient(transparent, rgba(0,0,0,0.6))",
+          zIndex: 10,
+        }}
+      >
+        <button
+          onClick={handleTogglePlay}
+          style={{
+            background: "none",
+            border: "none",
+            color: "#fff",
+            fontSize: 18,
+            cursor: "pointer",
+            padding: 4,
+            display: "flex",
+            alignItems: "center",
+          }}
+        >
+          {isPlaying ? <PauseCircleOutlined /> : <PlayCircleOutlined />}
+        </button>
+
+        <span
+          style={{
+            fontSize: 12,
+            color: "rgba(255,255,255,0.8)",
+            minWidth: 80,
+            fontVariantNumeric: "tabular-nums",
+          }}
+        >
+          {formatTime(currentTime)} / {formatTime(totalDuration)}
+        </span>
+
+        <div
+          ref={progressRef}
+          onMouseDown={handleMouseDown}
+          style={{
+            flex: 1,
+            height: 4,
+            background: "rgba(255,255,255,0.15)",
+            borderRadius: 2,
+            cursor: "pointer",
+            position: "relative",
+          }}
+        >
+          <div
+            style={{
+              height: "100%",
+              width: `${progressPercent}%`,
+              background: "#3b82f6",
+              borderRadius: 2,
+              transition: isDragging ? "none" : "width 0.1s linear",
+            }}
+          />
+          <div
+            style={{
+              position: "absolute",
+              top: "50%",
+              left: `${progressPercent}%`,
+              transform: "translate(-50%, -50%)",
+              width: 10,
+              height: 10,
+              borderRadius: "50%",
+              background: "#3b82f6",
+              border: "2px solid #fff",
+              opacity: isDragging ? 1 : 0,
+              transition: "opacity 0.15s",
+            }}
+          />
+        </div>
+      </div>
+    </>
+  )
+}
+
+export default FrontendPreviewPlayer

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

@@ -2,6 +2,8 @@
  * 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)
+ *
+ * V24: 移除 Step5 预览生成相关 props，改为纯标题样式编辑
  */
 import React from "react"
 import type { EditingTemplate } from "@/api/editing-planner"
@@ -9,7 +11,6 @@ import type { PresetVoiceItem } from "@/api/voices"
 import type { VoiceClone } from "@/api/voice-clone"
 import type { CoverConfig } from "../types/cover"
 import type { TitleSettings } from "../types"
-import type { PreviewItem, PreviewStatus } from "../hooks/useStep5Preview"
 import Step1TemplateSelect from "../components/Step1TemplateSelect"
 import Step2MaterialSelect from "../components/Step2MaterialSelect"
 import Step3VoiceSelect from "../components/Step5VoiceSelect"
@@ -35,6 +36,17 @@ export interface GenerateStepContentProps {
   /* 标题 */
   titleSettings: TitleSettings
   onTitleSettingsChange: (settings: TitleSettings) => void
+  /* 标题样式回调 — Step5 样式面板使用 */
+  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 }[]
   /* 封面 */
   coverSettings: CoverConfig
   onCoverSettingsChange: (settings: CoverConfig) => void
@@ -42,6 +54,7 @@ export interface GenerateStepContentProps {
   /* 配音 */
   selectedVoice: string
   onSelectedVoiceChange: (id: string) => void
+  totalVideoDuration?: number
   voiceMode: "preset" | "custom" | "clone"
   onVoiceModeChange: (mode: "preset" | "custom" | "clone") => void
   selectedClonedVoice: string
@@ -63,19 +76,6 @@ export interface GenerateStepContentProps {
   onDismissError: () => void
   /* 其他 */
   presetVoices: PresetVoiceItem[]
-  videoRatio: string
-  /* Step4 预览（多预览） */
-  previewCount: number
-  onPreviewCountChange: (count: number) => void
-  previewItems: PreviewItem[]
-  previewSelectedIndex: number
-  onSelectPreview: (index: number) => void
-  previewOverallStatus: PreviewStatus
-  previewOverallError: string
-  previewOverallProgress: number
-  previewAnyGenerating: boolean
-  onGeneratePreview: () => void
-  onRegeneratePreview: () => void
 }
 
 export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) => {
@@ -92,11 +92,22 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
     onSmartSelectedIdsChange,
     titleSettings,
     onTitleSettingsChange,
+    onUpdatePosition,
+    onUpdateFont,
+    onUpdateSize,
+    onToggleBold,
+    onToggleItalic,
+    onToggleStroke,
+    onToggleShadow,
+    onApplyPreset,
+    activePreset,
+    titlePresets,
     coverSettings,
     onCoverSettingsChange,
     duration,
     selectedVoice,
     onSelectedVoiceChange,
+    totalVideoDuration,
     voiceMode,
     selectedClonedVoice,
     clonedVoices,
@@ -110,18 +121,6 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
     onRetry,
     onDismissError,
     presetVoices,
-    videoRatio,
-    previewCount,
-    onPreviewCountChange,
-    previewItems,
-    previewSelectedIndex,
-    onSelectPreview,
-    previewOverallStatus,
-    previewOverallError,
-    previewOverallProgress,
-    previewAnyGenerating,
-    onGeneratePreview,
-    onRegeneratePreview,
   } = props
 
   switch (currentStep) {
@@ -149,6 +148,7 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
         <Step3VoiceSelect
           selectedVoice={selectedVoice}
           onSelectedVoiceChange={onSelectedVoiceChange}
+          totalVideoDuration={totalVideoDuration}
         />
       )
     case 4:
@@ -161,18 +161,17 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
     case 5:
       return (
         <Step5GeneratePreview
-          videoRatio={videoRatio}
-          previewCount={previewCount}
-          onPreviewCountChange={onPreviewCountChange}
-          items={previewItems}
-          selectedIndex={previewSelectedIndex}
-          onSelectPreview={onSelectPreview}
-          overallStatus={previewOverallStatus}
-          overallError={previewOverallError}
-          overallProgress={previewOverallProgress}
-          anyGenerating={previewAnyGenerating}
-          onGeneratePreview={onGeneratePreview}
-          onRegeneratePreview={onRegeneratePreview}
+          titleSettings={titleSettings}
+          onUpdatePosition={onUpdatePosition}
+          onUpdateFont={onUpdateFont}
+          onUpdateSize={onUpdateSize}
+          onToggleBold={onToggleBold}
+          onToggleItalic={onToggleItalic}
+          onToggleStroke={onToggleStroke}
+          onToggleShadow={onToggleShadow}
+          onApplyPreset={onApplyPreset}
+          activePreset={activePreset}
+          titlePresets={titlePresets}
         />
       )
     case 6:
@@ -183,6 +182,7 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
           duration={duration}
           assetIds={materialMode === "auto" ? smartSelectedIds : selectedMaterials}
           selectedTemplate={selectedTemplate}
+          titleSettings={titleSettings}
         />
       )
     case 7:

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

@@ -1,293 +1,256 @@
 /**
  * 右侧预览视频面板
- * Step4 生成预览后常驻显示预览视频
- * Step5+ 用 Canvas 绘制标题预览（替代 CSS overlay，与 ASS 渲染行为一致）
+ * 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
  *
- * 设计说明：标题预览仅在有视频时显示（叠加在视频画面上方）。
- * 无视频状态（idle/loading/error）下不再单独显示标题预览，这是有意为之的设计简化。
+ * 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
  *
- * Canvas 居中修复说明：
- * Canvas 的 CSS 位置和尺寸直接匹配视频实际渲染区域（通过 getBoundingClientRect），
- * 绘制坐标系基于 Canvas 自身尺寸，x = w/2 即可实现水平居中，
- * 避免容器与视频尺寸不一致时浏览器拉伸 Canvas 导致居中偏移。
+ * 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
  */
-import React, { useRef, useEffect, useCallback } from "react"
-import { PlayCircleOutlined, LoadingOutlined } from "@ant-design/icons"
-import type { PreviewResult, PreviewStatus } from "../hooks/useStep5Preview"
+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 type { TitleSettings } from "../types"
-import { drawTitleOnCanvas } from "../utils/drawTitleOnCanvas"
-import TitlePreviewCanvas from "./title/TitlePreviewCanvas"
+import { getFontFamily } from "../constants"
+import FrontendPreviewPlayer from "./FrontendPreviewPlayer"
 
 interface PreviewVideoPanelProps {
-  previewStatus: PreviewStatus
-  previewResult: PreviewResult | null
-  previewError: string
-  progress: number
+  /** 已加载的素材列表 */
+  assets: AssetItem[]
+  /** 当前模板 */
+  template: EditingTemplate | null
+  /** 视频比例 */
   videoRatio: string
-  onRegenerate: () => void
-  /** 标题文字 */
-  titleText?: string
-  /** 标题样式设置 */
+  /** 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 */
   titleSettings?: TitleSettings
-  /** Step4 标题预览模式 */
-  showTitlePreview?: boolean
-  /** 素材视频 URL（用于 Step4 标题预览背景） */
-  sourceVideoUrl?: string
+  /** 配音音频 URL */
+  voiceAudioUrl?: string
 }
 
-/* ── 组件 ── */
+/* ── 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
+
+/**
+ * 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
+
+  switch (position) {
+    case "bottom":
+      return {
+        bottom: `${(ASS_TITLE_MARGIN_BOTTOM / ASS_VIDEO_HEIGHT) * 100}%`,
+        left: `${sidePercent}%`,
+        right: `${sidePercent}%`,
+        textAlign: "center",
+      }
+    case "center":
+      return {
+        top: "50%",
+        transform: "translateY(-50%)",
+        left: `${sidePercent}%`,
+        right: `${sidePercent}%`,
+        textAlign: "center",
+      }
+    case "top":
+    default:
+      return {
+        top: `${(ASS_TITLE_MARGIN_TOP / ASS_VIDEO_HEIGHT) * 100}%`,
+        left: `${sidePercent}%`,
+        right: `${sidePercent}%`,
+        textAlign: "center",
+      }
+  }
+}
+
+/**
+ * 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
+  const fontSizePx =
+    containerHeight > 0
+      ? (Math.min(settings.size, 96) / ASS_VIDEO_HEIGHT) * containerHeight
+      : (Math.min(settings.size, 96) / ASS_VIDEO_HEIGHT) * 400 // fallback
+
+  const base: React.CSSProperties = {
+    fontFamily: getFontFamily(settings.font),
+    fontSize: `${fontSizePx}px`,
+    color: settings.color || "#ffffff",
+    fontWeight: settings.bold ? 700 : 400,
+    fontStyle: settings.italic ? "italic" : "normal",
+    lineHeight: 1.3,
+    wordBreak: "break-word",
+    pointerEvents: "none",
+    userSelect: "none",
+    paddingLeft: `${(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)"
+  }
+
+  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 }) => {
+  const containerRef = useRef<HTMLDivElement>(null)
+  const [containerHeight, setContainerHeight] = useState(400) // fallback
+
+  // ResizeObserver 获取容器实际高度
+  useEffect(() => {
+    const el = containerRef.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)
+    // 初始化也读一次
+    const rect = el.getBoundingClientRect()
+    if (rect.height > 0) setContainerHeight(rect.height)
+    return () => ro.disconnect()
+  }, [])
+
+  const positionStyle = useMemo(
+    () => getPositionStyle(titleSettings.position),
+    [titleSettings.position],
+  )
+  const titleStyle = useMemo(
+    () => buildTitleStyle(titleSettings, containerHeight),
+    // eslint-disable-next-line react-hooks/exhaustive-deps -- 已逐字段列出 titleSettings 依赖
+    [
+      containerHeight,
+      titleSettings.font,
+      titleSettings.size,
+      titleSettings.color,
+      titleSettings.bold,
+      titleSettings.italic,
+      titleSettings.stroke,
+      titleSettings.shadow,
+    ],
+  )
+
+  const displayTitle = titleSettings.title?.trim() || "标题预览"
+
+  return (
+    <div
+      ref={containerRef}
+      style={{
+        position: "absolute",
+        inset: 0,
+        zIndex: 20,
+        pointerEvents: "none",
+        overflow: "hidden",
+      }}
+    >
+      <div
+        style={{
+          ...positionStyle,
+          ...titleStyle,
+          position: "absolute",
+        }}
+      >
+        {displayTitle.split("/").map((part, i) => (
+          <span key={i}>
+            {i > 0 && <br />}
+            {part}
+          </span>
+        ))}
+      </div>
+    </div>
+  )
+}
+
+/* ── 主组件 ── */
 
 export const PreviewVideoPanel: React.FC<PreviewVideoPanelProps> = ({
-  previewStatus,
-  previewResult,
-  previewError,
-  progress,
+  assets,
+  template,
   videoRatio,
-  onRegenerate,
-  titleText,
+  assetsReady,
+  assetsLoading,
   titleSettings,
-  showTitlePreview,
-  sourceVideoUrl,
+  voiceAudioUrl,
 }) => {
-  const hasPreview = previewStatus === "ready" && previewResult
-  const isLoading = previewStatus === "pending" || previewStatus === "generating"
-  const isError = previewStatus === "error"
-  const videoAspectStyle = { aspectRatio: (videoRatio || "16:9").replace(":", "/") }
-
-  // video 模式 refs
-  const canvasRef = useRef<HTMLCanvasElement>(null)
-  const containerRef = useRef<HTMLDivElement>(null)
-  const videoRef = useRef<HTMLVideoElement>(null)
-
-  // 字体加载状态（ref 供 draw 回调同步读取，无需 state 避免触发不必要的重渲染）
-  const fontLoadedRef = useRef(false)
-
-  /** 在 video canvas 上绘制标题 */
-  const drawVideoTitle = useCallback(() => {
-    // 通过 ref 读取字体状态，避免 fontLoaded 进入依赖数组
-    if (!fontLoadedRef.current) return
-    const canvas = canvasRef.current
-    const container = containerRef.current
-    if (!canvas || !container || !titleSettings) return
-    const ctx = canvas.getContext("2d")
-    if (!ctx) return
-
-    const containerRect = container.getBoundingClientRect()
-    if (containerRect.width <= 0 || containerRect.height <= 0) return
-
-    // 使用 video 元素的 getBoundingClientRect 获取实际渲染尺寸和位置
-    const videoEl = videoRef.current
-    let drawW = containerRect.width
-    let drawH = containerRect.height
-    let offsetX = 0
-    let offsetY = 0
-
-    if (videoEl && videoEl.clientWidth > 0 && videoEl.clientHeight > 0) {
-      const videoRect = videoEl.getBoundingClientRect()
-      drawW = videoRect.width
-      drawH = videoRect.height
-      offsetX = videoRect.left - containerRect.left
-      offsetY = videoRect.top - containerRect.top
-    }
-
-    // 更新 Canvas CSS 位置和尺寸，使其与视频实际渲染区域完全对齐
-    canvas.style.left = `${offsetX}px`
-    canvas.style.top = `${offsetY}px`
-    canvas.style.width = `${drawW}px`
-    canvas.style.height = `${drawH}px`
-
-    // 绘制时，坐标系基于 Canvas 自身尺寸，无需额外偏移
-    drawTitleOnCanvas(
-      ctx,
-      drawW,
-      drawH,
-      titleText || "",
-      titleSettings,
-      40,
-      titleSettings.position,
-      60,
-    )
-  }, [titleText, titleSettings])
-
-  // video 模式：ResizeObserver 监听容器尺寸变化 → 重绘
-  useEffect(() => {
-    if (!showTitlePreview || !hasPreview) return
-    const container = containerRef.current
-    if (!container) return
-
-    const observer = new ResizeObserver(() => {
-      drawVideoTitle()
-    })
-    observer.observe(container)
-    requestAnimationFrame(drawVideoTitle)
-
-    return () => observer.disconnect()
-  }, [showTitlePreview, hasPreview, drawVideoTitle])
-
-  // 字体加载检测：字体变更时重新检测，确保 measureText 使用正确字体
-  useEffect(() => {
-    if (!showTitlePreview || !titleSettings) {
-      fontLoadedRef.current = false
-      return
-    }
-    let cancelled = false
-    fontLoadedRef.current = false
-
-    const fontWeight = titleSettings.bold ? "bold" : ""
-    const fontStyle = titleSettings.italic ? "italic" : ""
-    const fontSpec =
-      `${fontStyle} ${fontWeight} ${titleSettings.size}px "${titleSettings.font}"`.trim()
-
-    const onFontReady = () => {
-      if (cancelled) return
-      fontLoadedRef.current = true
-      // ref 已同步更新，显式触发重绘（draw 内部通过 ref 检查字体状态）
-      requestAnimationFrame(() => {
-        if (!cancelled) {
-          drawVideoTitle()
-        }
-      })
-    }
-
-    if (document.fonts.check(fontSpec)) {
-      onFontReady()
-      return
-    }
-
-    document.fonts
-      .load(fontSpec)
-      .then(() => onFontReady())
-      .catch(() => {
-        document.fonts.ready.then(() => onFontReady())
-      })
-
-    return () => {
-      cancelled = true
-    }
-  }, [showTitlePreview, titleSettings, drawVideoTitle])
-
-  // video 加载完成后重绘
-  const handleVideoLoaded = useCallback(() => {
-    if (showTitlePreview) {
-      requestAnimationFrame(drawVideoTitle)
-    }
-  }, [showTitlePreview, drawVideoTitle])
+  const videoAspectStyle = { aspectRatio: (videoRatio || "9:16").replace(":", "/") }
 
   return (
     <div className="xx-generate-preview">
       <div className="xx-preview-header">
-        <h3>{showTitlePreview && !hasPreview ? "标题预览" : "预览视频"}</h3>
-        {hasPreview && !showTitlePreview && <span className="xx-preview-badge">480p 预览版</span>}
+        <h3>预览视频</h3>
+        {assetsReady && assets.length > 0 && <span className="xx-preview-badge">实时预览</span>}
       </div>
 
-      {/* Step4 标题预览模式 */}
-      {showTitlePreview && titleSettings && titleText && (
-        <div style={{ padding: "0 16px 16px" }}>
-          <TitlePreviewCanvas
-            titleText={titleText}
-            titleSettings={titleSettings}
-            videoRatio="9:16"
-            sourceVideoUrl={sourceVideoUrl}
-          />
-        </div>
-      )}
-
-      {/* Step4 但无标题时的空状态 */}
-      {showTitlePreview && (!titleText || !titleSettings) && (
-        <div className="xx-preview-empty">
-          <PlayCircleOutlined
-            style={{ fontSize: 48, color: "var(--text-tertiary)", marginBottom: 12 }}
-          />
-          <p className="xx-preview-empty-title">请输入标题</p>
-          <p className="xx-preview-empty-desc">在左侧设置标题后，这里会实时预览效果</p>
-        </div>
-      )}
-
-      {/* 空状态：还没生成预览（非 Step4 模式） */}
-      {!showTitlePreview && previewStatus === "idle" && (
-        <div className="xx-preview-empty">
-          <PlayCircleOutlined
-            style={{ fontSize: 48, color: "var(--text-tertiary)", marginBottom: 12 }}
-          />
-          <p className="xx-preview-empty-title">暂无预览</p>
-          <p className="xx-preview-empty-desc">在左侧生成预览后在此查看</p>
-        </div>
-      )}
-
-      {/* 生成中（非 Step4 模式） */}
-      {!showTitlePreview && isLoading && (
-        <div className="xx-preview-loading-panel">
-          <div className="xx-preview-video" style={videoAspectStyle}>
-            <div className="xx-preview-loading-center">
-              <LoadingOutlined style={{ fontSize: 36, color: "#fff" }} spin />
-              <p style={{ marginTop: 12, color: "rgba(255,255,255,0.8)", fontSize: 14 }}>
-                {previewStatus === "pending" ? "排队中..." : `生成中 ${progress}%`}
-              </p>
-            </div>
+      {/* ✅ 预览容器 — 唯一的 .xx-preview-video 容器
+          内部所有内容（视频、控制条、标题叠加层）通过 absolute 定位填充 */}
+      <div className="xx-preview-video" style={{ ...videoAspectStyle, position: "relative" }}>
+        {/* 加载中状态 */}
+        {assetsLoading && (
+          <div
+            className="xx-preview-loading-center"
+            style={{
+              position: "absolute",
+              inset: 0,
+              display: "flex",
+              flexDirection: "column",
+              alignItems: "center",
+              justifyContent: "center",
+              zIndex: 5,
+            }}
+          >
+            <LoadingOutlined style={{ fontSize: 36, color: "#fff" }} spin />
+            <p style={{ marginTop: 12, color: "rgba(255,255,255,0.8)", fontSize: 14 }}>
+              加载素材中...
+            </p>
           </div>
-          <div className="xx-preview-progress-bar-wrap">
-            <div className="xx-preview-progress-fill" style={{ width: `${progress}%` }} />
-          </div>
-        </div>
-      )}
+        )}
 
-      {/* 生成失败（非 Step4 模式） */}
-      {!showTitlePreview && isError && (
-        <div className="xx-preview-error-panel">
-          <div className="xx-preview-video xx-preview-video--error" style={videoAspectStyle}>
-            <p style={{ color: "rgba(255,255,255,0.8)", fontSize: 14 }}>预览生成失败</p>
-          </div>
-          <p className="xx-preview-error-msg">
-            {typeof previewError === "string" && previewError ? previewError : "请重试"}
-          </p>
-          <button className="xx-btn xx-btn-ghost xx-btn-block" onClick={onRegenerate}>
-            重新生成
-          </button>
-        </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}
+        />
 
-      {/* 预览成功 + Canvas 标题叠加 */}
-      {hasPreview && (
-        <div ref={containerRef} style={{ position: "relative" }}>
-          <div className="xx-preview-video" style={videoAspectStyle}>
-            <video
-              ref={videoRef}
-              src={previewResult.videoUrl}
-              controls
-              preload="metadata"
-              onLoadedMetadata={handleVideoLoaded}
-            />
-          </div>
-          {showTitlePreview && (
-            <canvas
-              ref={canvasRef}
-              style={{
-                position: "absolute",
-                top: 0,
-                left: 0,
-                width: "100%",
-                height: "100%",
-                zIndex: 1,
-                pointerEvents: "none",
-              }}
-            />
-          )}
-        </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} />}
+      </div>
 
-      {/* 预览信息（非 Step4 模式） */}
-      {!showTitlePreview && hasPreview && previewResult && (
+      {/* 素材信息 */}
+      {assetsReady && assets.length > 0 && (
         <div className="xx-preview-info">
           <div className="xx-preview-info-row">
-            <span>时长</span>
-            <span>
-              {(typeof previewResult.duration === "number" ? previewResult.duration : 0).toFixed(1)}{" "}
-              秒
-            </span>
-          </div>
-          <div className="xx-preview-info-row">
-            <span>片段数</span>
-            <span>{previewResult.clipCount} 段</span>
+            <span>素材数</span>
+            <span>{assets.length} 个</span>
           </div>
           <div className="xx-preview-info-row">
             <span>比例</span>

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

@@ -1,13 +1,13 @@
 /**
  * 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）
  */
 import React from "react"
 import { AutoComplete } from "antd"
-import { POSITION_OPTIONS, FONT_OPTIONS } from "../constants"
 import type { TitleSettings } from "../types"
 import { useStep4Title } from "../hooks/useStep4Title"
 import AiTitleGenerator from "./title/AiTitleGenerator"
-import TitleStylePanel from "./title/TitleStylePanel"
 
 interface Step4TitleSettingsProps {
   titleSettings: TitleSettings
@@ -110,23 +110,6 @@ const Step4TitleSettings: React.FC<Step4TitleSettingsProps> = (props) => {
           </div>
         </>
       )}
-
-      {/* 样式面板始终可见，两种模式下都可调整标题展示样式 */}
-      <TitleStylePanel
-        settings={t.titleSettings}
-        onUpdatePosition={t.updatePosition}
-        onUpdateFont={t.updateFont}
-        onUpdateSize={t.updateSize}
-        onToggleBold={t.toggleBold}
-        onToggleItalic={t.toggleItalic}
-        onToggleStroke={t.toggleStroke}
-        onToggleShadow={t.toggleShadow}
-        onApplyPreset={t.applyPreset}
-        activePreset={t.activePreset}
-        titlePresets={t.titlePresets}
-        POSITION_OPTIONS={POSITION_OPTIONS}
-        FONT_OPTIONS={FONT_OPTIONS}
-      />
     </div>
   )
 }

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

@@ -1,274 +1,82 @@
 /**
- * Step 5 生成预览组件（支持多预览）
- * 调用后端预览生成接口，展示多个真实视频预览（网格布局）
+ * 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 {
-  CheckCircleFilled,
-  LoadingOutlined,
-  ReloadOutlined,
-  PlayCircleOutlined,
-  ExclamationCircleFilled,
-  ClockCircleOutlined,
-} from "@ant-design/icons"
-import { InputNumber } from "antd"
-import type { PreviewItem, PreviewStatus } from "../hooks/useStep5Preview"
+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 {
-  videoRatio: string
-  previewCount: number
-  onPreviewCountChange: (count: number) => void
-  items: PreviewItem[]
-  selectedIndex: number
-  onSelectPreview: (index: number) => void
-  overallStatus: PreviewStatus
-  overallError: string
-  overallProgress: number
-  anyGenerating: boolean
-  onGeneratePreview: () => void
-  onRegeneratePreview: () => void
+  /* 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 }[]
 }
 
-/** 预览数量选项 */
-const PREVIEW_COUNT_OPTIONS = [
-  { value: 1, label: "1个" },
-  { value: 2, label: "2个" },
-  { value: 3, label: "3个" },
-]
-
 const Step5GeneratePreview: React.FC<Step5GeneratePreviewProps> = ({
-  videoRatio,
-  previewCount,
-  onPreviewCountChange,
-  items,
-  selectedIndex,
-  onSelectPreview,
-  overallStatus,
-  overallError,
-  overallProgress,
-  anyGenerating,
-  onGeneratePreview,
-  onRegeneratePreview,
+  titleSettings,
+  onUpdatePosition,
+  onUpdateFont,
+  onUpdateSize,
+  onToggleBold,
+  onToggleItalic,
+  onToggleStroke,
+  onToggleShadow,
+  onApplyPreset,
+  activePreset,
+  titlePresets,
 }) => {
-  const aspectRatio = (videoRatio || "16:9").replace(":", "/") // "9:16" → "9/16", "16:9" → "16/9"
-  const isIdle = overallStatus === "idle"
-  const isError = overallStatus === "error" && !items.some((it) => it.status === "ready")
-
   return (
     <div className="xx-form-section">
-      <h3>🎬 生成预览</h3>
+      <h3>🎬 预览设置</h3>
 
-      {/* 预览数量选择器（仅在 idle 状态显示） */}
-      {isIdle && (
-        <div style={{ marginBottom: 16, display: "flex", alignItems: "center", gap: 12 }}>
-          <span style={{ fontSize: 14, color: "#666" }}>预览数量：</span>
-          <div style={{ display: "flex", gap: 8, alignItems: "center" }}>
-            {PREVIEW_COUNT_OPTIONS.map((opt) => (
-              <button
-                key={opt.value}
-                type="button"
-                onClick={() => onPreviewCountChange(opt.value)}
-                style={{
-                  padding: "4px 12px",
-                  borderRadius: 6,
-                  border: previewCount === opt.value ? "1px solid #1677ff" : "1px solid #d9d9d9",
-                  background: previewCount === opt.value ? "#e6f4ff" : "#fff",
-                  color: previewCount === opt.value ? "#1677ff" : "#666",
-                  cursor: "pointer",
-                  fontSize: 13,
-                  fontWeight: previewCount === opt.value ? 600 : 400,
-                }}
-              >
-                {opt.label}
-              </button>
-            ))}
-            <InputNumber
-              min={1}
-              max={10}
-              value={previewCount}
-              onChange={(val) => val && onPreviewCountChange(val)}
-              style={{ width: 70 }}
-              placeholder="自定义"
-            />
-            <span style={{ fontSize: 12, color: "#999", marginLeft: 4 }}>（1~10）</span>
-          </div>
-          {previewCount > 1 && (
-            <span style={{ fontSize: 12, color: "#999" }}>生成多个预览可对比不同剪辑效果</span>
-          )}
-        </div>
-      )}
+      {/* 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>
 
-      {/* 预览生成按钮（idle 状态） */}
-      {isIdle && (
-        <div className="xx-preview-generate-section">
-          <div className="xx-preview-generate-hint">
-            <PlayCircleOutlined style={{ fontSize: 32, color: "#3b82f6", marginBottom: 12 }} />
-            <p className="xx-preview-generate-title">一键生成剪辑预览</p>
-            <p className="xx-preview-generate-desc">
-              AI 将根据您选择的模板、素材和配音，智能生成
-              {previewCount > 1 ? `${previewCount}个不同版本的` : ""}视频预览（480p 低清版）
-            </p>
-          </div>
-          <button
-            className="xx-btn xx-btn-primary xx-preview-generate-btn"
-            onClick={onGeneratePreview}
-          >
-            ✨ 生成预览{previewCount > 1 ? `（${previewCount}个）` : ""}
-          </button>
-        </div>
-      )}
-
-      {/* 整体排队中（所有都在 pending） */}
-      {anyGenerating && items.every((it) => it.status === "pending") && (
-        <div className="xx-preview-loading">
-          <ClockCircleOutlined style={{ fontSize: 32, color: "#faad14" }} spin />
-          <p className="xx-preview-loading-text">预览排队中...</p>
-          <p className="xx-preview-loading-desc">正在等待渲染资源，请稍候</p>
-        </div>
-      )}
-
-      {/* 多预览网格（生成中/完成/部分完成） */}
-      {(anyGenerating || overallStatus === "ready") && items.length > 0 && (
-        <div
-          className="xx-preview-grid"
-          style={{
-            display: "grid",
-            gridTemplateColumns: `repeat(${Math.min(items.length, 3)}, 1fr)`,
-            gap: 12,
-            maxWidth: `${Math.min(items.length, 3) * 280 + (Math.min(items.length, 3) - 1) * 12}px`,
-            margin: "0 auto 16px",
-          }}
-        >
-          {items.map((item) => {
-            const isSelected = item.index === selectedIndex
-            return (
-              <div
-                key={item.index}
-                onClick={() => {
-                  if (item.status === "ready") onSelectPreview(item.index)
-                }}
-                style={{
-                  borderRadius: 8,
-                  border: isSelected ? "2px solid #1677ff" : "1px solid #e8e8e8",
-                  overflow: "hidden",
-                  cursor: item.status === "ready" ? "pointer" : "default",
-                  opacity: item.status === "error" ? 0.6 : 1,
-                  transition: "all 0.2s",
-                }}
-              >
-                {/* 轻量卡片：深色背景 + 状态指示 */}
-                <div
-                  style={{
-                    aspectRatio,
-                    background: "#1a1a2e",
-                    display: "flex",
-                    flexDirection: "column",
-                    alignItems: "center",
-                    justifyContent: "center",
-                    position: "relative",
-                    gap: 8,
-                  }}
-                >
-                  {/* 中心：预览编号 */}
-                  <span
-                    style={{
-                      fontSize: 24,
-                      fontWeight: 700,
-                      color: "#fff",
-                      opacity: 0.9,
-                    }}
-                  >
-                    预览 #{item.index + 1}
-                  </span>
-
-                  {/* 状态指示 */}
-                  {item.status === "generating" && (
-                    <div style={{ display: "flex", alignItems: "center", gap: 6 }}>
-                      <LoadingOutlined style={{ fontSize: 14, color: "#fff" }} spin />
-                      <span style={{ color: "rgba(255,255,255,0.8)", fontSize: 12 }}>
-                        生成中 {item.progress}%
-                      </span>
-                    </div>
-                  )}
-                  {item.status === "pending" && (
-                    <div style={{ display: "flex", alignItems: "center", gap: 6 }}>
-                      <ClockCircleOutlined
-                        style={{ fontSize: 14, color: "rgba(255,255,255,0.6)" }}
-                      />
-                      <span style={{ color: "rgba(255,255,255,0.6)", fontSize: 12 }}>
-                        排队中...
-                      </span>
-                    </div>
-                  )}
-                  {item.status === "ready" && (
-                    <CheckCircleFilled style={{ fontSize: 18, color: "#52c41a" }} />
-                  )}
-                  {item.status === "error" && (
-                    <ExclamationCircleFilled style={{ fontSize: 18, color: "#ef4444" }} />
-                  )}
-
-                  {/* 选中角标 */}
-                  {isSelected && item.status === "ready" && (
-                    <div
-                      style={{
-                        position: "absolute",
-                        top: 4,
-                        right: 4,
-                        background: "#1677ff",
-                        color: "#fff",
-                        fontSize: 10,
-                        padding: "2px 6px",
-                        borderRadius: 4,
-                      }}
-                    >
-                      预览 #{item.index + 1}
-                    </div>
-                  )}
-                </div>
-              </div>
-            )
-          })}
-        </div>
-      )}
-
-      {/* 整体进度条（多预览生成中） */}
-      {anyGenerating && (
-        <div className="xx-preview-progress-bar" style={{ marginBottom: 12 }}>
-          <div className="xx-preview-progress-fill" style={{ width: `${overallProgress}%` }} />
-        </div>
-      )}
-
-      {/* 全部完成提示 */}
-      {overallStatus === "ready" && (
-        <div className="xx-preview-tip">
-          <CheckCircleFilled style={{ color: "#52c41a", marginRight: 8 }} />
-          <span>
-            {items.filter((it) => it.status === "ready").length} 个预览生成成功
-            {items.length > 1 ? "，点击选择要查看的版本" : "，确认效果后进入下一步"}
-          </span>
-          <button
-            className="xx-preview-regenerate-btn"
-            onClick={onRegeneratePreview}
-            title="重新生成"
-          >
-            <ReloadOutlined /> 重新生成
-          </button>
-        </div>
-      )}
-
-      {/* 全部失败 */}
-      {isError && (
-        <div className="xx-preview-error">
-          <ExclamationCircleFilled style={{ fontSize: 28, color: "#ef4444" }} />
-          <p className="xx-preview-error-text">预览生成失败</p>
-          <p className="xx-preview-error-desc">
-            {typeof overallError === "string" && overallError ? overallError : "请稍后重试"}
-          </p>
-          <button className="xx-btn xx-btn-primary" onClick={onRegeneratePreview}>
-            <ReloadOutlined /> 重新生成
-          </button>
-        </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>
   )
 }

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

@@ -5,16 +5,28 @@
 import React, { useState, useRef, useCallback } from "react"
 import { useNavigate } from "react-router-dom"
 import { useQuery } from "@tanstack/react-query"
-import { AudioOutlined, SoundOutlined } from "@ant-design/icons"
+import { AudioOutlined, SoundOutlined, WarningOutlined } from "@ant-design/icons"
+import { Modal } from "antd"
 import { getAssetsByKind } from "@/api/assets"
 import type { AssetItem } from "@/api/assets"
 
 interface Step5VoiceSelectProps {
   selectedVoice: string
   onSelectedVoiceChange: (id: string) => void
+  totalVideoDuration?: number
 }
 
 /** 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)
@@ -34,10 +46,13 @@ const formatFileSize = (bytes?: number): string => {
 const Step5VoiceSelect: React.FC<Step5VoiceSelectProps> = ({
   selectedVoice,
   onSelectedVoiceChange,
+  totalVideoDuration = 0,
 }) => {
   const navigate = useNavigate()
   const [playingId, setPlayingId] = useState<string | null>(null)
   const audioRef = useRef<HTMLAudioElement | null>(null)
+  const [durationWarningOpen, setDurationWarningOpen] = useState(false)
+  const [pendingVoiceId, setPendingVoiceId] = useState<string | null>(null)
 
   // 获取用户上传的配音素材
   const { data: materials = [], isLoading } = useQuery({
@@ -78,14 +93,38 @@ const Step5VoiceSelect: React.FC<Step5VoiceSelectProps> = ({
     [playingId],
   )
 
-  /** 选中素材 */
+  /** 选中素材（含时长校验） */
   const handleSelect = useCallback(
     (id: string) => {
+      // 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 && getDuration(material) < totalVideoDuration) {
+          setPendingVoiceId(id)
+          setDurationWarningOpen(true)
+          return
+        }
+      }
       onSelectedVoiceChange(id)
     },
-    [onSelectedVoiceChange],
+    [onSelectedVoiceChange, totalVideoDuration, materials],
   )
 
+  /** 确认使用时长不足的配音 */
+  const handleConfirmUseAnyway = useCallback(() => {
+    if (pendingVoiceId) {
+      onSelectedVoiceChange(pendingVoiceId)
+    }
+    setDurationWarningOpen(false)
+    setPendingVoiceId(null)
+  }, [pendingVoiceId, onSelectedVoiceChange])
+
+  /** 取消选择 */
+  const handleCancelSelection = useCallback(() => {
+    setDurationWarningOpen(false)
+    setPendingVoiceId(null)
+  }, [])
+
   /** 跳转到配音库上传 */
   const handleGoToUpload = useCallback(() => {
     navigate("/app/voices")
@@ -238,15 +277,66 @@ const Step5VoiceSelect: React.FC<Step5VoiceSelectProps> = ({
                   justifyContent: "space-between",
                   fontSize: 12,
                   color: "#999",
+                  alignItems: "center",
                 }}
               >
-                <span>{formatDuration(item.duration)}</span>
-                <span>{formatFileSize(item.file_size)}</span>
+                <span style={{ display: "flex", alignItems: "center", gap: 4 }}>
+                  {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(getFileSize(item))}</span>
               </div>
             </div>
           )
         })}
       </div>
+
+      {/* 时长不足警告弹窗 */}
+      <Modal
+        title={
+          <span style={{ display: "flex", alignItems: "center", gap: 8 }}>
+            <WarningOutlined style={{ color: "#faad14" }} />
+            配音时长不足
+          </span>
+        }
+        open={durationWarningOpen}
+        onOk={handleConfirmUseAnyway}
+        onCancel={handleCancelSelection}
+        okText="仍要使用"
+        cancelText="重新选择"
+        okButtonProps={{ danger: true }}
+      >
+        {(() => {
+          const pendingMaterial = pendingVoiceId
+            ? materials.find((m) => m.id === pendingVoiceId)
+            : null
+          return (
+            <p>
+              该配音时长（
+              <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。
+            </p>
+          )
+        })()}
+      </Modal>
     </div>
   )
 }

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

@@ -14,6 +14,8 @@ 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
 }
 
 const Step6CoverSettings: React.FC<Step6CoverSettingsProps> = (props) => {
@@ -40,6 +42,7 @@ const Step6CoverSettings: React.FC<Step6CoverSettingsProps> = (props) => {
     duration: props.duration,
     assetIds: props.assetIds,
     selectedTemplate: props.selectedTemplate,
+    titleSettings: props.titleSettings,
   })
 
   const handleAutoGenerate = () => {

apps/web/src/pages/generate/components/cover-settings/UploadCoverPicker.tsx

@@ -30,7 +30,7 @@ export const UploadCoverPicker: React.FC<UploadCoverPickerProps> = ({ uploadUrl,
           <div className="xx-cover-upload-placeholder">
             <span style={{ fontSize: 32 }}>📤</span>
             <span className="xx-cover-upload-text">点击上传封面图片</span>
-            <span className="xx-cover-upload-hint">支持 JPG / PNG，建议 16:9 比例</span>
+            <span className="xx-cover-upload-hint">支持 JPG / PNG，建议 9:16 比例</span>
           </div>
         )}
         <input

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/TitlePreviewCanvas.tsx

@@ -1,192 +0,0 @@
-/**
- * 标题实时预览 Canvas 组件
- *
- * 在 Step4 标题设置面板中嵌入，让用户实时看到标题文字、字体、大小、颜色、
- * 位置、描边、阴影等样式的实际渲染效果（所见即所得）。
- *
- * 使用共享的 drawTitleOnCanvas 工具函数，与 PreviewVideoPanel 行为一致。
- */
-import React, { useRef, useEffect } from "react"
-import type { TitleSettings } from "../../types"
-import { drawTitleOnCanvas } from "../../utils/drawTitleOnCanvas"
-
-interface TitlePreviewCanvasProps {
-  /** 标题文字 */
-  titleText: string
-  /** 标题样式设置 */
-  titleSettings: TitleSettings
-  /** 视频比例，默认 "9:16"（竖屏） */
-  videoRatio?: string
-  /** 素材视频 URL（作为背景显示） */
-  sourceVideoUrl?: string
-}
-
-/**
- * 解析 videoRatio 字符串为 aspect-ratio CSS 值
- */
-function parseAspect(ratio: string): string {
-  return (ratio || "9:16").replace(":", "/")
-}
-
-const TitlePreviewCanvas: React.FC<TitlePreviewCanvasProps> = ({
-  titleText,
-  titleSettings,
-  videoRatio = "9:16",
-  sourceVideoUrl,
-}) => {
-  const containerRef = useRef<HTMLDivElement>(null)
-  const canvasRef = useRef<HTMLCanvasElement>(null)
-
-  // 字体加载状态
-  const fontLoadedRef = useRef(false)
-
-  /** 在 Canvas 上绘制标题 */
-  const draw = () => {
-    const canvas = canvasRef.current
-    const container = containerRef.current
-    if (!canvas || !container) return
-    const ctx = canvas.getContext("2d")
-    if (!ctx) return
-
-    const rect = container.getBoundingClientRect()
-    if (rect.width <= 0 || rect.height <= 0) return
-
-    const w = rect.width
-    const h = rect.height
-
-    // 更新 Canvas CSS 尺寸匹配容器
-    canvas.style.width = `${w}px`
-    canvas.style.height = `${h}px`
-
-    drawTitleOnCanvas(ctx, w, h, titleText, titleSettings, 24, titleSettings.position, 40)
-  }
-
-  // 字体加载：确保 measureText 使用正确字体
-  useEffect(() => {
-    let cancelled = false
-    fontLoadedRef.current = false
-
-    const fontWeight = titleSettings.bold ? "bold" : ""
-    const fontStyle = titleSettings.italic ? "italic" : ""
-    const fontSpec =
-      `${fontStyle} ${fontWeight} ${titleSettings.size}px "${titleSettings.font}"`.trim()
-
-    const onFontReady = () => {
-      if (cancelled) return
-      fontLoadedRef.current = true
-      requestAnimationFrame(() => {
-        if (!cancelled) draw()
-      })
-    }
-
-    // 用 FontFace API 加载字体，失败则降级
-    try {
-      const fontFace = new FontFace(titleSettings.font, `local("${titleSettings.font}")`)
-      fontFace
-        .load()
-        .then(() => {
-          if (!cancelled) {
-            // eslint-disable-next-line @typescript-eslint/no-explicit-any -- FontFaceSet.add() exists at runtime
-            ;(document.fonts as any).add(fontFace)
-            onFontReady()
-          }
-        })
-        .catch(() => {
-          // 字体加载失败，用默认字体继续
-          onFontReady()
-        })
-    } catch {
-      // FontFace 不可用，直接绘制
-      onFontReady()
-    }
-
-    // 同时检查 document.fonts 是否已有该字体
-    if (document.fonts.check(fontSpec)) {
-      onFontReady()
-      return
-    }
-
-    return () => {
-      cancelled = true
-    }
-    // eslint-disable-next-line react-hooks/exhaustive-deps -- draw uses refs, stable by design
-  }, [titleSettings.font, titleSettings.size, titleSettings.bold, titleSettings.italic])
-
-  // props 变化时重绘
-  useEffect(() => {
-    requestAnimationFrame(draw)
-    // eslint-disable-next-line react-hooks/exhaustive-deps -- draw uses refs, stable by design
-  }, [titleText, titleSettings])
-
-  // ResizeObserver 监听容器尺寸变化
-  useEffect(() => {
-    const container = containerRef.current
-    if (!container) return
-
-    const observer = new ResizeObserver(() => {
-      requestAnimationFrame(draw)
-    })
-    observer.observe(container)
-
-    return () => observer.disconnect()
-    // eslint-disable-next-line react-hooks/exhaustive-deps -- draw uses refs, stable by design
-  }, [])
-
-  return (
-    <div>
-      <div
-        style={{
-          fontSize: 12,
-          color: "var(--text-tertiary, #999)",
-          marginBottom: 6,
-          fontWeight: 500,
-        }}
-      >
-        预览效果
-      </div>
-      <div
-        ref={containerRef}
-        style={{
-          width: "100%",
-          aspectRatio: parseAspect(videoRatio),
-          background: sourceVideoUrl
-            ? "#000"
-            : "linear-gradient(135deg, #1a1a2e, #16213e, #0f3460)",
-          borderRadius: 8,
-          overflow: "hidden",
-          position: "relative",
-        }}
-      >
-        {sourceVideoUrl && (
-          <video
-            src={sourceVideoUrl}
-            muted
-            loop
-            autoPlay
-            playsInline
-            style={{
-              position: "absolute",
-              top: 0,
-              left: 0,
-              width: "100%",
-              height: "100%",
-              objectFit: "cover",
-            }}
-          />
-        )}
-        <canvas
-          ref={canvasRef}
-          style={{
-            position: "absolute",
-            top: 0,
-            left: 0,
-            width: "100%",
-            height: "100%",
-          }}
-        />
-      </div>
-    </div>
-  )
-}
-
-export default TitlePreviewCanvas

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

@@ -56,6 +56,21 @@ 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

@@ -891,17 +891,16 @@
 
 /* ── 视频预览 ── */
 .xx-preview-video {
-  aspect-ratio: 16 / 9;
+  aspect-ratio: 9 / 16;
   max-height: 400px;
   border-radius: var(--radius-md);
   background: linear-gradient(135deg, var(--color-gray-900), var(--color-primary-900));
-  display: grid;
-  place-items: center;
   color: var(--text-inverse);
   font-size: 36px;
   position: relative;
   overflow: hidden;
   margin-bottom: 16px;
+  isolation: isolate;
 }
 
 .xx-preview-video::before {
@@ -910,6 +909,7 @@
   inset: 0;
   background: radial-gradient(circle at 72% 28%, rgba(255, 255, 255, 0.2), transparent 40%);
   pointer-events: none;
+  z-index: 0;
 }
 
 .xx-preview-video video {
@@ -1516,7 +1516,7 @@
 .xx-smart-match-thumb {
   position: relative;
   width: 100%;
-  aspect-ratio: 16 / 9;
+  aspect-ratio: 9 / 16;
   background: #f1f5f9;
   overflow: hidden;
 }
@@ -2144,7 +2144,7 @@
 }
 
 .xx-cover-frame-placeholder {
-  aspect-ratio: 16 / 9;
+  aspect-ratio: 9 / 16;
   background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
   border-radius: var(--radius-md);
   display: flex;
@@ -2247,7 +2247,7 @@
 }
 
 .xx-cover-upload-area {
-  aspect-ratio: 16 / 9;
+  aspect-ratio: 9 / 16;
   border: 2px dashed var(--border-color);
   border-radius: var(--radius-md);
   display: flex;
@@ -2308,7 +2308,9 @@
 
 .xx-cover-preview-box {
   position: relative;
-  aspect-ratio: 16 / 9;
+  aspect-ratio: 9 / 16;
+  max-width: 180px;
+  margin: 0 auto;
   background: var(--bg-tertiary);
   border-radius: var(--radius-md);
   overflow: hidden;
@@ -2666,7 +2668,7 @@
 .xx-preview-video-wrapper .xx-preview-video {
   max-width: 300px;
   width: 100%;
-  aspect-ratio: 16 / 9;
+  aspect-ratio: 9 / 16;
   margin-bottom: 0;
 }
 
@@ -2772,7 +2774,7 @@
 
 .xx-video-thumb {
   position: relative;
-  aspect-ratio: 16 / 9;
+  aspect-ratio: 9 / 16;
   background: var(--bg-tertiary);
   overflow: hidden;
 }
@@ -2881,11 +2883,11 @@
   overflow: hidden;
 }
 
-.ant-modal-content {
+.xx-preview-modal .ant-modal-content {
   padding: 0 !important;
 }
 
-.ant-modal-close {
+.xx-preview-modal .ant-modal-close {
   color: #fff !important;
 }
 

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

@@ -19,8 +19,6 @@ export interface UseGenerateVideoProps {
   autoSubtitles: boolean
   bgm: boolean
   generateCount: number
-  /** 预览任务的 task_id（用于新确认生成 API） */
-  previewTaskId: string
 }
 
 /** 生成阶段 */

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

@@ -5,7 +5,7 @@
 import { useState, useCallback } from "react"
 import { message } from "antd"
 import type { GeneratedVideo } from "@/api/template-editor"
-import { confirmGeneration } from "@/api/generation"
+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"
@@ -55,21 +55,17 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
     clearTimer()
 
     try {
-      // 使用确认生成 API（基于预览任务）
-      // 解析分辨率：videoRatio 可能是 "9:16"（宽高比）或 "1080x1920"（分辨率）
+      // 解析分辨率
       const ratio = props.videoRatio || "9:16"
       let outputWidth: number
       let outputHeight: number
 
       if (ratio.includes(":")) {
-        // 宽高比格式，如 "9:16" → 基于基准高度 1920 计算
         const [rw, rh] = ratio.split(":").map(Number)
         if (rw > 0 && rh > 0) {
-          // 基准：长边 1920，短边按比例计算
           const [longSide, shortSide] = rw < rh ? [rh, rw] : [rw, rh]
           const baseLong = 1920
           const baseShort = Math.round((baseLong * shortSide) / longSide)
-          // 确保偶数（FFmpeg 要求）
           const evenShort = baseShort - (baseShort % 2)
           if (rw < rh) {
             outputWidth = evenShort
@@ -83,7 +79,6 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
           outputHeight = 1920
         }
       } else if (ratio.includes("x")) {
-        // 分辨率格式，如 "1080x1920"
         const [wStr, hStr] = ratio.split("x")
         outputWidth = parseInt(wStr, 10) || 1080
         outputHeight = parseInt(hStr, 10) || 1920
@@ -92,11 +87,33 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
         outputHeight = 1920
       }
 
-      await confirmGeneration(props.previewTaskId, {
+      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 || "",
+        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,
+              },
+            }
+          : {}),
       })
 
       startPolling()
@@ -109,7 +126,7 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
       setGenerateError(finalMsg)
       message.error(finalMsg)
     }
-  }, [props, clearTimer, startPolling])
+  }, [props, clearTimer, startPolling, selectedTemplate])
 
   /* 重新生成（失败后重试） */
   const retry = useCallback(() => {

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

@@ -0,0 +1,114 @@
+/**
+ * 预览素材加载 Hook
+ * 根据选中的素材 ID 列表，逐个获取素材详情（含 file_url、duration 等）
+ * 供前端预览播放器使用
+ *
+ * 注意：后端没有批量接口（/assets/batch 返回 405），
+ * 因此直接使用 Promise.allSettled 并发请求单个 GET /assets/{id}
+ */
+import { useState, useEffect, useCallback, useRef } from "react"
+import type { AssetItem } from "@/api/assets"
+import type { AxiosResponse } from "axios"
+
+/**
+ * 通过 ID 列表逐个获取素材（并发）
+ * 使用 Promise.allSettled 确保单个失败不影响整体
+ */
+async function fetchAssetsByIds(ids: string[]): Promise<AssetItem[]> {
+  if (!ids.length) return []
+
+  try {
+    const { default: apiClient } = await import("@/api/client")
+    const results = await Promise.allSettled(
+      ids.map((id) => apiClient.get<AssetItem>(`/assets/${id}`)),
+    )
+    return results
+      .filter(
+        (r): r is PromiseFulfilledResult<AxiosResponse<AssetItem>> =>
+          r.status === "fulfilled" && !!r.value?.data,
+      )
+      .map((r) => r.value.data)
+  } catch {
+    return []
+  }
+}
+
+interface UsePreviewAssetsReturn {
+  /** 加载后的素材列表 */
+  assets: AssetItem[]
+  /** 是否正在加载 */
+  loading: boolean
+  /** 是否已就绪（加载完成） */
+  ready: boolean
+  /** 手动触发重新加载 */
+  reload: () => void
+}
+
+/**
+ * usePreviewAssets — 加载选中素材的视频文件信息
+ */
+export function usePreviewAssets(assetIds: string[], enabled: boolean): UsePreviewAssetsReturn {
+  const [assets, setAssets] = useState<AssetItem[]>([])
+  const [loading, setLoading] = useState(false)
+  const [ready, setReady] = useState(false)
+  const requestIdRef = useRef(0)
+
+  // 稳定化 assetIds：只有内容真正变化时才更新引用
+  const stableAssetIds = useStableArray(assetIds)
+
+  const load = useCallback(async () => {
+    if (!stableAssetIds.length || !enabled) {
+      setAssets([])
+      setReady(false)
+      return
+    }
+
+    const thisRequestId = ++requestIdRef.current
+    setLoading(true)
+    setReady(false)
+
+    try {
+      const result = await fetchAssetsByIds(stableAssetIds)
+      // 防止竞态：只保留最新请求的结果
+      if (requestIdRef.current === thisRequestId) {
+        setAssets(result)
+        setReady(result.length > 0)
+      }
+    } catch {
+      if (requestIdRef.current === thisRequestId) {
+        setAssets([])
+        setReady(false)
+      }
+    } finally {
+      if (requestIdRef.current === thisRequestId) {
+        setLoading(false)
+      }
+    }
+  }, [stableAssetIds, enabled])
+
+  useEffect(() => {
+    load()
+  }, [load])
+
+  return { assets, loading, ready, reload: load }
+}
+
+/**
+ * useStableArray — 数组内容稳定化 Hook
+ * 只有数组内容真正变化时才返回新的引用，避免父组件 re-render 导致的无效更新
+ */
+function useStableArray<T>(array: T[]): T[] {
+  const ref = useRef<T[]>(array)
+
+  // 比较数组内容是否真正变化
+  const hasChanged =
+    array.length !== ref.current.length || array.some((item, index) => item !== ref.current[index])
+
+  if (hasChanged) {
+    ref.current = array
+  }
+
+  return ref.current
+}
+
+export default usePreviewAssets

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

@@ -0,0 +1,365 @@
+/**
+ * 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"
+
+export interface PlaybackSegment {
+  assetId: string
+  videoUrl: string
+  startTime: number
+  endTime: number
+  order: number
+}
+
+export interface SegmentSchedulerState {
+  isPlaying: boolean
+  currentTime: number
+  totalDuration: number
+  currentSegmentIndex: number
+  segmentLocalTime: number
+  isEnded: boolean
+  canPlay: boolean
+  play: () => void
+  pause: () => void
+  togglePlayPause: () => void
+  seekTo: (time: number) => void
+  videoRefs: React.MutableRefObject<(HTMLVideoElement | null)[]>
+}
+
+function findSegmentAtTime(
+  segments: PlaybackSegment[],
+  globalTime: number,
+): { index: number; localTime: number } {
+  let accumulated = 0
+  for (let i = 0; i < segments.length; i++) {
+    const seg = segments[i]
+    const segDuration = seg.endTime - seg.startTime
+    if (globalTime < accumulated + segDuration || i === segments.length - 1) {
+      return { index: i, localTime: seg.startTime + (globalTime - accumulated) }
+    }
+    accumulated += segDuration
+  }
+  return { index: segments.length - 1, localTime: segments[segments.length - 1].endTime }
+}
+
+function buildTimeline(segments: PlaybackSegment[]): number[] {
+  const starts: number[] = []
+  let acc = 0
+  for (const seg of segments) {
+    starts.push(acc)
+    acc += seg.endTime - seg.startTime
+  }
+  return starts
+}
+
+export function useSegmentScheduler(segments: PlaybackSegment[]): SegmentSchedulerState {
+  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(0)
+  const isSeekingRef = useRef(false)
+  const lastTimeUpdateRef = useRef(0)
+
+  // 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
+
+  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)
+    })
+  }, [])
+
+  const switchToSegment = useCallback(
+    async (index: number, seekToLocalTime?: number) => {
+      const segs = segmentsRef.current
+      const video = videoRefs.current[index]
+      if (!video || index >= segs.length) return
+
+      const seg = segs[index]
+      const localTime = seekToLocalTime ?? seg.startTime
+      const oldIdx = segIdxRef.current
+      const oldVideo = videoRefs.current[oldIdx]
+
+      if (oldVideo && oldVideo !== video) oldVideo.pause()
+
+      if (!video.src && seg.videoUrl) {
+        video.src = seg.videoUrl
+        video.load()
+      }
+
+      if (Math.abs(video.currentTime - localTime) > 0.05) {
+        video.currentTime = localTime
+      }
+
+      segIdxRef.current = index
+      setCurrentSegmentIndex(index)
+
+      await waitForReady(video)
+    },
+    [waitForReady],
+  )
+
+  // 稳定的 tick 函数，空依赖，所有值从 ref 读取
+  const tick = useCallback(() => {
+    const segs = segmentsRef.current
+    const idx = segIdxRef.current
+    const video = videoRefs.current[idx]
+
+    if (!video || isSeekingRef.current) {
+      rafRef.current = requestAnimationFrame(tick)
+      return
+    }
+
+    const seg = segs[idx]
+    if (!seg) return
+
+    // 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 =
+          (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(totalDurationRef.current)
+        return
+      }
+    }
+
+    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)
+  }, [])
+
+  const play = useCallback(async () => {
+    if (!canPlay) return
+    setIsEnded(false)
+    const idx = segIdxRef.current
+    const video = videoRefs.current[idx]
+    if (!video) return
+
+    if (idx === 0 && video.readyState < 2) {
+      if (!video.src && segmentsRef.current[0]?.videoUrl) {
+        video.src = segmentsRef.current[0].videoUrl
+        video.load()
+      }
+      await waitForReady(video)
+    }
+
+    try {
+      await video.play()
+      setIsPlaying(true)
+      cancelAnimationFrame(rafRef.current)
+      rafRef.current = requestAnimationFrame(tick)
+    } catch (err) {
+      console.warn("[useSegmentScheduler] 播放失败:", err)
+    }
+  }, [canPlay, waitForReady, tick])
+
+  const pause = useCallback(() => {
+    const video = videoRefs.current[segIdxRef.current]
+    if (video) video.pause()
+    setIsPlaying(false)
+    cancelAnimationFrame(rafRef.current)
+  }, [])
+
+  const togglePlayPause = useCallback(() => {
+    if (isPlayingRef.current) {
+      pause()
+    } else {
+      if (isEnded) {
+        setIsEnded(false)
+        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()
+      }
+    }
+  }, [isEnded, pause, play, tick])
+
+  const seekTo = useCallback(
+    async (time: number) => {
+      if (!canPlay) return
+      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 !== segIdxRef.current) {
+        await switchToSegment(index, localTime)
+      } else {
+        const video = videoRefs.current[index]
+        if (video) video.currentTime = localTime
+      }
+
+      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, switchToSegment, tick],
+  )
+
+  useEffect(() => {
+    videoRefs.current = videoRefs.current.slice(0, segments.length)
+    while (videoRefs.current.length < segments.length) {
+      videoRefs.current.push(null)
+    }
+  }, [segments])
+
+  useEffect(() => {
+    return () => {
+      cancelAnimationFrame(rafRef.current)
+    }
+  }, [])
+
+  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: totalDurationData,
+    currentSegmentIndex,
+    segmentLocalTime,
+    isEnded,
+    canPlay,
+    play,
+    pause,
+    togglePlayPause,
+    seekTo,
+    videoRefs,
+  }
+}
+
+export default useSegmentScheduler

apps/web/src/pages/generate/hooks/useStep5Preview.ts

@@ -1,476 +0,0 @@
-/**
- * Step 5 生成预览 Hook（支持多预览 + voice_ids）
- * 调用 /generation/preview 接口创建多个预览任务，轮询状态直到全部完成
- */
-import { useState, useCallback, useMemo, useEffect, useRef } from "react"
-import { createPreview, getPreviewStatus } from "@/api/generation"
-import { updateEditPlan } from "@/api/template-editor/editPlans"
-import type { PreviewTaskResponse, PreviewStatus as ApiPreviewStatus } from "@/api/generation"
-import type { EditingTemplate } from "@/api/editing-planner"
-import { safeExtractError } from "./generate-video/errorUtils"
-import type { TitleSettings } from "../types"
-
-/** 安全地将值转为字符串，防止对象被直接渲染导致 React Error #31 */
-const safeString = (val: unknown, fallback: string): string => {
-  if (val == null) return fallback
-  const s = safeExtractError(val)
-  return s || fallback
-}
-
-/** 安全地将值转为数字，防止非数字值进入渲染 */
-const safeNumber = (val: unknown, fallback = 0): number => {
-  if (typeof val === "number" && !Number.isNaN(val)) return val
-  if (typeof val === "string") {
-    const n = Number(val)
-    return Number.isNaN(n) ? fallback : n
-  }
-  return fallback
-}
-
-interface UseStep5PreviewProps {
-  templates: EditingTemplate[]
-  selectedTemplate: string
-  materialMode: "manual" | "auto"
-  selectedMaterials: string[]
-  smartSelectedIds: string[]
-  duration: number
-  videoRatio: string
-  /** 配音 voice_ids（传给后端，让预览包含配音音频） */
-  voiceIds?: string[]
-  /** 配音素材库ID（用户选择的上传音频或AI配音素材） */
-  voiceLibraryId?: string
-  /** 要生成的预览数量 */
-  previewCount?: number
-  /** 标题设置（传递给后端，让预览视频包含标题） */
-  titleSettings?: TitleSettings
-}
-
-export type PreviewStatus = "idle" | "pending" | "generating" | "ready" | "error"
-
-/** 单个预览生成结果 */
-export interface PreviewResult {
-  taskId: string
-  videoUrl: string
-  clipCount: number
-  transitionCount: number
-  materialUsage: number
-  duration: number
-  fileSize: number
-  generateDuration: number
-  progress: number
-}
-
-/** 单个预览项的完整状态（用于多预览） */
-export interface PreviewItem {
-  index: number
-  status: PreviewStatus
-  result: PreviewResult | null
-  error: string
-  progress: number
-}
-
-// 轮询超时时间（10 分钟）
-const POLL_TIMEOUT_MS = 10 * 60 * 1000
-
-/** 初始单项状态 */
-const createInitialItem = (index: number): PreviewItem => ({
-  index,
-  status: "idle",
-  result: null,
-  error: "",
-  progress: 0,
-})
-
-export function useStep5Preview({
-  templates,
-  selectedTemplate,
-  materialMode,
-  selectedMaterials,
-  smartSelectedIds,
-  duration,
-  videoRatio,
-  voiceIds,
-  voiceLibraryId,
-  previewCount = 1,
-  titleSettings,
-}: UseStep5PreviewProps) {
-  const templateName = useMemo(
-    () => templates.find((t) => t.id === selectedTemplate)?.name ?? "未选择",
-    [templates, selectedTemplate],
-  )
-
-  const materialCount = useMemo(() => {
-    if (materialMode === "auto") {
-      return `${smartSelectedIds.length} 个素材（智能匹配）`
-    }
-    return `${selectedMaterials.length} 个素材`
-  }, [materialMode, selectedMaterials.length, smartSelectedIds.length])
-
-  const materialTotal = materialMode === "auto" ? smartSelectedIds.length : selectedMaterials.length
-
-  /* ── 多预览状态 ── */
-  const [items, setItems] = useState<PreviewItem[]>(() =>
-    Array.from({ length: previewCount }, (_, i) => createInitialItem(i)),
-  )
-  const [selectedIndex, setSelectedIndex] = useState(0)
-
-  // 每个任务 ID + 轮询定时器，用于防止竞态条件（按 index 存储）
-  const taskIdsRef = useRef<Map<number, string>>(new Map())
-  const pollTimersRef = useRef<Map<number, ReturnType<typeof setTimeout>>>(new Map())
-  const startTimeRef = useRef<number>(0)
-
-  const clearPollTimer = useCallback((index?: number) => {
-    if (index !== undefined) {
-      const timer = pollTimersRef.current.get(index)
-      if (timer) {
-        clearTimeout(timer)
-        pollTimersRef.current.delete(index)
-      }
-    } else {
-      pollTimersRef.current.forEach((timer) => clearTimeout(timer))
-      pollTimersRef.current.clear()
-    }
-  }, [])
-
-  // 同步 previewCount 变化（增减项）
-  useEffect(() => {
-    setItems((prev) => {
-      if (prev.length === previewCount) return prev
-      if (prev.length > previewCount) return prev.slice(0, previewCount)
-      return [
-        ...prev,
-        ...Array.from({ length: previewCount - prev.length }, (_, i) =>
-          createInitialItem(prev.length + i),
-        ),
-      ]
-    })
-    // 如果 selectedIndex 超出范围，重置
-    setSelectedIndex((prev) => Math.min(prev, previewCount - 1))
-  }, [previewCount])
-
-  /* ── 参数变化时重置所有预览状态 ── */
-  const prevDepsRef = useRef({
-    selectedTemplate,
-    materialMode,
-    selectedMaterials: [...selectedMaterials].sort().join(","),
-    smartSelectedIds: [...smartSelectedIds].sort().join(","),
-    duration,
-    videoRatio,
-    voiceIds: [...(voiceIds || [])].sort().join(","),
-    titleSettings: titleSettings?.title || "",
-  })
-
-  useEffect(() => {
-    const currentKey = [
-      selectedTemplate,
-      materialMode,
-      [...selectedMaterials].sort().join(","),
-      [...smartSelectedIds].sort().join(","),
-      duration,
-      videoRatio,
-      [...(voiceIds || [])].sort().join(","),
-      titleSettings?.title || "",
-    ].join("|")
-
-    const prevKey = [
-      prevDepsRef.current.selectedTemplate,
-      prevDepsRef.current.materialMode,
-      prevDepsRef.current.selectedMaterials,
-      prevDepsRef.current.smartSelectedIds,
-      prevDepsRef.current.duration,
-      prevDepsRef.current.videoRatio,
-      prevDepsRef.current.voiceIds,
-      prevDepsRef.current.titleSettings,
-    ].join("|")
-
-    if (prevKey !== currentKey && items.some((it) => it.status !== "idle")) {
-      taskIdsRef.current.clear()
-      clearPollTimer()
-      setItems(Array.from({ length: previewCount }, (_, i) => createInitialItem(i)))
-      setSelectedIndex(0)
-    }
-
-    prevDepsRef.current = {
-      selectedTemplate,
-      materialMode,
-      selectedMaterials: [...selectedMaterials].sort().join(","),
-      smartSelectedIds: [...smartSelectedIds].sort().join(","),
-      duration,
-      videoRatio,
-      voiceIds: [...(voiceIds || [])].sort().join(","),
-      titleSettings: titleSettings?.title || "",
-    }
-    // eslint-disable-next-line react-hooks/exhaustive-deps
-  }, [
-    selectedTemplate,
-    materialMode,
-    selectedMaterials,
-    smartSelectedIds,
-    duration,
-    videoRatio,
-    voiceIds,
-    previewCount,
-    titleSettings,
-  ])
-
-  // 组件卸载时清理所有轮询
-  useEffect(() => {
-    return () => {
-      clearPollTimer()
-    }
-  }, [clearPollTimer])
-
-  /** 轮询单个预览任务状态 */
-  const pollPreviewStatus = useCallback(
-    (index: number, taskId: string) => {
-      const poll = async () => {
-        // 竞态检查
-        if (taskIdsRef.current.get(index) !== taskId) return
-
-        // 超时检查
-        if (Date.now() - startTimeRef.current > POLL_TIMEOUT_MS) {
-          setItems((prev) =>
-            prev.map((it) =>
-              it.index === index ? { ...it, status: "error", error: "预览生成超时，请重试" } : it,
-            ),
-          )
-          return
-        }
-
-        try {
-          const data: PreviewTaskResponse = await getPreviewStatus(taskId)
-
-          if (taskIdsRef.current.get(index) !== taskId) return
-
-          const status = data.status as ApiPreviewStatus
-
-          if (status === "completed") {
-            const result: PreviewResult = {
-              taskId: safeString(data.task_id, ""),
-              videoUrl: safeString(data.video_url, ""),
-              clipCount: safeNumber(data.clip_count),
-              transitionCount: safeNumber(data.transition_count),
-              materialUsage: safeNumber(data.material_usage),
-              duration: safeNumber(data.duration),
-              fileSize: safeNumber(data.file_size),
-              generateDuration: safeNumber(data.generate_duration),
-              progress: 100,
-            }
-            setItems((prev) =>
-              prev.map((it) =>
-                it.index === index ? { ...it, status: "ready", result, progress: 100 } : it,
-              ),
-            )
-
-            // 保存预览视频 URL 到 plan config，供封面生成使用
-            if (result.videoUrl && selectedTemplate) {
-              updateEditPlan(selectedTemplate, {
-                config: { rendered_storage_key: result.videoUrl },
-              }).catch((err) => {
-                console.warn("[Step4] 保存预览视频URL到plan config失败:", err)
-              })
-            }
-            return
-          }
-
-          if (status === "failed") {
-            setItems((prev) =>
-              prev.map((it) =>
-                it.index === index
-                  ? {
-                      ...it,
-                      status: "error",
-                      error: safeString(data.error_message, "预览生成失败，请重试"),
-                    }
-                  : it,
-              ),
-            )
-            return
-          }
-
-          if (status === "cancelled") {
-            setItems((prev) =>
-              prev.map((it) =>
-                it.index === index ? { ...it, status: "error", error: "预览任务已取消" } : it,
-              ),
-            )
-            return
-          }
-
-          // pending / generating 状态继续轮询
-          const prog = safeNumber(data.progress)
-          const nextStatus: PreviewStatus = status === "pending" ? "pending" : "generating"
-          setItems((prev) =>
-            prev.map((it) =>
-              it.index === index ? { ...it, status: nextStatus, progress: prog } : it,
-            ),
-          )
-          const delay = status === "pending" ? 5000 : 2000
-          pollTimersRef.current.set(index, setTimeout(poll, delay))
-        } catch {
-          if (taskIdsRef.current.get(index) !== taskId) return
-          pollTimersRef.current.set(index, setTimeout(poll, 3000))
-        }
-      }
-
-      pollTimersRef.current.set(index, setTimeout(poll, 1000))
-    },
-    [selectedTemplate],
-  )
-
-  /** 生成所有预览 */
-  const generatePreview = useCallback(async () => {
-    if (!selectedTemplate) {
-      setItems((prev) => prev.map((it) => ({ ...it, status: "error", error: "请先选择模板" })))
-      return
-    }
-    if (materialTotal === 0) {
-      setItems((prev) => prev.map((it) => ({ ...it, status: "error", error: "请先选择素材" })))
-      return
-    }
-
-    // 取消之前的所有轮询
-    clearPollTimer()
-    taskIdsRef.current.clear()
-
-    // 初始化所有项为 pending
-    setItems(
-      Array.from({ length: previewCount }, (_, i) => ({
-        index: i,
-        status: "pending" as PreviewStatus,
-        result: null,
-        error: "",
-        progress: 0,
-      })),
-    )
-    setSelectedIndex(0)
-    startTimeRef.current = Date.now()
-
-    const assetIds = materialMode === "auto" ? smartSelectedIds : selectedMaterials
-
-    // 并发创建所有预览任务（Promise.all 并行请求，减少串行等待）
-    const createTasks = Array.from({ length: previewCount }, async (_, i) => {
-      try {
-        const response = await createPreview({
-          template_id: selectedTemplate,
-          asset_ids: assetIds,
-          duration: duration || undefined,
-          video_ratio: videoRatio,
-          voice_ids: voiceIds && voiceIds.length > 0 ? voiceIds : undefined,
-          voice_library_id: voiceLibraryId || undefined,
-          // 标题烧录配置
-          title_config: titleSettings?.title
-            ? {
-                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,
-              }
-            : undefined,
-        })
-
-        if (startTimeRef.current === 0) return
-
-        taskIdsRef.current.set(i, response.task_id)
-        pollPreviewStatus(i, response.task_id)
-      } catch (e) {
-        const errMsg = safeString(e instanceof Error ? e.message : e, "预览生成失败")
-        setItems((prev) =>
-          prev.map((it) => (it.index === i ? { ...it, status: "error", error: errMsg } : it)),
-        )
-      }
-    })
-    await Promise.all(createTasks)
-  }, [
-    selectedTemplate,
-    materialTotal,
-    materialMode,
-    smartSelectedIds,
-    selectedMaterials,
-    duration,
-    videoRatio,
-    voiceIds,
-    voiceLibraryId,
-    previewCount,
-    titleSettings,
-    clearPollTimer,
-    pollPreviewStatus,
-  ])
-
-  /** 重新生成所有预览 */
-  const regeneratePreview = useCallback(() => {
-    generatePreview()
-  }, [generatePreview])
-
-  /** 是否所有预览都已完成 */
-  const allReady = items.length > 0 && items.every((it) => it.status === "ready")
-  /** 是否至少有一个预览已完成 */
-  const anyReady = items.some((it) => it.status === "ready")
-  /** 是否有任一正在生成中 */
-  const anyGenerating = items.some((it) => it.status === "pending" || it.status === "generating")
-
-  /** 当前选中的预览结果 */
-  const selectedResult = items[selectedIndex]?.result ?? null
-
-  /** 综合状态（兼容旧逻辑） */
-  const previewStatus: PreviewStatus = useMemo(() => {
-    if (items.every((it) => it.status === "idle")) return "idle"
-    if (items.some((it) => it.status === "pending" || it.status === "generating"))
-      return "generating"
-    if (allReady) return "ready"
-    if (items.every((it) => it.status === "error")) return "error"
-    // 部分完成部分出错
-    if (anyReady) return "ready"
-    return "error"
-  }, [items, allReady, anyReady])
-
-  /** 综合进度（取平均） */
-  const progress = useMemo(() => {
-    if (items.length === 0) return 0
-    return Math.round(items.reduce((sum, it) => sum + it.progress, 0) / items.length)
-  }, [items])
-
-  /** 综合错误信息 */
-  const previewError = useMemo(() => {
-    const errorItems = items.filter((it) => it.status === "error" && it.error)
-    if (errorItems.length === 0) return ""
-    if (errorItems.length === 1) return errorItems[0].error
-    return `${errorItems.length} 个预览生成失败`
-  }, [items])
-
-  const canProceed = anyReady
-
-  /** 当前选中预览的 taskId（用于确认生成时复用预览产物） */
-  const selectedTaskId = selectedResult?.taskId ?? ""
-
-  return {
-    templateName,
-    materialCount,
-    duration,
-    videoRatio,
-    // 多预览状态
-    items,
-    selectedIndex,
-    setSelectedIndex,
-    previewCount,
-    // 综合状态
-    previewStatus,
-    previewResult: selectedResult,
-    previewError,
-    progress,
-    canProceed,
-    allReady,
-    anyReady,
-    anyGenerating,
-    generatePreview,
-    regeneratePreview,
-    // 确认生成复用预览产物
-    selectedTaskId,
-  }
-}
-
-export default useStep5Preview

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

@@ -6,6 +6,9 @@ import { useCallback, useEffect, 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,
@@ -21,6 +24,8 @@ 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
 }
 
 export function useStep6Cover({
@@ -29,6 +34,7 @@ export function useStep6Cover({
   duration,
   assetIds = [],
   selectedTemplate = "",
+  titleSettings,
 }: UseStep6CoverProps) {
   const [generating, setGenerating] = useState(false)
 
@@ -109,31 +115,146 @@ export function useStep6Cover({
       console.error("[Step6] 智能封面生成失败:", err)
 
       // 提取详细错误信息
-      let errorMsg = "封面生成失败"
-      const e = err as {
-        response?: { data?: { detail?: string; message?: string }; status?: number }
-        request?: unknown
-        message?: string
-      }
-      if (e.response) {
-        // 后端返回错误
-        const detail = e.response.data?.detail || e.response.data?.message || ""
-        errorMsg = detail || `后端错误 (${e.response.status})`
-        console.error("[Step6] 后端返回:", e.response.data)
-      } else if (e.request) {
-        // 请求已发送但无响应
-        errorMsg = "服务器无响应，请检查网络连接"
-        console.error("[Step6] 请求无响应:", e.request)
-      } else if (e.message) {
-        errorMsg = e.message
-      }
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+      const anyErr = err as any
+      const statusCode = anyErr?.response?.status
 
-      message.error(errorMsg)
+      // 400 错误：精确判断是否为"预览缺失"，避免误判其他 400 错误
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+      const errCode = anyErr?.response?.data?.code as string | undefined
+      // eslint-disable-next-line @typescript-eslint/no-explicit-any
+      const errMsg = (anyErr?.response?.data?.message ||
+        anyErr?.response?.data?.detail ||
+        "") as string
+      const isPreviewMissing =
+        statusCode === 400 &&
+        (errCode?.includes("PREVIEW") ||
+          /预览.*(?:缺失|不存在|未找到)|(?:missing|not found|does not exist).*preview/i.test(
+            errMsg,
+          ))
+
+      if (isPreviewMissing) {
+        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 {
+          const previewResp = await createPreview({
+            template_id: selectedTemplate,
+            asset_ids: assetIds,
+            duration: duration || 30,
+            ...(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,
+                  },
+                }
+              : {}),
+          })
+          // 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
+            const done = (fn: () => void) => {
+              if (finished) return
+              finished = true
+              clearTimeout(timeoutId)
+              fn()
+            }
+            const timeoutId = setTimeout(() => {
+              done(() => reject(new Error("预览生成超时，请稍后重试")))
+            }, 120_000)
+            const poll = async () => {
+              if (finished) return
+              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")))
+                } else {
+                  setTimeout(poll, 2000)
+                }
+              } catch (e) {
+                done(() => reject(e))
+              }
+            }
+            poll()
+          })
+          message.success("预览视频就绪，重新生成封面...")
+          // 重试封面生成
+          const retryResp = await generateCover(selectedTemplate, {
+            asset_ids: assetIds,
+            cover_type: "ai_frame",
+          })
+          const retryUrl = retryResp.cover?.image_url || ""
+          if (retryUrl) {
+            onCoverSettingsChange({
+              ...coverSettings,
+              thumbnail_url: retryUrl,
+              ai_suggested_time: retryResp.cover?.frame_time ?? null,
+            })
+            message.success("封面生成成功")
+          } else {
+            message.warning("封面生成未返回图片，请重试")
+          }
+        } catch (retryErr) {
+          console.error("[Step6] 自动创建预览后重试失败:", retryErr)
+          message.error("预览视频创建失败，请稍后重试")
+        }
+      } else if (anyErr?.__msgShown) {
+        // 拦截器已处理，不再重复弹出
+      } else {
+        let errorMsg = "封面生成失败"
+        const e = anyErr as {
+          response?: { data?: { detail?: string; message?: string }; status?: number }
+          request?: unknown
+          message?: string
+        }
+        if (e.response) {
+          const detail = e.response.data?.detail || e.response.data?.message || ""
+          errorMsg = detail || `后端错误 (${e.response.status})`
+          console.error("[Step6] 后端返回:", e.response.data)
+        } else if (e.request) {
+          errorMsg = "服务器无响应，请检查网络连接"
+          console.error("[Step6] 请求无响应:", e.request)
+        } else if (e.message) {
+          errorMsg = e.message
+        }
+        message.error(errorMsg)
+      }
     } finally {
       clearTimeout(timeoutId)
       setGenerating(false)
     }
-  }, [selectedTemplate, assetIds, coverSettings, onCoverSettingsChange, generating])
+  }, [
+    selectedTemplate,
+    assetIds,
+    coverSettings,
+    onCoverSettingsChange,
+    generating,
+    duration,
+    titleSettings,
+  ])
 
   // ── 模板操作方法 ──
   const handleSelectTemplate = useCallback((id: string) => {

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

@@ -2,6 +2,8 @@
  * 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)
+ *
+ * V24: previewReady 改为前端素材加载状态
  */
 import { message } from "antd"
 import type { TitleSettings } from "../types"
@@ -14,7 +16,7 @@ export interface UseStepNavigationOptions {
   selectedMaterials: string[]
   smartSelectedIds: string[]
   titleSettings: TitleSettings
-  /** Step4 是否已生成预览 */
+  /** 预览是否就绪（前端素材已加载） */
   previewReady: boolean
 }
 
@@ -54,7 +56,7 @@ export const useStepNavigation = (options: UseStepNavigationOptions): UseStepNav
       return
     }
     if (currentStep === 5 && !previewReady) {
-      message.warning("请先生成剪辑预览")
+      message.warning("请先选择素材以预览效果")
       return
     }
     if (currentStep < 7) {

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

@@ -0,0 +1,54 @@
+/**
+ * 共享：根据素材列表和模板片段计算总视频时长
+ * GeneratePage（配音校验）和 FrontendPreviewPlayer（播放控制）共用
+ */
+
+export interface DurationAsset {
+  id?: string
+  duration?: number
+  metadata?: { duration?: number }
+}
+
+export interface DurationTemplateSegment {
+  duration_min?: number
+  duration_max?: number
+}
+
+/**
+ * 计算总视频时长
+ * @param assets 素材列表
+ * @param template 模板（含 segments）
+ * @returns 总时长（秒），无有效数据时返回 0
+ */
+export function calculateTotalVideoDuration(
+  assets: DurationAsset[] | undefined,
+  template: { segments?: DurationTemplateSegment[] } | undefined,
+): number {
+  if (!assets || assets.length === 0 || !template) return 0
+
+  const templateSegments = template.segments || []
+
+  return assets.reduce((sum, asset, i) => {
+    const assetDuration = asset.duration || asset.metadata?.duration || 30
+    const tplSeg = templateSegments[i] || templateSegments[templateSegments.length - 1]
+    const segDuration = tplSeg
+      ? Math.min(
+          tplSeg.duration_max ?? assetDuration,
+          Math.max(tplSeg.duration_min ?? 0, assetDuration),
+        )
+      : Math.min(assetDuration, 10)
+    return sum + segDuration
+  }, 0)
+}
+
+/**
+ * 估算总视频时长（仅依赖模板 segments）
+ * 当素材未加载或加载失败时，用各片段 duration_max 之和作为估算值
+ * 确保配音时长校验不会因素材未就绪而跳过
+ */
+export function estimateTotalVideoDuration(
+  template: { segments?: DurationTemplateSegment[] } | undefined,
+): number {
+  if (!template?.segments || template.segments.length === 0) return 0
+  return template.segments.reduce((sum, seg) => sum + (seg.duration_max || 0), 0)
+}

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

@@ -1,126 +0,0 @@
-/**
- * Canvas 标题绘制工具函数（共享模块）
- *
- * 供 PreviewVideoPanel（预览视频标题叠加）和 TitlePreviewCanvas（标题设置实时预览）共用。
- * 绘制行为与 ASS 字幕引擎一致：逐字换行、居中、描边/阴影。
- */
-import type { TitleSettings } from "../types"
-
-/**
- * 将文本按 maxWidth 逐字换行，返回行数组。
- * 与 ASS 字幕引擎的逐字换行行为一致。
- */
-export function wrapText(ctx: CanvasRenderingContext2D, text: string, maxWidth: number): string[] {
-  const lines: string[] = []
-  let currentLine = ""
-  for (const char of text) {
-    const testLine = currentLine + char
-    if (ctx.measureText(testLine).width > maxWidth && currentLine) {
-      lines.push(currentLine)
-      currentLine = char
-    } else {
-      currentLine = testLine
-    }
-  }
-  if (currentLine) lines.push(currentLine)
-  return lines
-}
-
-/**
- * 在 canvas 上绘制标题文字（含描边/阴影/多行居中）
- *
- * @param ctx canvas 上下文
- * @param w canvas CSS 宽度
- * @param h canvas CSS 高度
- * @param text 标题文字
- * @param settings 标题样式
- * @param paddingX 左右边距（px），与 ASS 的 MarginL/MarginR 对应
- * @param position "top" | "center" | "bottom"
- * @param topOffset 顶部/底部偏移量
- */
-export function drawTitleOnCanvas(
-  ctx: CanvasRenderingContext2D,
-  w: number,
-  h: number,
-  text: string,
-  settings: TitleSettings,
-  paddingX: number,
-  position: string,
-  topOffset: number,
-) {
-  const dpr = window.devicePixelRatio || 1
-
-  // 设置 canvas 物理像素尺寸（高清屏适配）
-  ctx.canvas.width = Math.round(w * dpr)
-  ctx.canvas.height = Math.round(h * dpr)
-  ctx.scale(dpr, dpr)
-
-  // 清除
-  ctx.clearRect(0, 0, w, h)
-
-  // 可用宽度 = 总宽 - 左右边距
-  const availableWidth = w - paddingX * 2
-  if (availableWidth <= 0) return
-
-  // 字体设置
-  const fontSize = Math.round(Math.min(settings.size, 36))
-  const fontWeight = settings.bold ? "bold" : "normal"
-  const fontStyle = settings.italic ? "italic" : "normal"
-  ctx.font = `${fontStyle} ${fontWeight} ${fontSize}px "${settings.font}"`
-
-  // 文字属性
-  ctx.textAlign = "center"
-  ctx.textBaseline = "middle"
-  ctx.fillStyle = settings.color
-
-  const lineHeight = fontSize * 1.4
-
-  // 描边 & 阴影
-  if (settings.stroke) {
-    ctx.strokeStyle = "rgba(0,0,0,0.6)"
-    ctx.lineWidth = 2
-    ctx.lineJoin = "round"
-  }
-  if (settings.shadow) {
-    ctx.shadowColor = "rgba(0,0,0,0.7)"
-    ctx.shadowBlur = 4
-    ctx.shadowOffsetX = 2
-    ctx.shadowOffsetY = 2
-  }
-
-  // 换行
-  const displayText = text && text.trim() ? text : "请选择或输入标题"
-  const lines = wrapText(ctx, displayText, availableWidth)
-
-  // 起始 Y：根据 position 计算
-  const totalTextHeight = lines.length * lineHeight
-  let startY: number
-  switch (position) {
-    case "top":
-      startY = topOffset
-      break
-    case "center":
-      startY = (h - totalTextHeight) / 2 + lineHeight / 2
-      break
-    case "bottom":
-    default:
-      startY = h - topOffset - totalTextHeight + lineHeight / 2
-      break
-  }
-
-  // 居中 x = w/2
-  const x = w / 2
-  lines.forEach((line, i) => {
-    const y = startY + i * lineHeight
-    if (settings.stroke) ctx.strokeText(line, x, y)
-    ctx.fillText(line, x, y)
-  })
-
-  // 重置 shadow（避免影响后续绘制）
-  if (settings.shadow) {
-    ctx.shadowColor = "transparent"
-    ctx.shadowBlur = 0
-    ctx.shadowOffsetX = 0
-    ctx.shadowOffsetY = 0
-  }
-}

apps/web/src/pages/voice-materials/hooks/useVoiceMaterials/actions/useVoiceUpload.ts

@@ -2,14 +2,13 @@ import { useState, useCallback } from "react"
 import { useMutation, useQueryClient } from "@tanstack/react-query"
 import { message } from "antd"
 import {
-  createAsset,
   uploadAssetDirect,
   getAssetLibraries,
+  getIngestJob,
   type AssetLibraryItem,
 } from "@/api/assets"
 import { tagAsset } from "@/api/tags"
-import { type VoiceGender, type VoiceMaterial, buildMetadata } from "../../../types"
-import { getAudioDuration } from "../../../utils/audio"
+import { type VoiceGender, type VoiceMaterial } from "../../../types"
 
 interface UseVoiceUploadOptions {
   voiceLibrary?: { id: string; kind: string }
@@ -48,32 +47,34 @@ export function useVoiceUpload({ voiceLibrary, createLibMutation }: UseVoiceUplo
           if (!lib) throw new Error("无法创建配音库")
         }
 
-        // 2. 上传文件（带进度）
-        const { storage_key } = await uploadAssetDirect({
+        // 2. 上传文件（带进度，后端自动创建 ingest job）
+        const { ingest_job_id } = await uploadAssetDirect({
           file: data.file,
           library_id: lib.id,
           onProgress: (p) => setUploadProgress(p),
         })
 
-        // 3. 获取音频时长
-        const duration = await getAudioDuration(data.file)
+        // 3. 轮询 ingest job 状态
+        let job: Awaited<ReturnType<typeof getIngestJob>> | null = null
+        let retries = 0
+        const maxRetries = 60 // 最多等待 5 分钟
+        while (retries < maxRetries) {
+          await new Promise((r) => setTimeout(r, 5000))
+          job = await getIngestJob(ingest_job_id)
+          if (job.status === "completed" || job.status === "failed") break
+          retries++
+        }
 
-        // 4. 创建素材记录
-        const asset = await createAsset({
-          library_id: lib.id,
-          name: data.name,
-          storage_key,
-          mime_type: data.file.type || "audio/mpeg",
-          metadata: buildMetadata({
-            gender: data.gender,
-            description: data.description,
-            duration,
-          }),
-        })
+        if (!job || job.status === "failed") {
+          throw new Error("音频处理失败，请重试")
+        }
+        if (retries >= maxRetries) {
+          throw new Error("音频处理超时，请稍后在素材库查看")
+        }
 
-        // 5. 打标签（标签走独立 API）
-        if (data.tagIds.length > 0) {
-          await tagAsset(asset.id, data.tagIds)
+        // 4. 打标签（标签走独立 API）
+        if (data.tagIds.length > 0 && job.result_asset_id) {
+          await tagAsset(job.result_asset_id, data.tagIds)
         }
       } finally {
         setUploadProgress(null)

apps/web/src/pages/voice-materials/types.ts

@@ -21,7 +21,7 @@ export interface VoiceMaterial {
   fileUrl?: string
 }
 
-/** 配音素材上传元数据（传递给 createAsset 的 metadata） */
+/** 配音素材上传元数据（上传素材的 metadata） */
 export interface VoiceAssetMetadata {
   gender: VoiceGender
   description: string

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

@@ -1,8 +1,6 @@
 import { useState, useCallback } from "react"
 import { useMutation, useQueryClient } from "@tanstack/react-query"
-import { uploadAssetDirect, getAssetLibraries, createAsset } from "@/api/assets"
-import { getAudioDuration } from "../utils/audio"
-import { buildVoiceMetadata } from "../types"
+import { uploadAssetDirect, getAssetLibraries, getIngestJob } from "@/api/assets"
 
 /**
  * 配音上传 Hook
@@ -33,27 +31,30 @@ export function useVoiceUpload({ showToast }: UseVoiceUploadProps) {
         const lib = libs.find((l) => l.kind === "voice")
         if (!lib) throw new Error("配音库不存在，请先在配音库页面创建")
 
-        /* 直传文件 */
-        const { storage_key } = await uploadAssetDirect({
+        /* 直传文件（后端会自动创建 ingest job） */
+        const { ingest_job_id } = await uploadAssetDirect({
           file: data.file,
           library_id: lib.id,
           onProgress: (p) => setUploadProgress(p),
         })
 
-        /* 获取音频时长 */
-        const duration = await getAudioDuration(data.file)
+        /* 轮询 ingest job 状态，等待 Worker 处理完成 */
+        let jobStatus = ""
+        let retries = 0
+        const maxRetries = 60 // 最多等待 5 分钟（60 * 5秒）
+        while (jobStatus !== "ready" && jobStatus !== "failed" && retries < maxRetries) {
+          await new Promise((r) => setTimeout(r, 5000))
+          const job = await getIngestJob(ingest_job_id)
+          jobStatus = job.status
+          retries++
+        }
 
-        /* 创建素材记录 */
-        await createAsset({
-          library_id: lib.id,
-          name: data.name,
-          storage_key,
-          mime_type: data.file.type || "audio/mpeg",
-          metadata: buildVoiceMetadata({
-            description: data.description,
-            duration,
-          }),
-        })
+        if (jobStatus === "failed") {
+          throw new Error("音频处理失败，请重试")
+        }
+        if (retries >= maxRetries) {
+          throw new Error("音频处理超时，请稍后在素材库查看")
+        }
       } finally {
         setUploadProgress(null)
       }

apps/web/src/pages/voices/types.ts

@@ -39,7 +39,7 @@ export interface ClonedVoiceDisplay {
   sampleUrl?: string
 }
 
-/** 音色上传元数据（传递给 createAsset 的 metadata） */
+/** 音色上传元数据（上传素材的 metadata） */
 export interface VoiceUploadMetadata {
   gender?: string
   description?: string

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

@@ -7,11 +7,9 @@ import {
   deleteAssetLibrary,
   getAssets,
   getAssetsByKind,
-  createAsset,
   updateAsset,
   updateAssetReviewStatus,
   deleteAsset,
-  uploadAsset,
   prepareDirectUpload,
   completeDirectUpload,
   uploadAssetDirect,
@@ -175,22 +173,6 @@ describe("assets API", () => {
     })
   })
 
-  describe("createAsset", () => {
-    it("should resolve successfully", async () => {
-      await expect(createAsset({ 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(createAsset({ name: "test-item" })).rejects.toThrow()
-    })
-  })
-
   describe("updateAsset", () => {
     it("should resolve successfully", async () => {
       await expect(updateAsset("test-assetId")).resolves.not.toThrow()
@@ -239,22 +221,6 @@ describe("assets API", () => {
     })
   })
 
-  describe("uploadAsset", () => {
-    it("should resolve successfully", async () => {
-      await expect(uploadAsset(new FormData())).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(uploadAsset(new FormData())).rejects.toThrow()
-    })
-  })
-
   describe("prepareDirectUpload", () => {
     it("should resolve successfully", async () => {
       await expect(prepareDirectUpload({ name: "test-item" })).resolves.not.toThrow()

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

@@ -20,9 +20,16 @@ vi.mock("@/api/auth", () => ({
   refreshAccessToken: vi.fn(),
 }))
 
+vi.mock("@/api/auth/tokenRefresh", () => ({
+  scheduleProactiveRefresh: vi.fn(),
+  cancelProactiveRefresh: vi.fn(),
+  executeTokenRefresh: vi.fn(),
+}))
+
 import { message } from "antd"
 import { useAuthStore } from "@/store/authStore"
 import { refreshAccessToken } from "@/api/auth"
+import { executeTokenRefresh } from "@/api/auth/tokenRefresh"
 import apiClient from "@/api/client"
 
 // 从真实实例取出拦截器回调
@@ -264,23 +271,28 @@ describe("apiClient - 401 token refresh", () => {
     expect(window.location.href).toBe("/")
   })
 
-  it("refreshes token on 401 and calls setAuth", async () => {
+  it("refreshes token on 401 and calls executeTokenRefresh", async () => {
     const mockSetAuth = vi.fn()
-    vi.mocked(useAuthStore.getState).mockReturnValue({
+    let currentAccessToken = "old-access"
+    vi.mocked(useAuthStore.getState).mockImplementation(() => ({
       user: { id: "1", email: "test@test.com" },
-      accessToken: "old-access",
+      accessToken: currentAccessToken,
       refreshToken: "old-refresh",
       isAuthenticated: true,
       clearAuth: vi.fn(),
-      setAuth: mockSetAuth,
-    } as any)
-    vi.mocked(refreshAccessToken).mockResolvedValue({
-      access_token: "new-access",
-      refresh_token: "new-refresh",
-    } as never)
+      setAuth: ((_user: any, newAccess: string, _newRefresh: string) => {
+        currentAccessToken = newAccess
+        mockSetAuth(_user, newAccess, _newRefresh)
+      }) as any,
+    }))
+
+    // Mock executeTokenRefresh to simulate successful refresh
+    vi.mocked(executeTokenRefresh).mockImplementation(() => {
+      currentAccessToken = "new-access"
+      mockSetAuth({ id: "1", email: "test@test.com" }, "new-access", "new-refresh")
+      return Promise.resolve()
+    })
 
-    // 拦截器重试时会调用 apiClient(config)，会真的发请求，最终会 reject
-    // 但我们只关心刷新逻辑是否正确执行
     const err = makeAxiosError(401, { detail: "Unauthorized" })
 
     try {
@@ -289,31 +301,45 @@ describe("apiClient - 401 token refresh", () => {
       // 重试会因为没有真实网络而失败，忽略
     }
 
-    expect(refreshAccessToken).toHaveBeenCalledWith("old-refresh")
+    expect(executeTokenRefresh).toHaveBeenCalled()
     expect(mockSetAuth).toHaveBeenCalled()
   })
 
   it("handles refresh failure by logging out", async () => {
     const mockClearAuth = vi.fn()
-    vi.mocked(useAuthStore.getState).mockReturnValue({
+    // After executeTokenRefresh fails, it clears auth (sets accessToken to null)
+    // and redirects to /login. The promise resolves (doesn't reject).
+    let currentAccessToken: string | null = "old-access"
+    vi.mocked(useAuthStore.getState).mockImplementation(() => ({
       user: { id: "1", email: "test@test.com" },
-      accessToken: "old-access",
+      accessToken: currentAccessToken,
       refreshToken: "old-refresh",
-      isAuthenticated: true,
-      clearAuth: mockClearAuth,
+      isAuthenticated: currentAccessToken !== null,
+      clearAuth: (() => {
+        currentAccessToken = null
+        mockClearAuth()
+        window.location.href = "/login"
+      }) as any,
       setAuth: vi.fn(),
-    } as any)
-    vi.mocked(refreshAccessToken).mockRejectedValue(new Error("refresh failed") as never)
+    }))
+    // Mock executeTokenRefresh: simulates failure → clears auth + redirects
+    vi.mocked(executeTokenRefresh).mockImplementation(() => {
+      currentAccessToken = null
+      mockClearAuth()
+      window.location.href = "/login"
+      return Promise.resolve()
+    })
 
     const err = makeAxiosError(401, { detail: "Unauthorized" })
 
     try {
       await responseErrorInterceptor(err)
     } catch {
-      // expected
+      // expected - rejects because accessToken is null after failed refresh
     }
 
+    expect(executeTokenRefresh).toHaveBeenCalled()
     expect(mockClearAuth).toHaveBeenCalled()
-    expect(window.location.href).toBe("/")
+    expect(window.location.href).toBe("/login")
   })
 })

apps/web/src/test/components/voice/CloneModal.test.tsx

@@ -10,7 +10,9 @@ vi.mock("@/api/voice-clone", () => ({
 }))
 
 vi.mock("@/api/assets", () => ({
-  uploadAsset: vi.fn(),
+  uploadAssetDirect: vi
+    .fn()
+    .mockResolvedValue({ storage_key: "test", ingest_job_id: "test", url: "http://test" }),
 }))
 
 vi.mock("@/components/ui", () => ({

apps/web/src/test/hooks/useAuth.test.tsx

@@ -31,14 +31,27 @@ vi.mock("react-router-dom", async () => {
 })
 
 vi.mock("@/store/authStore", () => ({
-  useAuthStore: (selector: any) =>
-    selector({
-      user: { id: "1", username: "testuser" },
-      token: "mock-token",
-      isAuthenticated: true,
-      setAuth: mockSetAuth,
-      clearAuth: mockClearAuth,
-    }),
+  useAuthStore: Object.assign(
+    (selector: any) =>
+      selector({
+        user: { id: "1", username: "testuser" },
+        token: "mock-token",
+        refreshToken: "mock-refresh-token",
+        isAuthenticated: true,
+        setAuth: mockSetAuth,
+        clearAuth: mockClearAuth,
+      }),
+    {
+      getState: () => ({
+        user: { id: "1", username: "testuser" },
+        token: "mock-token",
+        refreshToken: "mock-refresh-token",
+        isAuthenticated: true,
+        setAuth: mockSetAuth,
+        clearAuth: mockClearAuth,
+      }),
+    },
+  ),
 }))
 
 vi.mock("@tanstack/react-query", () => ({

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

@@ -180,7 +180,6 @@ vi.mock("@/api/assets", () => ({
   getAssets: vi.fn().mockResolvedValue({ items: [], total: 0 }),
   getAssetsByKind: vi.fn().mockResolvedValue({ items: [], total: 0 }),
   smartMatchAssets: vi.fn().mockResolvedValue({ items: [] }),
-  createAsset: vi.fn().mockResolvedValue({}),
   updateAsset: vi.fn().mockResolvedValue({}),
   deleteAsset: vi.fn().mockResolvedValue({}),
   uploadAssetDirect: vi.fn().mockResolvedValue({}),

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

@@ -165,7 +165,6 @@ vi.mock("@/api/assets", () => ({
   deleteAssetLibrary: vi.fn().mockResolvedValue({}),
   getAssetsByKind: vi.fn().mockResolvedValue({ items: [], total: 0 }),
   getAssets: vi.fn().mockResolvedValue({ items: [], total: 0 }),
-  createAsset: vi.fn().mockResolvedValue({}),
   updateAsset: vi.fn().mockResolvedValue({}),
   deleteAsset: vi.fn().mockResolvedValue({}),
   uploadAssetDirect: vi.fn().mockResolvedValue({}),

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

@@ -47,7 +47,8 @@ describe("GeneratePage module smoke test", () => {
   })
 })
 import "@/pages/generate/hooks/useGenerateVideo"
-import "@/pages/generate/hooks/useStep5Preview"
+import "@/pages/generate/hooks/usePreviewAssets"
+import "@/pages/generate/hooks/useSegmentScheduler"
 import "@/pages/generate/hooks/generate-video/useGenerationPolling"
 import "@/pages/generate/hooks/useGenerateFormState"
 import "@/pages/generate/hooks/useGenerateFormState/useTemplateSelection"

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

@@ -5,7 +5,9 @@
 import { describe, it, expect } from "vitest"
 
 import "@/pages/generate/components/Step5GeneratePreview"
-import "@/pages/generate/hooks/useStep5Preview"
+import "@/pages/generate/hooks/usePreviewAssets"
+import "@/pages/generate/hooks/useSegmentScheduler"
+import "@/pages/generate/components/FrontendPreviewPlayer"
 import "@/pages/generate/hooks/useStepNavigation"
 import "@/pages/generate/components/GenerateStepContent"
 import "@/pages/generate/GeneratePage"

apps/web/src/test/pages/generate/useStep5Voice.test.tsx

@@ -93,5 +93,5 @@ describe("useStep5Voice smoke test", () => {
     )
     expect(result.current).toBeDefined()
     expect(typeof result.current.handlePlayCloneSample).toBe("function")
-  })
+  }, 15_000)
 })

apps/web/vite.config.ts

@@ -13,6 +13,7 @@ export default defineConfig({
     environment: "jsdom",
     globals: true,
     setupFiles: ["./src/test/setup.ts"],
+    testTimeout: 15_000, // 全局 15 秒，防止 CI 高负载时偶发超时
   },
   plugins: [
     react({

apps/worker/video_processing/oss_helpers.py

@@ -22,7 +22,7 @@ logger = logging.getLogger(__name__)
 
 # OSS 上传配置
 OSS_CONNECT_TIMEOUT = 10  # 连接超时（秒），防止 TCP 握手挂死
-OSS_UPLOAD_TOTAL_TIMEOUT = 300  # 单文件上传总超时（秒），防止网络慢时无限卡住
+OSS_UPLOAD_TOTAL_TIMEOUT = 900  # 单文件上传总超时（秒），防止网络慢时无限卡住
 OSS_MULTIPART_THRESHOLD = 100 * 1024 * 1024  # 分片上传阈值：100MB 以上走分片
 OSS_PART_SIZE = 8 * 1024 * 1024  # 分片大小：8MB
 OSS_MULTIPART_NUM_THREADS = 3  # 分片上传并发数
@@ -127,10 +127,10 @@ def download_asset(asset_storage_key: str, local_path: Path) -> bool:
 def _download_via_http(url: str, local_path: Path) -> bool:
     """通过 HTTP 下载文件（支持预签名 URL）。
 
-    使用流式下载避免大文件内存溢出，超时 300s。
+    使用流式下载避免大文件内存溢出，超时 900s。
     """
     try:
-        resp = requests.get(url, stream=True, timeout=300)
+        resp = requests.get(url, stream=True, timeout=900)
         resp.raise_for_status()
         with open(local_path, "wb") as f:
             for chunk in resp.iter_content(chunk_size=8 * 1024 * 1024):
@@ -146,7 +146,7 @@ def upload_to_oss(local_path: Path | str, storage_key: str) -> str | None:
     """上传文件到 OSS，返回公开 URL。
 
     大文件（>100MB）自动走分片上传，降低内存峰值，减少 OOM 风险。
-    上传加总超时保护（默认 300s），防止网络异常时无限挂死。
+    上传加总超时保护（默认 900s），防止网络异常时无限挂死。
 
     Args:
         local_path: 本地文件路径（Path 或 str 均可）

apps/worker/video_processing/render_adapter.py

@@ -585,14 +585,11 @@ 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, num_frames=3, title_text=""
             )
             if cover_candidates:
                 logger.info(

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 +13,38 @@ 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,
+    *,
+    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（白色 + 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,
+        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,
@@ -134,3 +167,92 @@ def _format_seek_time(seconds: float) -> str:
     m = int((seconds % 3600) // 60)
     s = seconds % 60
     return f"{h:02d}:{m:02d}:{s:05.2f}"
+
+
+def generate_and_upload_thumbnail(
+    video_path: str,
+    storage_key: str,
+    *,
+    seek_ratio: float = 0.15,
+) -> str:
+    """从视频中提取一帧缩略图并上传到 OSS。
+
+    Args:
+        video_path: 视频文件路径
+        storage_key: OSS 存储 key
+        seek_ratio: 抽帧位置比例（默认 0.15）
+
+    Returns:
+        上传后的 URL 字符串
+
+    Raises:
+        RuntimeError: 抽帧或上传失败
+    """
+    from video_processing.oss_helpers import upload_to_oss
+
+    tmp = tempfile.NamedTemporaryFile(suffix=".jpg", delete=False)
+    tmp.close()
+    try:
+        frame_path = extract_first_frame(video_path, output_path=tmp.name, seek_ratio=seek_ratio)
+        url = upload_to_oss(frame_path, storage_key)
+        if not url:
+            raise RuntimeError(f"上传缩略图到 OSS 失败: {storage_key}")
+        return url
+    finally:
+        Path(tmp.name).unlink(missing_ok=True)
+
+
+def extract_and_upload_cover_frames(
+    video_path: str,
+    plan_id: str,
+    *,
+    num_frames: int = 3,
+    title_text: str = "",
+) -> list[dict]:
+    """从视频中抽取多帧作为封面候选，上传到 OSS。
+
+    Args:
+        video_path: 视频文件路径
+        plan_id: 编辑计划 ID（用于生成 storage key）
+        num_frames: 抽取帧数（默认 3）
+        title_text: 标题文字；非空时用 Pillow 叠加到每帧（白色 + 黑色描边）。
+            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。
+
+    Returns:
+        封面候选列表，每项包含 {"url": str, "position": float}
+    """
+    from video_processing.ffmpeg_utils import probe_duration
+    from video_processing.oss_helpers import upload_to_oss
+
+    try:
+        duration = probe_duration(video_path)
+    except Exception:
+        duration = 0.0
+
+    candidates: list[dict] = []
+    # 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)
+        tmp = tempfile.NamedTemporaryFile(suffix=".jpg", delete=False)
+        tmp.close()
+        try:
+            frame_path = extract_first_frame(
+                video_path,
+                output_path=tmp.name,
+                seek_ratio=ratio,
+                min_seek_seconds=0.5,
+            )
+            # 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)
+            storage_key = f"covers/{plan_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
+                candidates.append({"url": url, "position": round(seek_time, 2)})
+        except Exception as e:
+            logger.warning("[thumbnail] 封面候选帧 %d 提取失败: %s", i, e)
+        finally:
+            Path(tmp.name).unlink(missing_ok=True)
+
+    return candidates

apps/worker/worker_app/celery_app.py

@@ -19,4 +19,14 @@ celery_app.conf.imports = (
     "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/_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

apps/worker/worker_app/tasks/cleanup.py

@@ -0,0 +1,35 @@
+"""定期清理任务 — Celery Beat 调度。
+
+包含：
+- cleanup_stale_pending_tasks: 定期清理卡在 pending 超时的 generation_tasks
+"""
+
+import logging
+
+from celery import shared_task
+from worker_app.tasks._startup import (
+    PENDING_TASK_TIMEOUT_MINUTES,
+    cleanup_stale_pending_tasks,
+)
+
+logger = logging.getLogger(__name__)
+
+
+@shared_task(name="worker.cleanup_stale_pending_tasks")
+def scheduled_cleanup_stale_pending(timeout_minutes: int = PENDING_TASK_TIMEOUT_MINUTES) -> dict:
+    """Celery Beat 调度的定期任务：清理超时的 pending 任务。
+
+    每 10 分钟执行一次（由 celery_app.py 的 beat_schedule 配置），
+    查找所有 status='pending' 且 created_at < NOW() - timeout_minutes
+    的 generation_tasks，批量更新为 failed。
+
+    Args:
+        timeout_minutes: 超时时间（分钟），默认 30 分钟
+
+    Returns:
+        {"cleaned": int}
+    """
+    count = cleanup_stale_pending_tasks(timeout_minutes)
+    if count > 0:
+        logger.info("[Beat] 清理了 %d 个超时 pending 任务（超时阈值 %d 分钟）", count, timeout_minutes)
+    return {"cleaned": count}

apps/worker/worker_app/tasks/edit_plan_generation.py

@@ -98,6 +98,7 @@ def _finalize_render_success(
     output_path: Path,
     engine: str,
     thumbnail_url: str = "",
+    cover_candidates: list[dict] | None = None,
 ) -> dict:
     """渲染成功后的统一收尾：查重 + 更新状态 + 返回结果。"""
     # 创建 GeneratedVideo 记录 + 查重
@@ -163,6 +164,19 @@ def _finalize_render_success(
                 clip_count=len(rendered_clip_ids),
             )
             gen_task.completed_at = datetime.now(timezone.utc)
+
+            # 回写封面 URL 到 GenerationTask，供封面生成接口读取
+            if cover_candidates:
+                first_cover = cover_candidates[0].get("image_url") or cover_candidates[0].get("url") or ""
+                if first_cover:
+                    gen_task.cover_url = first_cover
+                    logger.info(
+                        "预览渲染完成，回写 cover_url: plan_id=%s task_id=%s url=%s",
+                        plan_id,
+                        generation_task_id,
+                        first_cover[:80],
+                    )
+
             gen_task_repo.update(gen_task)
 
     logger.info(
@@ -288,6 +302,7 @@ def _render_with_unified(
         output_path=output_path,
         engine="unified",
         thumbnail_url=thumbnail_url,
+        cover_candidates=result.cover_candidates,
     )
 
 

apps/worker/worker_app/tasks/generation.py

@@ -15,7 +15,6 @@ from __future__ import annotations
 
 import json
 import logging
-import os
 import tempfile
 import time
 from pathlib import Path
@@ -37,8 +36,6 @@ from packages.domain.bgm_utils import merge_bgm_config
 OUTPUT_WIDTH = 1280
 OUTPUT_HEIGHT = 720
 OUTPUT_FPS = 25.0
-OUTPUT_DURATION_SECONDS = 5.0
-GENERATED_FILES_DIR = Path(os.getenv("GENERATED_FILES_DIR", "/app/generated"))
 
 logger = logging.getLogger(__name__)
 
@@ -155,7 +152,6 @@ def _flush_logs(task_id: str, gen_task) -> None:
 # ── 共享工具模块导入 ──────────────────────────────────────────────────────────
 
 from video_processing.dedup_helpers import create_video_record_and_dedup
-from video_processing.ffmpeg_utils import FFMPEG_BIN, run_ffmpeg
 from video_processing.oss_helpers import (
     download_asset,
     get_signed_download_url,
@@ -330,61 +326,6 @@ def _build_plan_and_clips_from_task(
     return plan, clips, asset_path_map
 
 
-def _create_fallback_clip(output_path: Path, title: str) -> None:
-    """创建 fallback 视频（无素材时）"""
-    safe_title = title.replace(":", "\\:").replace("'", "\\'")[:80]
-    run_ffmpeg(
-        [
-            FFMPEG_BIN,
-            "-y",
-            "-f",
-            "lavfi",
-            "-i",
-            f"color=c=#111827:s={OUTPUT_WIDTH}x{OUTPUT_HEIGHT}:d={OUTPUT_DURATION_SECONDS}:r={int(OUTPUT_FPS)}",
-            "-vf",
-            f"drawtext=text='{safe_title}':fontcolor=white:fontsize=48:x=(w-text_w)/2:y=(h-text_h)/2",
-            "-c:v",
-            "libx264",
-            "-pix_fmt",
-            "yuv420p",
-            "-movflags",
-            "+faststart",
-            str(output_path),
-        ]
-    )
-
-
-def _mux_audio_track(video_path: Path, audio_path: str, output_path: Path) -> None:
-    """将音频轨混入已渲染的视频（后处理步骤）。
-
-    使用 FFmpeg 将视频和音频合并，视频时长为准，音频不足则循环，
-    音频过长则截断。
-    """
-    command = [
-        FFMPEG_BIN,
-        "-y",
-        "-i",
-        str(video_path),
-        "-i",
-        audio_path,
-        "-c:v",
-        "copy",
-        "-c:a",
-        "aac",
-        "-b:a",
-        "192k",
-        "-shortest",
-        "-map",
-        "0:v:0",
-        "-map",
-        "1:a:0",
-        "-movflags",
-        "+faststart",
-        str(output_path),
-    ]
-    run_ffmpeg(command)
-
-
 def _download_voice_asset(voice_library_id: str, local_path: Path) -> bool:
     """下载配音文件。
 
@@ -428,105 +369,6 @@ def _download_voice_asset(voice_library_id: str, local_path: Path) -> bool:
     return download_asset(storage_key, local_path)
 
 
-def _prepare_bgm_track(
-    *,
-    bgm_config: dict,
-    temp_path: Path,
-    task_id: str = "",
-) -> str | None:
-    """准备 BGM 音频文件（下载到本地）.
-
-    支持 3 种来源（按优先级）：
-    1. audio_url — 外部直链 URL（最高优先级）
-    2. asset_id — 素材库中的音频素材
-    3. preset_id — 预设 BGM 库
-
-    Returns:
-        BGM 本地文件路径，准备失败返回 None
-    """
-    from urllib.parse import urlparse
-
-    audio_url = bgm_config.get("audio_url", "") or ""
-    asset_id = bgm_config.get("asset_id", "") or ""
-    preset_id = bgm_config.get("preset_id", "") or ""
-
-    bgm_file = temp_path / f"bgm_{task_id or 'track'}.mp3"
-
-    # 优先级1：外部直链 URL
-    if audio_url:
-        try:
-            parsed = urlparse(audio_url)
-            if parsed.scheme in ("http", "https"):
-                from video_processing.url_security import (
-                    ALLOWED_AUDIO_MIME_TYPES,
-                    safe_download_file,
-                )
-
-                logger.info("[task_id=%s] [BGM] 从URL下载: %s", task_id, audio_url[:80])
-                safe_download_file(
-                    audio_url,
-                    str(bgm_file),
-                    purpose="bgm_download",
-                    allowed_mime_types=ALLOWED_AUDIO_MIME_TYPES,
-                    timeout=60.0,
-                )
-                if bgm_file.exists() and bgm_file.stat().st_size > 0:
-                    return str(bgm_file)
-        except Exception as e:
-            logger.warning("[task_id=%s] [BGM] URL下载失败: %s", task_id, e)
-
-    # 优先级2：素材库素材
-    if asset_id:
-        try:
-            from app.core.db import SessionLocal
-
-            from packages.adapters.sqlalchemy_impl.models import AssetModel
-
-            session = SessionLocal()
-            try:
-                model = session.query(AssetModel).filter(AssetModel.id == asset_id).first()
-                if model and (model.storage_key or model.file_url):
-                    # 兼容存量数据：storage_key 为空时 fallback 到 file_url
-                    storage_key = model.storage_key or model.file_url
-                    logger.info("[task_id=%s] [BGM] 从素材库下载: asset_id=%s", task_id, asset_id)
-                    ok = download_asset(storage_key, bgm_file)
-                    if ok and bgm_file.exists() and bgm_file.stat().st_size > 0:
-                        return str(bgm_file)
-            finally:
-                session.close()
-        except Exception as e:
-            logger.warning("[task_id=%s] [BGM] 素材库下载失败: %s", task_id, e)
-
-    # 优先级3：预设 BGM 库
-    if preset_id:
-        try:
-            from packages.domain.preset_bgm import get_preset_bgm
-
-            preset = get_preset_bgm(preset_id)
-            if preset and preset.audio_url:
-                from video_processing.url_security import (
-                    ALLOWED_AUDIO_MIME_TYPES,
-                    safe_download_file,
-                )
-
-                logger.info("[task_id=%s] [BGM] 从预设库下载: preset_id=%s", task_id, preset_id)
-                safe_download_file(
-                    preset.audio_url,
-                    str(bgm_file),
-                    purpose="bgm_preset_download",
-                    allowed_mime_types=ALLOWED_AUDIO_MIME_TYPES,
-                    timeout=60.0,
-                )
-                if bgm_file.exists() and bgm_file.stat().st_size > 0:
-                    return str(bgm_file)
-        except Exception as e:
-            logger.warning("[task_id=%s] [BGM] 预设库下载失败: %s", task_id, e)
-
-    # 所有来源都失败
-    logger.warning("[task_id=%s] [BGM] 所有来源都无法获取BGM，跳过", task_id)
-    return None
-
-
 def _verify_url_accessible(
     url: str,
     timeout: float = 10.0,
@@ -1045,6 +887,7 @@ def _load_task_info(task_id: str) -> dict | None:
             "output_height": getattr(gen_task, "output_height", OUTPUT_HEIGHT) or OUTPUT_HEIGHT,
             "cover_url": getattr(gen_task, "cover_url", "") or "",
             "custom_title": getattr(gen_task, "custom_title", "") or "",
+            "title_config": dict(getattr(gen_task, "title_config", {}) or {}),
             "voice_ids": list(getattr(gen_task, "voice_ids", []) or []),
         }
     finally:
@@ -1125,14 +968,15 @@ def _render_video(
     bgm_config: dict | None = None,
     voice_ids: list[str] | None = None,
     custom_title: str = "",
-) -> tuple[Path, float]:
+    title_config: dict | None = None,
+) -> tuple[Path, float, list[dict] | None]:
     """渲染视频（含配音混音）。
 
     使用 RenderAdapter 统一渲染入口，复用 BGM/ASR/分辨率/缩略图逻辑。
 
     Args:
     Returns:
-        (output_path, render_duration)
+        (output_path, render_duration, cover_candidates)
     """
     if not downloaded_videos:
         raise RuntimeError(f"素材下载结果为空: task_id={task_id}")
@@ -1157,27 +1001,34 @@ def _render_video(
                 list(template_config.keys()),
             )
 
-    # ── 用户自定义标题覆盖模板标题配置 ──────────────────────────────────
-    if custom_title:
+    # ── 用户自定义标题：title_config 优先，custom_title 兜底 ─────────────
+    effective_title_cfg: dict | None = None
+    if title_config and isinstance(title_config, dict) and title_config.get("text", "").strip():
+        effective_title_cfg = dict(title_config)
+    elif custom_title:
         try:
-            user_title_cfg = json.loads(custom_title) if isinstance(custom_title, str) else custom_title
-            if isinstance(user_title_cfg, dict) and user_title_cfg.get("text", "").strip():
-                # 字段名归一化: 前端 font_size/font_color → 后端 size/color
-                if "font_size" in user_title_cfg and "size" not in user_title_cfg:
-                    user_title_cfg["size"] = user_title_cfg["font_size"]
-                if "font_color" in user_title_cfg and "color" not in user_title_cfg:
-                    user_title_cfg["color"] = user_title_cfg["font_color"]
-                plan_cfg = dict(virtual_plan.config or {})
-                plan_cfg["title"] = user_title_cfg
-                virtual_plan.config = plan_cfg
-                logger.info(
-                    "[task_id=%s] [渲染] 用户自定义标题已注入: text=%s",
-                    task_id,
-                    user_title_cfg.get("text", "")[:30],
-                )
+            parsed = json.loads(custom_title) if isinstance(custom_title, str) else custom_title
+            if isinstance(parsed, dict) and parsed.get("text", "").strip():
+                effective_title_cfg = parsed
         except (json.JSONDecodeError, TypeError):
             logger.warning("[task_id=%s] custom_title JSON解析失败: %s", task_id, custom_title[:100])
 
+    if effective_title_cfg:
+        # 字段名归一化: 前端 font_size/font_color → 后端 size/color
+        if "font_size" in effective_title_cfg and "size" not in effective_title_cfg:
+            effective_title_cfg["size"] = effective_title_cfg["font_size"]
+        if "font_color" in effective_title_cfg and "color" not in effective_title_cfg:
+            effective_title_cfg["color"] = effective_title_cfg["font_color"]
+        plan_cfg = dict(virtual_plan.config or {})
+        plan_cfg["title"] = effective_title_cfg
+        virtual_plan.config = plan_cfg
+        logger.info(
+            "[task_id=%s] [渲染] 标题配置已注入(source=%s): text=%s",
+            task_id,
+            "title_config" if title_config else "custom_title",
+            effective_title_cfg.get("text", "")[:30],
+        )
+
     # 用户自定义 BGM 覆盖模板 BGM（用户指定优先级最高）
     if bgm_config:
         plan_cfg = virtual_plan.config or {}
@@ -1271,8 +1122,10 @@ def _render_video(
 
     # 配音素材库音频已在统一渲染引擎内部通过 audio 图层混音处理
     output_path = render_output_path
+    # RenderAdapter 在渲染完成后用本地 ffmpeg 抽取的封面候选帧（已上传 OSS）
+    cover_candidates = getattr(render_result, "cover_candidates", None)
 
-    return output_path, render_duration
+    return output_path, render_duration, cover_candidates
 
 
 def _upload_and_record(
@@ -1543,7 +1396,7 @@ def generate_video(self, task_id: str) -> dict:
             else:
                 _resolved_resolution = task_info.get("resolution", "")
 
-            output_path, render_duration = _render_video(
+            output_path, render_duration, cover_candidates = _render_video(
                 task_id=task_id,
                 downloaded_videos=downloaded_videos,
                 voice_path=audio_path,
@@ -1557,6 +1410,7 @@ def generate_video(self, task_id: str) -> dict:
                 bgm_config=task_info.get("bgm_config", {}),
                 voice_ids=task_info.get("voice_ids", []),
                 custom_title=task_info.get("custom_title", ""),
+                title_config=task_info.get("title_config", {}),
             )
 
             if gen_task:
@@ -1588,51 +1442,47 @@ def generate_video(self, task_id: str) -> dict:
 
             _update_task_progress(task_id, 95, "上传完成")
 
-            # ── 4.5 封面抽帧 ────────────────────────────────────────────────
-            # 预览视频上传完成后，提取封面帧写入 gen_task.cover_url
-            # 这样封面路由（generation_cover.py 步骤A）可以通过 generation_task_id 直接找到
+            # ── 4.5 封面帧持久化 ────────────────────────────────────────────
+            # RenderAdapter 在渲染完成后已用本地 ffmpeg 从 output_path 抽帧
+            # （标题通过 ASS 烧录，帧天然带标题），并上传 OSS 返回 cover_candidates。
+            # 这里把第一帧写入 gen_task.cover_url，完整列表写入 metadata，
+            # 封面路由（generation_cover.py）的 A/B/C/D 步骤即可直接命中。
             try:
-                from packages.shared.mediakit_client import get_mediakit_client
+                if cover_candidates:
+                    first = cover_candidates[0]
+                    # 候选帧字段兼容：RenderAdapter 用 image_url，thumbnail_generator 用 url
+                    cover_frame_url = first.get("image_url") or first.get("url") or ""
+                    if cover_frame_url:
+                        _cover_session = SessionLocal()
+                        try:
+                            from packages.adapters.sqlalchemy_impl.models import (
+                                GenerationTaskModel,
+                            )
 
-                mk_client = get_mediakit_client()
-                if mk_client.is_available:
-                    _update_task_progress(task_id, 96, "提取封面帧")
-                    snapshots = mk_client.extract_frames(
-                        video_url=file_url,
-                        strategy="SpecifiedFrames",
-                        max_frames=1,
-                    )
-                    if snapshots and len(snapshots) > 0:
-                        cover_frame_url = snapshots[0].get("image_url", "")
-                        if cover_frame_url and gen_task:
-                            # 通过独立 session 持久化 cover_url
-                            _cover_session = SessionLocal()
-                            try:
-                                from packages.adapters.sqlalchemy_impl.models import (
-                                    GenerationTaskModel,
+                            _cover_model = (
+                                _cover_session.query(GenerationTaskModel)
+                                .filter(GenerationTaskModel.id == task_id)
+                                .first()
+                            )
+                            if _cover_model:
+                                _cover_model.cover_url = cover_frame_url
+                                # 持久化完整候选列表到 metadata
+                                meta = dict(_cover_model.metadata or {})
+                                meta["cover_candidates"] = cover_candidates
+                                _cover_model.metadata = meta
+                                _cover_session.commit()
+                                logger.info(
+                                    "[task_id=%s] 封面帧已持久化(ffmpeg本地抽帧): cover_url=%s candidates=%d",
+                                    task_id,
+                                    cover_frame_url[:80],
+                                    len(cover_candidates),
                                 )
-
-                                _cover_model = (
-                                    _cover_session.query(GenerationTaskModel)
-                                    .filter(GenerationTaskModel.id == task_id)
-                                    .first()
-                                )
-                                if _cover_model:
-                                    _cover_model.cover_url = cover_frame_url
-                                    _cover_session.commit()
-                                    logger.info(
-                                        "[task_id=%s] 封面帧提取成功: %s",
-                                        task_id,
-                                        cover_frame_url[:80],
-                                    )
-                            finally:
-                                _cover_session.close()
-                    else:
-                        logger.warning("[task_id=%s] 封面帧提取返回空结果", task_id)
+                        finally:
+                            _cover_session.close()
                 else:
-                    logger.warning("[task_id=%s] MediaKit 未配置，跳过封面帧提取", task_id)
+                    logger.warning("[task_id=%s] 渲染未产出 cover_candidates，封面将依赖 API 兜底", task_id)
             except Exception:
-                logger.warning("[task_id=%s] 封面帧提取失败（不影响主流程）", task_id, exc_info=True)
+                logger.warning("[task_id=%s] 封面帧持久化失败（不影响主流程）", task_id, exc_info=True)
 
             # ── 5. 标记完成 ──────────────────────────────────────────────────
             _update_task_status(task_id, "mark_completed", result_count=video_count)

apps/worker/worker_app/tasks/ingest.py

@@ -1,3 +1,4 @@
+import shutil
 import subprocess
 import tempfile
 from datetime import datetime, timezone
@@ -233,6 +234,183 @@ def ingest_asset(job_id: str) -> dict:
                         job_id,
                         thumb_err,
                     )
+
+            # ── HEVC 自动转码为 1080p H.264 ──────────────────────────────
+            # 浏览器 WebCodecs 硬件解码 HEVC 输出黑帧，上传时自动转码
+            # 失败时降级使用原始文件，不阻塞上传流程
+            if media_type == "video" and local_file and local_file.exists():
+                codec = (metadata.get("codec") or "").lower()
+                if codec in ("hevc", "h265", "hvh1"):
+                    logger.info(
+                        "检测到 HEVC 编码 (codec=%s)，启动转码: job_id=%s",
+                        codec,
+                        job_id,
+                    )
+                    _tc_tmp = None
+                    _needs_rotation = False
+
+                    # ── Step 1: 磁盘空间检查（独立 try/except，失败仍尝试转码）──
+                    try:
+                        _disk_usage = shutil.disk_usage("/tmp")
+                        _free_gb = _disk_usage.free / (1024**3)
+                        if _free_gb < 2:
+                            raise RuntimeError(f"磁盘空间不足 ({_free_gb:.1f}GB < 2GB)")
+                    except Exception as _disk_err:
+                        logger.warning("磁盘检查失败，仍尝试转码: job_id=%s err=%s", job_id, _disk_err)
+
+                    # ── Step 2: ffprobe 旋转检测（独立 try/except，失败不阻塞转码）──
+                    try:
+                        _probe_cmd = [
+                            "ffprobe",
+                            "-v",
+                            "error",
+                            "-select_streams",
+                            "v:0",
+                            "-show_entries",
+                            "side_data=rotation",
+                            "-show_entries",
+                            "stream_tags=rotate",
+                            "-of",
+                            "default=noprint_wrappers=1:nokey=1",
+                            str(local_file),
+                        ]
+                        _probe_result = subprocess.run(
+                            _probe_cmd,
+                            stdout=subprocess.PIPE,
+                            stderr=subprocess.DEVNULL,
+                            text=True,
+                            timeout=60,  # 大文件在容器 overlay 文件系统上解析可能较慢
+                        )
+                        _rotation_str = (_probe_result.stdout or "").strip().split("\n")[0]
+                        if _rotation_str in ("90", "270", "-90"):
+                            _needs_rotation = True
+                            logger.info(
+                                "检测到竖屏视频 (rotation=%s)，将物理旋转画面: job_id=%s",
+                                _rotation_str,
+                                job_id,
+                            )
+                    except subprocess.TimeoutExpired:
+                        logger.warning(
+                            "ffprobe 旋转检测超时(60s)，跳过旋转继续转码: job_id=%s",
+                            job_id,
+                        )
+                        _needs_rotation = False
+                    except Exception as _probe_err:
+                        logger.warning(
+                            "ffprobe 旋转检测异常，跳过旋转继续转码: job_id=%s err=%s",
+                            job_id,
+                            _probe_err,
+                        )
+                        _needs_rotation = False
+
+                    # ── Step 3: ffmpeg 转码（独立 try/except）──
+                    try:
+                        _tc_tmp_file = tempfile.NamedTemporaryFile(delete=False, suffix="_h264.mp4")
+                        _tc_tmp = Path(_tc_tmp_file.name)
+                        _tc_tmp_file.close()  # 关闭文件描述符，ffmpeg 会自己打开
+
+                        # 构建 video filter：竖屏先旋转再缩放
+                        if _needs_rotation:
+                            _vf = "transpose=1,scale='if(gt(ih,1080),-2,iw)':'if(gt(ih,1080),1080,ih)'"
+                        else:
+                            _vf = "scale='if(gt(ih,1080),-2,iw)':'if(gt(ih,1080),1080,ih)'"
+
+                        _cmd = [
+                            "ffmpeg",
+                            "-y",
+                            "-i",
+                            str(local_file),
+                            "-c:v",
+                            "libx264",
+                            "-preset",
+                            "fast",
+                            "-crf",
+                            "18",
+                            "-vf",
+                            _vf + ",format=yuv420p",
+                            "-colorspace",
+                            "bt709",
+                            "-color_primaries",
+                            "bt709",
+                            "-color_trc",
+                            "bt709",
+                            "-pix_fmt",
+                            "yuv420p",
+                            "-level",
+                            "4.2",
+                        ]
+                        # 竖屏视频：清除旋转元数据
+                        if _needs_rotation:
+                            _cmd.extend(["-metadata:s:v:0", "rotate=0"])
+                        _cmd.extend(
+                            [
+                                "-c:a",
+                                "aac",
+                                "-b:a",
+                                "128k",
+                                "-movflags",
+                                "+faststart",
+                                str(_tc_tmp),
+                            ]
+                        )
+                        _proc = subprocess.run(
+                            _cmd,
+                            stdout=subprocess.DEVNULL,
+                            stderr=subprocess.PIPE,
+                            text=True,
+                            timeout=900,
+                        )
+                        if _proc.returncode == 0 and _tc_tmp.exists() and _tc_tmp.stat().st_size > 0:
+                            from video_processing.oss_helpers import upload_to_oss
+
+                            _p = Path(job.storage_key)
+                            _new_key = str(_p.parent / (_p.stem + "_h264" + _p.suffix))
+                            _url = upload_to_oss(_tc_tmp, _new_key)
+                            if _url:
+                                # 先提取元数据，确认成功后再更新 storage_key（避免脏数据）
+                                _new_metadata, _new_extract_success = extract_media_metadata(
+                                    str(_tc_tmp),
+                                    media_type,
+                                )
+                                if _new_extract_success:
+                                    job.storage_key = _new_key
+                                    metadata = _new_metadata
+                                    extract_success = _new_extract_success
+                                logger.info(
+                                    "HEVC→H.264 转码完成: job_id=%s key=%s",
+                                    job_id,
+                                    _new_key[:80],
+                                )
+                            else:
+                                logger.warning(
+                                    "转码文件上传 OSS 失败，使用原始文件: job_id=%s",
+                                    job_id,
+                                )
+                        else:
+                            _tail = _proc.stderr[-300:] if _proc.stderr else ""
+                            logger.warning(
+                                "FFmpeg 转码失败 rc=%s stderr=%s: job_id=%s",
+                                _proc.returncode,
+                                _tail,
+                                job_id,
+                            )
+                    except subprocess.TimeoutExpired:
+                        logger.warning(
+                            "FFmpeg 转码超时(900s)，降级原始文件: job_id=%s",
+                            job_id,
+                        )
+                    except Exception as _e:
+                        logger.warning(
+                            "HEVC 转码异常（降级原始文件）: job_id=%s err=%s",
+                            job_id,
+                            _e,
+                        )
+                    finally:
+                        if _tc_tmp and _tc_tmp.exists():
+                            try:
+                                _tc_tmp.unlink()
+                            except OSError:
+                                pass
         finally:
             if local_file and local_file.exists():
                 try:

infra/docker/api-base.Dockerfile

@@ -0,0 +1,45 @@
+# ============================================================
+# API 基础镜像（预构建）
+# 预装系统依赖 + Python 依赖，业务构建从此镜像开始
+# 当 requirements-base.txt 或 requirements.txt 变更时重新构建
+# 目标：将 API Image 构建时间从 15-20 分钟降至 3-5 分钟
+# ============================================================
+
+FROM git.xiaoxiajianji.com/xiaoxia/base/python:3.12-slim
+
+# 使用阿里云镜像加速
+RUN sed -i 's|deb.debian.org|mirrors.aliyun.com|g' /etc/apt/sources.list.d/debian.sources 2>/dev/null || \
+    sed -i 's|deb.debian.org|mirrors.aliyun.com|g' /etc/apt/sources.list 2>/dev/null || true
+
+# 预装系统依赖（gcc 编译 psycopg/pg 扩展，libpq-dev 编译期，libpq5 运行期，ffmpeg 封面取帧）
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    gcc \
+    libpq-dev \
+    libpq5 \
+    ffmpeg \
+    fonts-noto-cjk \
+    fontconfig \
+    && rm -rf /var/lib/apt/lists/*
+
+# 创建虚拟环境
+RUN python -m venv /opt/venv
+ENV PATH="/opt/venv/bin:$PATH"
+
+WORKDIR /tmp
+
+# 预装 Python 基础依赖
+COPY requirements-base.txt requirements.txt ./
+RUN pip install --no-cache-dir \
+    -i https://mirrors.aliyun.com/pypi/simple/ \
+    --trusted-host mirrors.aliyun.com \
+    -r requirements-base.txt -r requirements.txt
+
+# 虚拟环境瘦身
+RUN find /opt/venv -name "*.so" -type f -exec strip --strip-all {} \; 2>/dev/null || true
+RUN find /opt/venv -type d -name "__pycache__" -exec rm -rf {} + 2>/dev/null; \
+    find /opt/venv -name "*.pyc" -delete 2>/dev/null || true
+
+# 清理临时文件
+RUN rm -f /tmp/requirements-base.txt /tmp/requirements.txt
+
+ENV PYTHONPATH=/app

infra/docker/api.Dockerfile

@@ -1,81 +1,28 @@
 # ============================================================
 # API Dockerfile - FastAPI 应用
-# 优化：多阶段构建 + pip cache mount + 依赖分层缓存
+# 优化：从预构建基础镜像开始，仅叠加业务代码
+# 基础镜像包含所有系统依赖和 Python 依赖，构建时间 < 5 分钟
 # ============================================================
 
-# ==================== Builder 阶段 ====================
-FROM git.xiaoxiajianji.com/xiaoxia/base/python:3.12-slim AS builder
-
-# 使用阿里云镜像加速
-RUN sed -i 's|deb.debian.org|mirrors.aliyun.com|g' /etc/apt/sources.list.d/debian.sources 2>/dev/null || \
-    sed -i 's|deb.debian.org|mirrors.aliyun.com|g' /etc/apt/sources.list 2>/dev/null || true
-
-# 安装编译依赖（仅 builder 需要）
-RUN apt-get update && apt-get install -y --no-install-recommends \
-    gcc \
-    libpq-dev \
-    && rm -rf /var/lib/apt/lists/*
-
-# 创建虚拟环境
-RUN python -m venv /opt/venv
-ENV PATH="/opt/venv/bin:$PATH"
-
-WORKDIR /tmp
-
-# ---- 依赖分层：基础依赖（变化少，缓存命中率高）----
-COPY requirements-base.txt /tmp/requirements-base.txt
-
-RUN --mount=type=cache,target=/root/.cache/pip,sharing=locked \
-    pip install --no-cache-dir -i https://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com \
-    -r /tmp/requirements-base.txt \
-    && rm /tmp/requirements-base.txt
-
-# ---- 依赖分层：业务依赖（变化频繁）----
-COPY requirements.txt /tmp/requirements.txt
-
-RUN --mount=type=cache,target=/root/.cache/pip,sharing=locked \
-    pip install --no-cache-dir -i https://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com \
-    -r /tmp/requirements.txt \
-    && rm /tmp/requirements.txt
-
-# ---- Python 依赖瘦身 ----
-RUN find /opt/venv -name "*.so" -type f -exec strip --strip-all {} \; 2>/dev/null || true
-RUN find /opt/venv -type d -name "__pycache__" -exec rm -rf {} + 2>/dev/null; \
-    find /opt/venv -name "*.pyc" -delete 2>/dev/null || true
-
-# ==================== Runtime 阶段 ====================
-FROM git.xiaoxiajianji.com/xiaoxia/base/python:3.12-slim AS runtime
+FROM xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com/xiaoxiakeji/saas-api-base:latest
 
 # 构建参数：版本号（CI 传入 commit hash）
 ARG APP_VERSION=dev
 
-# 使用阿里云镜像加速
-RUN sed -i 's|deb.debian.org|mirrors.aliyun.com|g' /etc/apt/sources.list.d/debian.sources 2>/dev/null || \
-    sed -i 's|deb.debian.org|mirrors.aliyun.com|g' /etc/apt/sources.list 2>/dev/null || true
-
-# 只装运行时需要的库（libpq5 是 psycopg2 运行时依赖，ffmpeg 用于封面兜底取帧）
-RUN apt-get update && apt-get install -y --no-install-recommends \
-    libpq5 \
-    ffmpeg \
-    && rm -rf /var/lib/apt/lists/*
-
-# 从 builder 复制虚拟环境
-COPY --from=builder /opt/venv /opt/venv
-
 # 设置工作目录
 WORKDIR /app
 
-# 复制应用代码
-COPY apps/api/ /app/apps/api/
-COPY packages/ /app/packages/
-COPY alembic.ini /app/alembic.ini
-COPY migrations/ /app/migrations/
-COPY alembic/ /app/alembic/
-COPY scripts/ /app/scripts/
+# 复制应用代码（按变化频率从低到高排序，最大化层缓存命中）
+COPY alembic.ini ./alembic.ini
+COPY migrations/ ./migrations/
+COPY alembic/ ./alembic/
+COPY scripts/ ./scripts/
+COPY packages/ ./packages/
+COPY apps/api/ ./apps/api/
 
 # 设置环境变量
 ENV PATH="/opt/venv/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin"
-ENV PYTHONPATH=/app
+ENV PYTHONPATH=/app:/app/apps/api
 ENV PYTHONUNBUFFERED=1
 ENV APP_VERSION=$APP_VERSION
 
@@ -84,5 +31,4 @@ HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
     CMD python -c "import urllib.request; urllib.request.urlopen('http://localhost:8000/health', timeout=5)"
 
 # API 入口点
-WORKDIR /app/apps/api
-CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000"]
+CMD ["uvicorn", "apps.api.main:app", "--host", "0.0.0.0", "--port", "8000"]

infra/docker/entrypoint-worker.sh

@@ -6,8 +6,13 @@ set -e
 
 CONCURRENCY="${WORKER_CONCURRENCY:-2}"
 
+# ⚠️ 部署约束：此 Worker 必须且只能运行单实例（replicas=1）
+# -B 标志嵌入 celery beat，beat 负责定期触发 pending 超时清理等定时任务
+# 多实例部署会导致每个 Worker 独立运行 Beat，造成定时任务重复执行
+# 若需横向扩展 Worker，必须将 Beat 拆分为独立服务（celery beat -A worker_app.celery_app）
 exec celery \
   -A worker_app.celery_app \
   worker \
   --loglevel=info \
+  "-B" \
   "--concurrency=${CONCURRENCY}"

infra/docker/worker-base.Dockerfile

@@ -0,0 +1,49 @@
+# ============================================================
+# Worker 统一基础镜像（预构建）
+# 预装系统依赖 + Python 全部依赖 + CJK 字体
+# 业务构建从此镜像开始，只需 COPY 业务代码，构建时间 < 5 分钟
+# 当 requirements-*.txt 变更时重新构建
+# ============================================================
+
+FROM git.xiaoxiajianji.com/xiaoxia/base/python:3.12-slim
+
+# 使用阿里云镜像加速
+RUN sed -i 's|deb.debian.org|mirrors.aliyun.com|g' /etc/apt/sources.list.d/debian.sources 2>/dev/null || \
+    sed -i 's|deb.debian.org|mirrors.aliyun.com|g' /etc/apt/sources.list 2>/dev/null || true
+
+# 预装系统依赖（编译工具 + 运行时 + CJK 字体用于 ASS 字幕渲染）
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    gcc \
+    g++ \
+    python3-dev \
+    binutils \
+    ffmpeg \
+    libglib2.0-0 \
+    fonts-noto-cjk \
+    && fc-cache -fv \
+    && rm -rf /var/lib/apt/lists/*
+
+# 创建虚拟环境
+RUN python -m venv /opt/venv
+ENV PATH="/opt/venv/bin:$PATH"
+
+WORKDIR /tmp
+
+# 预装全部 Python 依赖（基础 + Worker 大包 + 业务依赖）
+COPY requirements-base.txt requirements.txt requirements-worker.txt ./
+RUN pip install --no-cache-dir \
+    -i https://mirrors.aliyun.com/pypi/simple/ \
+    --trusted-host mirrors.aliyun.com \
+    -r requirements-base.txt \
+    -r requirements-worker.txt \
+    -r requirements.txt
+
+# 虚拟环境瘦身
+RUN find /opt/venv -name "*.so" -type f -exec strip --strip-all {} \; 2>/dev/null || true
+RUN find /opt/venv -type d -name "__pycache__" -exec rm -rf {} + 2>/dev/null; \
+    find /opt/venv -name "*.pyc" -delete 2>/dev/null || true
+
+# 清理临时文件
+RUN rm -f /tmp/requirements-base.txt /tmp/requirements.txt /tmp/requirements-worker.txt
+
+ENV PYTHONPATH=/app:/app/packages

infra/docker/worker.Dockerfile

@@ -1,51 +1,16 @@
 # ============================================================
-# Worker Dockerfile - 分层缓存优化版
-# 优化：基础依赖 + Worker大包预构建为基础镜像，业务构建仅叠加业务依赖
-# 基础镜像：worker-base-builder / worker-base-runtime
+# Worker Dockerfile - 极简化
+# 从预构建统一基础镜像开始，仅叠加业务代码
+# 基础镜像包含：系统依赖 + 全部 Python 依赖 + CJK 字体
+# 构建时间目标：< 5 分钟
 # ============================================================
 
-# ==================== Builder 阶段 ====================
-# 从预构建的builder基础镜像开始，已经包含：
-#   - 编译工具 (gcc/g++/python3-dev/binutils)
-#   - requirements-base.txt 全部依赖
-#   - requirements-worker.txt 全部依赖 (numpy/scipy/opencv)
-#   - 预strip的.so文件
-FROM xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com/xiaoxiakeji/worker-base-builder:latest AS builder
+FROM xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com/xiaoxiakeji/saas-worker-base:latest
 
-ENV PATH="/opt/venv/bin:$PATH"
-
-WORKDIR /tmp
-
-# ---- 安装业务依赖（变化频繁，单独一层）----
-COPY requirements.txt /tmp/requirements.txt
-RUN --mount=type=cache,target=/root/.cache/pip,sharing=locked \
-    pip install --no-cache-dir -i https://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com \
-    -r /tmp/requirements.txt \
-    && rm /tmp/requirements.txt
-
-# ---- 增量瘦身（清理新增业务依赖的冗余文件）----
-RUN find /opt/venv -type d -name "__pycache__" -exec rm -rf {} + 2>/dev/null; \
-    find /opt/venv -name "*.pyc" -delete 2>/dev/null || true
-
-# ==================== Runtime 阶段 ====================
-# 从预构建的runtime基础镜像开始，已经包含：
-#   - ffmpeg
-#   - libglib2.0-0
-FROM xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com/xiaoxiakeji/worker-base-runtime:latest AS runtime
-
-# 构建参数：版本号
+# 构建参数：版本号（CI 传入 commit hash）
 ARG APP_VERSION=dev
 
-# 从 builder 复制 Python 虚拟环境
-COPY --from=builder /opt/venv /opt/venv
-
-# 设置 Python 环境变量
-ENV PATH="/opt/venv/bin:$PATH"
-ENV PYTHONPATH=/app:/app/packages
-ENV PYTHONUNBUFFERED=1
-ENV APP_VERSION=$APP_VERSION
-
-# 创建非 root 用户（极少变化，放最前）
+# 创建非 root 用户
 RUN groupadd -r celery \
     && useradd -r -g celery -d /app -s /sbin/nologin celery \
     && mkdir -p /app/generated \
@@ -53,14 +18,20 @@ RUN groupadd -r celery \
 
 WORKDIR /app
 
-# 复制文件按变化频率从低到高排序，最大化层缓存命中
+# 设置 Python 环境变量
+ENV PATH="/opt/venv/bin:$PATH"
+ENV PYTHONPATH=/app:/app/packages
+ENV PYTHONUNBUFFERED=1
+ENV APP_VERSION=$APP_VERSION
+
+# 复制文件（按变化频率从低到高排序，最大化层缓存命中）
 COPY alembic.ini /app/alembic.ini
 COPY migrations/ /app/migrations/
 COPY packages/ /app/packages/
 COPY apps/api/app/config.py /app/apps/api/app/config.py
 COPY apps/api/app/core/ /app/apps/api/app/core/
 
-# 复制 Worker 启动脚本
+# Worker 启动脚本
 COPY infra/docker/entrypoint-worker.sh /usr/local/bin/entrypoint-worker.sh
 RUN chmod +x /usr/local/bin/entrypoint-worker.sh
 
@@ -68,7 +39,5 @@ RUN chmod +x /usr/local/bin/entrypoint-worker.sh
 COPY apps/worker/ /app/apps/worker/
 
 USER celery
-
-# Worker 入口点
 WORKDIR /app/apps/worker
 CMD ["/usr/local/bin/entrypoint-worker.sh"]

packages/adapters/sqlalchemy_impl/asset_repository.py

@@ -101,6 +101,13 @@ class SQLAlchemyAssetRepository:
             return None
         return self._to_domain(model)
 
+    def find_by_ids(self, asset_ids: list[str]) -> list[Asset]:
+        """批量查询素材（单次 SQL IN 查询，避免 N+1）。"""
+        if not asset_ids:
+            return []
+        models = self.session.query(AssetModel).filter(AssetModel.id.in_(asset_ids)).all()
+        return [self._to_domain(m) for m in models]
+
     def get(self, asset_id: str) -> Asset | None:
         return self.find_by_id(asset_id)
 

packages/adapters/sqlalchemy_impl/generation_task_repository.py

@@ -42,6 +42,7 @@ def _to_domain(model: GenerationTaskModel) -> GenerationTask:
         output_height=getattr(model, "output_height", 720) or 720,
         cover_url=getattr(model, "cover_url", "") or "",
         custom_title=getattr(model, "custom_title", "") or "",
+        title_config=dict(getattr(model, "title_config", {}) or {}),
         logs=model.logs or "[]",
         created_at=model.created_at,
         updated_at=model.updated_at,
@@ -86,6 +87,7 @@ class SQLAlchemyGenerationTaskRepository:
             output_height=task.output_height,
             cover_url=task.cover_url or "",
             custom_title=task.custom_title or "",
+            title_config=dict(task.title_config) if task.title_config else {},
             logs=task.logs,
             created_at=task.created_at,
             updated_at=task.updated_at,
@@ -273,6 +275,7 @@ class SQLAlchemyGenerationTaskRepository:
         model.output_height = task.output_height
         model.cover_url = task.cover_url or ""
         model.custom_title = task.custom_title or ""
+        model.title_config = dict(task.title_config) if task.title_config else {}
         model.logs = task.logs
         self.session.commit()
         return task
@@ -310,3 +313,42 @@ class SQLAlchemyGenerationTaskRepository:
             model.completed_at = datetime.now(timezone.utc)
         self.session.commit()
         return len(models)
+
+    def cleanup_stale_pending(self, timeout_minutes: int = 30) -> int:
+        """清理超时的 pending 任务（未被 Worker 拉取的任务）。
+
+        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。将超时的 pending 任务标记为 failed。
+
+        Args:
+            timeout_minutes: 超时时间（分钟），默认 30 分钟
+
+        Returns:
+            清理的任务数量
+        """
+        from datetime import timedelta
+
+        cutoff = datetime.now(timezone.utc) - timedelta(minutes=timeout_minutes)
+        error_info = {
+            "error_type": "PendingTimeout",
+            "message": f"任务在 pending 状态停留超过 {timeout_minutes} 分钟，自动清理",
+            "failed_at": datetime.now(timezone.utc).isoformat(),
+        }
+        count = (
+            self.session.query(GenerationTaskModel)
+            .filter(
+                GenerationTaskModel.status == GenerationTaskStatus.PENDING.value,
+                GenerationTaskModel.created_at < cutoff,
+            )
+            .update(
+                {
+                    GenerationTaskModel.status: GenerationTaskStatus.FAILED.value,
+                    GenerationTaskModel.error_message: "pending timeout: auto cleanup",
+                    GenerationTaskModel.error_info: error_info,
+                    GenerationTaskModel.completed_at: datetime.now(timezone.utc),
+                },
+                synchronize_session=False,
+            )
+        )
+        self.session.commit()
+        return count

packages/adapters/sqlalchemy_impl/models.py

@@ -298,6 +298,7 @@ class GenerationTaskModel(Base):
     output_height = Column(Integer, nullable=False, default=720)
     cover_url = Column(String(1000), nullable=False, default="")
     custom_title = Column(String(500), nullable=False, default="")
+    title_config = Column(JSON, nullable=False, default=dict)
     bgm_config = Column(JSON, nullable=False, default=dict)
     extra_meta = Column("metadata", JSON, nullable=False, default=dict)
     logs = Column(Text, nullable=False, default="[]", server_default="[]")

packages/application/generation_tasks.py

@@ -69,6 +69,7 @@ class CreateGenerationTaskUseCase:
             output_height=command.output_height,
             cover_url=command.cover_url,
             custom_title=command.custom_title,
+            title_config=command.title_config,
         )
         return self.generation_task_repository.create(task)
 

packages/domain/generation_task.py

@@ -121,6 +121,8 @@ class GenerationTask:
     output_height: int = 720
     cover_url: str = ""
     custom_title: str = ""
+    title_config: dict = field(default_factory=dict)
+    extra_meta: dict = field(default_factory=dict)
     logs: str = "[]"
     created_at: datetime = field(default_factory=lambda: datetime.now(timezone.utc))
     updated_at: datetime = field(default_factory=lambda: datetime.now(timezone.utc))
@@ -152,6 +154,8 @@ class GenerationTask:
         output_height: int = 720,
         cover_url: str = "",
         custom_title: str = "",
+        title_config: dict | None = None,
+        extra_meta: dict | None = None,
     ) -> "GenerationTask":
         if not project_id.strip() and not template_id.strip():
             raise ValueError("project_id 或 template_id 至少需要提供一个")
@@ -182,6 +186,8 @@ class GenerationTask:
             output_height=output_height,
             cover_url=cover_url,
             custom_title=custom_title,
+            title_config=dict(title_config) if title_config else {},
+            extra_meta=dict(extra_meta) if extra_meta else {},
         )
 
     # ── 状态查询 ────────────────────────────────────────────────────────────
@@ -301,6 +307,7 @@ class GenerationTask:
         *,
         cover_url: str = "",
         custom_title: str = "",
+        extra_meta: dict | None = None,
         output_width: int = 0,
         output_height: int = 0,
     ) -> None:
@@ -318,6 +325,8 @@ class GenerationTask:
             self.output_width = output_width
         if output_height > 0:
             self.output_height = output_height
+        if extra_meta:
+            self.extra_meta.update(extra_meta)
         self.updated_at = datetime.now(timezone.utc)
 
     # ── 日志辅助 ────────────────────────────────────────────────────────────

packages/shared/mediakit_client.py

@@ -364,7 +364,7 @@ class MediaKitClient:
                 data = response.json()
 
                 status = data.get("status")
-                if status == "success":
+                if status in ("completed", "success"):
                     result = data.get("result", {})
                     snapshots = result.get("snapshots", [])
                     logger.info(

packages/shared/title_overlay.py

@@ -0,0 +1,153 @@
+"""封面标题文字叠加（Pillow）— API / Worker 共用。
+
+在封面帧上绘制白色标题文字 + 黑色描边/阴影，支持 CJK 字体和自动换行。
+从已渲染视频抽帧时通常不需要调用（标题已烧录）；
+从源素材抽帧（API E2 兜底）时调用，保证封面带标题。
+"""
+
+from __future__ import annotations
+
+import logging
+from pathlib import Path
+from typing import Optional
+
+logger = logging.getLogger(__name__)
+
+# 按优先级查找 CJK 字体（Debian/Ubuntu fonts-noto-cjk 安装路径）
+_FONT_CANDIDATES = (
+    "/usr/share/fonts/opentype/noto/NotoSansCJK-Bold.ttc",
+    "/usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc",
+    "/usr/share/fonts/truetype/noto/NotoSansCJK-Bold.ttc",
+    "/usr/share/fonts/truetype/noto/NotoSansCJK-Regular.ttc",
+    "/usr/share/fonts/truetype/wqy/wqy-zenhei.ttc",
+)
+
+
+def find_title_font(size: int):
+    """查找可用的 CJK 字体并返回 PIL ImageFont，找不到返回 None。"""
+    try:
+        from PIL import ImageFont
+    except ImportError:
+        return None
+
+    for fp in _FONT_CANDIDATES:
+        if Path(fp).exists():
+            try:
+                return ImageFont.truetype(fp, size=size)
+            except Exception:
+                continue
+    logger.warning("未找到 CJK 字体，标题叠加将使用 PIL 默认字体（中文可能显示为方块）")
+    return ImageFont.load_default()
+
+
+def wrap_title_text(text: str, font, max_width: int) -> list[str]:
+    """按像素宽度对中英文混合文本自动换行，支持显式 \\n。"""
+    lines: list[str] = []
+    current = ""
+    for ch in text:
+        if ch == "\n":
+            if current:
+                lines.append(current)
+            current = ""
+            continue
+        trial = current + ch
+        try:
+            bbox = font.getbbox(trial)
+            width = bbox[2] - bbox[0]
+        except Exception:
+            width = len(trial) * (font.size // 2)
+        if width <= max_width:
+            current = trial
+        else:
+            if current:
+                lines.append(current)
+            current = ch
+    if current:
+        lines.append(current)
+    return lines
+
+
+def apply_title_to_image(
+    image_path: str,
+    title_text: str,
+    *,
+    position: str = "bottom",
+    font_size: Optional[int] = None,
+    margin_ratio: float = 0.06,
+    stroke_width_ratio: float = 0.04,
+) -> Optional[str]:
+    """在图片上绘制标题文字并覆盖保存。
+
+    Args:
+        image_path: 图片路径（处理结果覆盖写回）
+        title_text: 标题文字；为空直接返回 None 表示跳过
+        position: top / center / bottom
+        font_size: 字号，None 时按图片宽度自动计算
+        margin_ratio: 边缘留白占短边比例
+        stroke_width_ratio: 描边宽度占字号比例
+
+    Returns:
+        成功返回 image_path；标题为空或 PIL 不可用返回 None。
+    """
+    if not title_text or not title_text.strip():
+        return None
+
+    try:
+        from PIL import Image, ImageDraw
+    except ImportError:
+        logger.warning("Pillow 未安装，跳过标题叠加: image=%s", image_path)
+        return None
+
+    img = Image.open(image_path).convert("RGB")
+    draw = ImageDraw.Draw(img)
+    img_w, img_h = img.size
+
+    if font_size is None:
+        font_size = max(28, min(72, img_w // 16))
+
+    font = find_title_font(font_size)
+    if font is None:
+        return None
+
+    stroke_width = max(2, int(font_size * stroke_width_ratio))
+    margin = int(min(img_w, img_h) * margin_ratio)
+    max_text_width = img_w - 2 * margin
+
+    lines = wrap_title_text(title_text.strip(), font, max_text_width)
+    if not lines:
+        return None
+
+    line_heights = []
+    for ln in lines:
+        bbox = font.getbbox(ln)
+        line_heights.append(bbox[3] - bbox[1])
+    line_height = max(line_heights) if line_heights else font_size
+    line_gap = int(line_height * 0.3)
+    total_height = len(lines) * line_height + (len(lines) - 1) * line_gap
+
+    if position == "top":
+        y_start = margin
+    elif position == "center":
+        y_start = (img_h - total_height) // 2
+    else:
+        y_start = img_h - total_height - margin
+
+    for i, ln in enumerate(lines):
+        bbox = font.getbbox(ln)
+        line_w = bbox[2] - bbox[0]
+        x = (img_w - line_w) // 2
+        y = y_start + i * (line_height + line_gap)
+        # 阴影
+        draw.text((x + 2, y + 2), ln, font=font, fill=(0, 0, 0))
+        # 白色文字 + 黑色描边
+        draw.text(
+            (x, y),
+            ln,
+            font=font,
+            fill=(255, 255, 255),
+            stroke_width=stroke_width,
+            stroke_fill=(0, 0, 0),
+        )
+
+    img.save(image_path, "JPEG", quality=92)
+    return image_path

requirements-base.txt

@@ -2,7 +2,6 @@
 # 修改此文件会触发完整重新构建，请谨慎修改
 
 # 数据库（基础层）
-psycopg2-binary==2.9.9
 psycopg[binary]==3.2.2
 sqlalchemy==2.0.35
 alembic==1.13.3
@@ -30,3 +29,4 @@ httpx==0.27.2
 
 # Prometheus monitoring
 prometheus-client==0.21.1
+Pillow==10.4.0

scripts/ci/docker_build_push.sh

@@ -3,6 +3,9 @@
 # 用法: docker_build_push.sh [--no-cache] <Dockerfile> <image_tag> <cache_ref> [build_arg...]
 set -eu
 
+# 单次 build 超时时间（秒），防止 docker buildx build 无限挂起
+BUILD_TIMEOUT=1500
+
 NO_CACHE_FLAG=""
 if [ "$1" = "--no-cache" ]; then
   NO_CACHE_FLAG="--no-cache"
@@ -43,7 +46,7 @@ build_with_cache_retry() {
     local build_output
     local exit_code
     set +e
-    build_output=$(docker buildx build \
+    build_output=$(timeout ${BUILD_TIMEOUT} docker buildx build \
       $NO_CACHE_FLAG \
       $BUILD_ARGS \
       --cache-from "type=local,src=${LOCAL_CACHE_DIR}" \
@@ -60,6 +63,12 @@ build_with_cache_retry() {
       echo "$build_output"
       return 0
     fi
+    # 超时退出（exit code 124）
+    if [ $exit_code -eq 124 ]; then
+      echo "❌ Docker build TIMEOUT after ${BUILD_TIMEOUT}s - build hung and was killed"
+      echo "$build_output" | tail -20
+      return $exit_code
+    fi
     # 检测到缓存损坏类错误，清掉本地缓存重试
     if echo "$build_output" | grep -qE "parent snapshot.*not found|snapshot.*does not exist|cache.*corrupt|failed to compute cache key"; then
       echo "$build_output"
@@ -68,7 +77,7 @@ build_with_cache_retry() {
       rm -rf "${LOCAL_CACHE_DIR}"
       mkdir -p "${LOCAL_CACHE_DIR}"
       # 清理buildx builder的内部snapshot状态
-      docker buildx prune -f -a >/dev/null 2>&1 || true
+      docker buildx prune -f -a > /dev/null 2>&1 || true
       attempt=$((attempt + 1))
     else
       # 非缓存类错误，直接输出并返回
@@ -78,7 +87,7 @@ build_with_cache_retry() {
   done
   # 重试完还是失败，不用本地缓存最后试一次（只从registry读）
   echo "⚠️  All cached attempts failed, building without local cache..."
-  docker buildx build \
+  timeout ${BUILD_TIMEOUT} docker buildx build \
     $NO_CACHE_FLAG \
     $BUILD_ARGS \
     --cache-from "type=registry,ref=${CACHE_REF}" \
@@ -93,6 +102,7 @@ build_with_cache_retry() {
 echo "=== Step 1: Build & push image (local cache + registry cache, with auto-repair) ==="
 echo "Local cache: ${LOCAL_CACHE_DIR}"
 echo "Registry cache: ${CACHE_REF}"
+echo "Build timeout: ${BUILD_TIMEOUT}s"
 echo ""
 
 build_with_cache_retry

scripts/ci/rebuild-api-base-image.sh

@@ -0,0 +1,40 @@
+#!/bin/bash
+# ============================================================
+# 重建 API 基础镜像脚本
+# 用途：当 requirements-base.txt 或 requirements.txt 变更时手动触发
+# 前提：需要在已登录 ACR 的构建服务器上执行
+# ============================================================
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+REPO_ROOT="$(cd "$SCRIPT_DIR/../.." && pwd)"
+
+REGISTRY="xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com/xiaoxiakeji"
+IMAGE_NAME="saas-api-base"
+TAG="latest"
+FULL_TAG="${REGISTRY}/${IMAGE_NAME}:${TAG}"
+
+echo "========================================="
+echo "🔨 Rebuilding API base image"
+echo "   Registry: ${REGISTRY}"
+echo "   Image:    ${FULL_TAG}"
+echo "   Context:  ${REPO_ROOT}"
+echo "========================================="
+
+cd "$REPO_ROOT"
+
+# 构建并推送
+docker buildx build \
+    --platform linux/amd64 \
+    --tag "${FULL_TAG}" \
+    --push \
+    -f infra/docker/api-base.Dockerfile \
+    .
+
+echo ""
+echo "✅ API base image pushed: ${FULL_TAG}"
+
+# 显示镜像大小
+docker pull "${FULL_TAG}" > /dev/null 2>&1
+docker images "${FULL_TAG}" --format "table {{.Repository}}:{{.Tag}}\t{{.Size}}"

tests/integration/test_assets_api.py

@@ -260,6 +260,30 @@ def _make_user(**overrides) -> User:
     return User(**defaults)
 
 
+def _direct_insert_asset(client, name="test-video.mp4", storage_key=None, mime_type="video/mp4", status=None):
+    """Helper: insert asset directly into repo (bypass deprecated create_asset API)."""
+    import uuid as _uuid
+
+    app = client.app
+    asset_repo = app.dependency_overrides[get_asset_repository]()
+    kw = {}
+    if status is not None:
+        kw["status"] = status
+    else:
+        kw["status"] = AssetStatus.READY
+    asset = Asset(
+        id=_uuid.uuid4().hex,
+        project_id="proj-1",
+        library_id="lib-1",
+        name=name,
+        storage_key=storage_key or f"uploads/{name}",
+        mime_type=mime_type,
+        **kw,
+    )
+    asset_repo.create(asset)
+    return asset.id
+
+
 def _make_project(id: str = "proj-1", owner_user_id: str = "user-test-001") -> Project:
     return Project(id=id, name="Test Project", owner_user_id=owner_user_id)
 
@@ -344,8 +368,8 @@ def client(mock_storage):
 class TestCreateAsset:
     """创建素材端点测试。"""
 
-    def test_create_asset_success(self, client):
-        """正常创建素材成功。"""
+    def test_create_asset_returns_410_gone(self, client):
+        """create_asset 已废弃，返回 410 Gone 提示使用 ingest-jobs。"""
         resp = client.post(
             "/api/v1/assets",
             json={
@@ -354,59 +378,23 @@ class TestCreateAsset:
                 "name": "new-video.mp4",
                 "storage_key": "uploads/new-video.mp4",
                 "mime_type": "video/mp4",
-                "file_size": 2048,
-                "duration": 15.0,
             },
         )
-        assert resp.status_code == 200
-        data = resp.json()
-        assert data["name"] == "new-video.mp4"
-        assert data["project_id"] == "proj-1"
-        assert data["library_id"] == "lib-1"
-        assert data["mime_type"] == "video/mp4"
-        assert "id" in data
-        assert data["status"] == "uploading"
+        assert resp.status_code == 410
 
-    def test_create_asset_project_not_found(self, client):
-        """项目不存在返回 404。"""
+    def test_create_asset_any_type_returns_410(self, client):
+        """所有类型都返回 410 Gone（图片/音频也废弃）。"""
         resp = client.post(
             "/api/v1/assets",
             json={
-                "project_id": "nonexistent",
+                "project_id": "proj-1",
                 "library_id": "lib-1",
-                "name": "test.mp4",
-                "storage_key": "uploads/test.mp4",
-                "mime_type": "video/mp4",
+                "name": "photo.jpg",
+                "storage_key": "uploads/photo.jpg",
+                "mime_type": "image/jpeg",
             },
         )
-        assert resp.status_code == 404
-        assert "Project" in resp.json()["detail"]
-
-    def test_create_asset_library_not_found(self, client):
-        """素材库不存在返回 404。"""
-        resp = client.post(
-            "/api/v1/assets",
-            json={
-                "project_id": "proj-1",
-                "library_id": "nonexistent",
-                "name": "test.mp4",
-                "storage_key": "uploads/test.mp4",
-                "mime_type": "video/mp4",
-            },
-        )
-        assert resp.status_code == 404
-        assert "AssetLibrary" in resp.json()["detail"]
-
-    def test_create_asset_missing_required_fields(self, client):
-        """缺少必填字段返回 422。"""
-        resp = client.post(
-            "/api/v1/assets",
-            json={
-                "project_id": "proj-1",
-                "name": "test.mp4",
-            },
-        )
-        assert resp.status_code == 422
+        assert resp.status_code == 410
 
 
 # ---------------------------------------------------------------------------
@@ -418,20 +406,26 @@ class TestListAssets:
     """获取素材列表端点测试。"""
 
     def _create_test_assets(self, client, count: int = 3):
-        """辅助方法：创建测试素材（status=ready）。"""
+        """辅助方法：直接插入测试素材到 repository（绕过已废弃的 create_asset API）。"""
+        # 通过依赖覆盖获取 asset_repo
+        app = client.app
+        asset_repo = app.dependency_overrides.get(get_asset_repository, lambda: None)()
+        if asset_repo is None:
+            return
         for i in range(count):
-            client.post(
-                "/api/v1/assets",
-                json={
-                    "project_id": "proj-1",
-                    "library_id": "lib-1",
-                    "name": f"video-{i}.mp4",
-                    "storage_key": f"uploads/video-{i}.mp4",
-                    "mime_type": "video/mp4",
-                    "file_size": 1024 * (i + 1),
-                    "status": "ready",
-                },
+            import uuid
+
+            asset = Asset(
+                id=uuid.uuid4().hex,
+                project_id="proj-1",
+                library_id="lib-1",
+                name=f"video-{i}.mp4",
+                storage_key=f"uploads/video-{i}.mp4",
+                mime_type="video/mp4",
+                file_size=1024 * (i + 1),
+                status=AssetStatus.READY,
             )
+            asset_repo.create(asset)
 
     def test_empty_list(self, client):
         """无素材时返回空列表。"""
@@ -517,17 +511,7 @@ class TestListAssets:
 
     def test_list_status_filter_uploading_visible(self, client):
         """uploading状态的素材默认能看到（上传后立即显示处理中）。"""
-        client.post(
-            "/api/v1/assets",
-            json={
-                "project_id": "proj-1",
-                "library_id": "lib-1",
-                "name": "uploading-test.mp4",
-                "storage_key": "uploads/uploading-test.mp4",
-                "mime_type": "video/mp4",
-                "status": "uploading",
-            },
-        )
+        _direct_insert_asset(client, name="uploading-test.mp4", status=AssetStatus.UPLOADING)
 
         resp = client.get("/api/v1/assets?library_id=lib-1")
         assert resp.status_code == 200
@@ -537,28 +521,8 @@ class TestListAssets:
 
     def test_list_with_keyword_filter(self, client):
         """按名称关键词过滤。"""
-        client.post(
-            "/api/v1/assets",
-            json={
-                "project_id": "proj-1",
-                "library_id": "lib-1",
-                "name": "hello-world.mp4",
-                "storage_key": "uploads/hello.mp4",
-                "mime_type": "video/mp4",
-                "status": "ready",
-            },
-        )
-        client.post(
-            "/api/v1/assets",
-            json={
-                "project_id": "proj-1",
-                "library_id": "lib-1",
-                "name": "goodbye.mp4",
-                "storage_key": "uploads/goodbye.mp4",
-                "mime_type": "video/mp4",
-                "status": "ready",
-            },
-        )
+        _direct_insert_asset(client, name="hello-world.mp4")
+        _direct_insert_asset(client, name="goodbye.mp4", mime_type="video/mp4", status=AssetStatus.READY)
 
         resp = client.get("/api/v1/assets?library_id=lib-1&keyword=hello")
         assert resp.status_code == 200
@@ -576,22 +540,8 @@ class TestGetAsset:
     """获取单个素材详情端点测试。"""
 
     def _create_asset(self, client) -> str:
-        resp = client.post(
-            "/api/v1/assets",
-            json={
-                "project_id": "proj-1",
-                "library_id": "lib-1",
-                "name": "detail-test.mp4",
-                "storage_key": "uploads/detail-test.mp4",
-                "mime_type": "video/mp4",
-                "file_size": 5000,
-                "duration": 25.0,
-                "width": 1280,
-                "height": 720,
-                "fps": 30.0,
-            },
-        )
-        return resp.json()["id"]
+        """Direct insert into repo (create_asset API is deprecated/410)."""
+        return _direct_insert_asset(client)
 
     def test_get_asset_success(self, client):
         """获取存在的素材详情成功。"""
@@ -601,11 +551,7 @@ class TestGetAsset:
         assert resp.status_code == 200
         data = resp.json()
         assert data["id"] == asset_id
-        assert data["name"] == "detail-test.mp4"
-        assert data["file_size"] == 5000
-        assert data["duration"] == 25.0
-        assert data["width"] == 1280
-        assert data["height"] == 720
+        assert data["name"] == "test-video.mp4"
         assert "file_url" in data
         assert "status" in data
 
@@ -625,17 +571,8 @@ class TestUpdateAsset:
     """更新素材端点测试。"""
 
     def _create_asset(self, client) -> str:
-        resp = client.post(
-            "/api/v1/assets",
-            json={
-                "project_id": "proj-1",
-                "library_id": "lib-1",
-                "name": "old-name.mp4",
-                "storage_key": "uploads/old-name.mp4",
-                "mime_type": "video/mp4",
-            },
-        )
-        return resp.json()["id"]
+        """Direct insert into repo (create_asset API is deprecated/410)."""
+        return _direct_insert_asset(client)
 
     def test_update_asset_name(self, client):
         """更新素材名称成功。"""
@@ -675,7 +612,7 @@ class TestUpdateAsset:
 
         resp = client.put(f"/api/v1/assets/{asset_id}", json={})
         assert resp.status_code == 200
-        assert resp.json()["name"] == "old-name.mp4"
+        assert resp.json()["name"] == "test-video.mp4"
 
 
 # ---------------------------------------------------------------------------
@@ -687,17 +624,8 @@ class TestDeleteAsset:
     """删除素材端点测试。"""
 
     def _create_asset(self, client) -> str:
-        resp = client.post(
-            "/api/v1/assets",
-            json={
-                "project_id": "proj-1",
-                "library_id": "lib-1",
-                "name": "delete-test.mp4",
-                "storage_key": "uploads/delete-test.mp4",
-                "mime_type": "video/mp4",
-            },
-        )
-        return resp.json()["id"]
+        """Direct insert into repo (create_asset API is deprecated/410)."""
+        return _direct_insert_asset(client)
 
     def test_delete_asset_success(self, client):
         """删除存在的素材成功，返回 204。"""
@@ -737,17 +665,8 @@ class TestBatchDeleteAssets:
     def _create_assets(self, client, count: int = 3) -> list[str]:
         ids = []
         for i in range(count):
-            resp = client.post(
-                "/api/v1/assets",
-                json={
-                    "project_id": "proj-1",
-                    "library_id": "lib-1",
-                    "name": f"batch-{i}.mp4",
-                    "storage_key": f"uploads/batch-{i}.mp4",
-                    "mime_type": "video/mp4",
-                },
-            )
-            ids.append(resp.json()["id"])
+            aid = _direct_insert_asset(client, name=f"batch-{i}.mp4")
+            ids.append(aid)
         return ids
 
     def test_batch_delete_success(self, client):
@@ -802,17 +721,8 @@ class TestAssetTags:
     """素材标签相关端点测试。"""
 
     def _create_asset(self, client) -> str:
-        resp = client.post(
-            "/api/v1/assets",
-            json={
-                "project_id": "proj-1",
-                "library_id": "lib-1",
-                "name": "tag-test.mp4",
-                "storage_key": "uploads/tag-test.mp4",
-                "mime_type": "video/mp4",
-            },
-        )
-        return resp.json()["id"]
+        """Direct insert into repo (create_asset API is deprecated/410)."""
+        return _direct_insert_asset(client)
 
     def test_add_tags_to_asset(self, client):
         """给素材打标签。需要先在 tag_repo 中创建标签。"""
@@ -847,22 +757,8 @@ class TestAssetsCRUDFlow:
 
     def test_full_crud_flow(self, client):
         """测试完整的创建 → 列表 → 详情 → 更新 → 删除流程。"""
-        # 1. 创建
-        create_resp = client.post(
-            "/api/v1/assets",
-            json={
-                "project_id": "proj-1",
-                "library_id": "lib-1",
-                "name": "crud-flow.mp4",
-                "storage_key": "uploads/crud-flow.mp4",
-                "mime_type": "video/mp4",
-                "file_size": 8192,
-                "metadata": {"source": "test"},
-                "status": "ready",
-            },
-        )
-        assert create_resp.status_code == 200
-        asset_id = create_resp.json()["id"]
+        # 1. 创建 (direct insert since create_asset is 410)
+        asset_id = _direct_insert_asset(client, name="crud-flow.mp4")
 
         # 2. 列表中应包含
         list_resp = client.get("/api/v1/assets?library_id=lib-1")

tests/integration/test_full_pipeline.py

@@ -14,11 +14,7 @@ import pytest
 from video_processing.unified_render_service import (
     UnifiedRenderService,
 )
-from worker_app.tasks.generation import (
-    _build_plan_and_clips_from_task,
-    _create_fallback_clip,
-    _mux_audio_track,
-)
+from worker_app.tasks.generation import _build_plan_and_clips_from_task
 
 pytestmark = pytest.mark.skipif(
     not shutil.which("ffmpeg"),
@@ -62,61 +58,6 @@ def _generate_test_audio(path: Path, duration: float = 5.0) -> None:
     subprocess.run(cmd, check=True, capture_output=True, timeout=30)
 
 
-# ── 测试 _create_fallback_clip ────────────────────────────────────────────────
-
-
-class TestFallbackClip:
-    """测试 fallback 视频生成。"""
-
-    def test_fallback_clip_creates_video(self):
-        with tempfile.TemporaryDirectory() as tmpdir:
-            output = Path(tmpdir) / "fallback.mp4"
-            _create_fallback_clip(output, "Test Fallback")
-
-            assert output.exists()
-            assert output.stat().st_size > 0
-
-
-# ── 测试 _mux_audio_track ────────────────────────────────────────────────────
-
-
-class TestMuxAudioTrack:
-    """测试视频+音频混合。"""
-
-    def test_mux_audio_into_video(self):
-        with tempfile.TemporaryDirectory() as tmpdir:
-            video_path = Path(tmpdir) / "video.mp4"
-            audio_path = Path(tmpdir) / "audio.aac"
-            output_path = Path(tmpdir) / "output.mp4"
-
-            _generate_test_video(video_path, duration=3.0)
-            _generate_test_audio(audio_path, duration=5.0)
-
-            _mux_audio_track(video_path, str(audio_path), output_path)
-
-            assert output_path.exists()
-            assert output_path.stat().st_size > 0
-
-            # 验证输出文件包含音频轨
-            probe_cmd = [
-                "ffprobe",
-                "-v",
-                "quiet",
-                "-show_streams",
-                "-select_streams",
-                "a",
-                "-of",
-                "csv=p=0",
-                str(output_path),
-            ]
-            result = subprocess.run(probe_cmd, capture_output=True, text=True, timeout=10)
-            # 如果有音频流，输出非空
-            assert result.stdout.strip() != "" or result.returncode == 0
-
-
-# ── 测试 PlanGenerator → UnifiedRenderService 全链路 ─────────────────────────
-
-
 class TestFullPipeline:
     """验证从虚拟 plan 构建到渲染输出的完整流程。"""
 
@@ -175,12 +116,26 @@ class TestFullPipeline:
             )
             render_result = service.render()
 
-            # 混音
+            # 混音 - 直接用 ffmpeg（_mux_audio_track 已被清理）
             audio_path = work_dir / "voice.aac"
             _generate_test_audio(audio_path, duration=5.0)
 
             final_path = work_dir / "final.mp4"
-            _mux_audio_track(render_result.output_path, str(audio_path), final_path)
+            mux_cmd = [
+                "ffmpeg",
+                "-y",
+                "-i",
+                str(render_result.output_path),
+                "-i",
+                str(audio_path),
+                "-c:v",
+                "copy",
+                "-c:a",
+                "aac",
+                "-shortest",
+                str(final_path),
+            ]
+            subprocess.run(mux_cmd, check=True, capture_output=True, timeout=30)
 
             assert final_path.exists()
             assert final_path.stat().st_size > 0

tests/unit/test_1294_preview_voice_injection.py

@@ -158,7 +158,7 @@ class TestRenderVideoVoiceInjection:
 
             from packages.domain import EditingMode
 
-            output_path, render_duration = _render_video(
+            output_path, render_duration, cover_candidates = _render_video(
                 task_id="test_task_123",
                 downloaded_videos=[Path("/tmp/video1.mp4")],
                 voice_path=None,

tests/unit/test_asset_dedup.py

@@ -147,6 +147,7 @@ def _build_app(
     storage._normalize_storage_key = lambda key: key
     storage.file_exists = lambda key: True
     storage.upload_file = MagicMock(return_value="https://oss.example.com/file.mp4")
+    storage.get_url = MagicMock(return_value="https://oss.example.com/file.mp4")
 
     mock_user = MagicMock(spec=AuthenticatedUser)
     mock_user.id = "user-1"

tests/unit/test_batch_download.py

@@ -108,21 +108,30 @@ def _run_with_fakes(
         session = MagicMock()
         session_maker = MagicMock(return_value=session)
 
-    captured = {"upload_calls": [], "session": session_maker()}
+    captured: dict = {"upload_calls": []}
 
     def _tracking_upload(local_path, storage_key):
         captured["upload_calls"].append((local_path, storage_key))
         return upload_fn(local_path, storage_key)
 
+    # Wrap session_maker to capture the session INSIDE the patch context.
+    # This avoids stale PromiseProxy cache issues in full-suite runs.
+    _created_sessions: list = []
+    _orig_sm = session_maker
+
+    def _tracking_sm(*a, **kw):
+        s = _orig_sm(*a, **kw)
+        _created_sessions.append(s)
+        return s
+
     bound_task = _make_bound_task()
 
-    with patch(
-        "packages.adapters.sqlalchemy_impl.generated_video_repository.SQLAlchemyGeneratedVideoRepository",
-        return_value=repo,
-    ):
+    import worker_app.db as _db_mod
+
+    with patch.object(_db_mod, "SessionLocal", _tracking_sm):
         with patch(
-            "worker_app.db.SessionLocal",
-            session_maker,
+            "packages.adapters.sqlalchemy_impl.generated_video_repository.SQLAlchemyGeneratedVideoRepository",
+            return_value=repo,
         ):
             with patch(
                 "video_processing.oss_helpers.upload_to_oss",
@@ -134,6 +143,7 @@ def _run_with_fakes(
                 ):
                     result = _call_task(batch_download_videos, bound_task, [v.id for v in videos], user_id)
 
+    captured["session"] = _created_sessions[0] if _created_sessions else None
     captured["result"] = result
     return captured
 
@@ -244,19 +254,66 @@ def test_batch_download_single_video():
 
 
 def test_batch_download_session_closed():
-    """DB session is always closed (via finally block)."""
+    """DB session is always closed (via finally block).
+
+    Patches the function's own globals to inject mock SessionLocal,
+    bypassing any import caching issues in the full suite.
+    """
+    import sys
+
+    import worker_app.db as _db_mod
+
+    from apps.worker.worker_app.tasks.batch_download import batch_download_videos
+
     videos = [_FakeVideo("v1", "v.mp4")]
+    repo = _FakeGeneratedVideoRepository(videos)
 
     session = MagicMock()
-    session_maker = MagicMock(return_value=session)
+    mock_session_factory = MagicMock(return_value=session)
 
-    _run_with_fakes(videos, session_maker=session_maker)
+    def _noop_download(url, dest):
+        Path(dest).parent.mkdir(parents=True, exist_ok=True)
+        Path(dest).write_bytes(b"fake video data")
 
+    bound_task = _make_bound_task()
+
+    # Get the raw function to patch its globals
+    raw_fn = _get_raw_task_fn(batch_download_videos)
+
+    # Patch SessionLocal in ALL possible module locations
+    _db_mod.SessionLocal = mock_session_factory
+    if "worker_app.db" in sys.modules:
+        sys.modules["worker_app.db"].SessionLocal = mock_session_factory
+
+    # Also patch in the function's own globals if it has a reference there
+    if "SessionLocal" in raw_fn.__globals__:
+        raw_fn.__globals__["SessionLocal"] = mock_session_factory
+
+    try:
+        with patch(
+            "packages.adapters.sqlalchemy_impl.generated_video_repository.SQLAlchemyGeneratedVideoRepository",
+            return_value=repo,
+        ):
+            with patch("video_processing.oss_helpers.upload_to_oss", return_value="https://oss.example.com/zip"):
+                with patch(
+                    "apps.worker.worker_app.tasks.batch_download._download_video_to_file",
+                    _noop_download,
+                ):
+                    raw_fn(bound_task, ["v1"], "user_1")
+    finally:
+        pass  # Don't restore - other tests handle their own patches
+
+    # Diagnostic: check if our mock factory was actually called
+    assert mock_session_factory.called, "SessionLocal mock was never called! " f"raw_fn={raw_fn}, type={type(raw_fn)}"
     session.close.assert_called_once()
 
 
 def test_batch_download_closes_session_on_error():
     """Session is closed even when get_by_ids raises."""
+    import sys
+
+    import worker_app.db as _db_mod
+
     from apps.worker.worker_app.tasks.batch_download import batch_download_videos
 
     class _ExplodingRepo:
@@ -264,17 +321,25 @@ def test_batch_download_closes_session_on_error():
             raise RuntimeError("db down")
 
     session = MagicMock()
-    session_maker = MagicMock(return_value=session)
+    mock_session_factory = MagicMock(return_value=session)
     bound_task = _make_bound_task()
 
+    raw_fn = _get_raw_task_fn(batch_download_videos)
+
+    _db_mod.SessionLocal = mock_session_factory
+    if "worker_app.db" in sys.modules:
+        sys.modules["worker_app.db"].SessionLocal = mock_session_factory
+    if "SessionLocal" in raw_fn.__globals__:
+        raw_fn.__globals__["SessionLocal"] = mock_session_factory
+
     with patch(
         "packages.adapters.sqlalchemy_impl.generated_video_repository.SQLAlchemyGeneratedVideoRepository",
         return_value=_ExplodingRepo(),
     ):
-        with patch("worker_app.db.SessionLocal", session_maker):
-            with pytest.raises(RuntimeError, match="db down"):
-                _call_task(batch_download_videos, bound_task, ["v1"], "u")
+        with pytest.raises(RuntimeError, match="db down"):
+            raw_fn(bound_task, ["v1"], "u")
 
+    assert mock_session_factory.called, "SessionLocal mock was never called!"
     session.close.assert_called_once()
 
 

tests/unit/test_clips_api_response_structure.py

@@ -0,0 +1,496 @@
+"""片段管理路由 clips.py 增量覆盖率测试.
+
+覆盖 PR fix/clips-api-response-structure 新增代码:
+- _clip_to_response: 枚举转换、日期格式化、asset_url 参数
+- _build_asset_url_map: 批量素材 URL 解析（空列表/异常/正常路径）
+- 路由层 asset_repo 注入与 URL 拼接逻辑
+"""
+
+from __future__ import annotations
+
+import os
+import sys
+from enum import Enum
+from pathlib import Path
+from unittest.mock import MagicMock, patch
+
+os.environ.setdefault("JWT_SECRET_KEY", "unit-test-secret-key-for-testing")
+os.environ.setdefault("DATABASE_URL", "sqlite:///test.db")
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[2] / "apps" / "api"))
+
+import pytest
+from fastapi import FastAPI
+from fastapi.testclient import TestClient
+
+# ---------------------------------------------------------------------------
+# 常量与工厂
+# ---------------------------------------------------------------------------
+
+TEST_TEMPLATE_ID = "tmpl-test-001"
+TEST_PLAN_ID = "plan-draft-001"
+TEST_USER_ID = "user-001"
+
+
+def _auth_user():
+    u = MagicMock()
+    u.user.id = TEST_USER_ID
+    u.user_id = TEST_USER_ID
+    return u
+
+
+def _clip(**overrides):
+    """构造 mock clip，支持 Enum 类型字段"""
+    c = MagicMock()
+    c.id = overrides.get("id", "clip-001")
+    c.plan_id = overrides.get("plan_id", TEST_PLAN_ID)
+    c.clip_type = overrides.get("clip_type", "video")
+    c.order = overrides.get("order", 0)
+    c.duration = overrides.get("duration", 10.0)
+    c.start_time = overrides.get("start_time", 0.0)
+    c.text_content = overrides.get("text_content", "")
+    c.transition_effect = overrides.get("transition_effect", "cut")
+    c.transition_duration = overrides.get("transition_duration", 0.0)
+    c.playback_speed = overrides.get("playback_speed", 1.0)
+    c.asset_id = overrides.get("asset_id", "")
+    c.status = overrides.get("status", "ready")
+    c.template_clip_config_id = overrides.get("template_clip_config_id", "")
+    c.config = overrides.get("config", {})
+    c.created_at = overrides.get("created_at", None)
+    c.updated_at = overrides.get("updated_at", None)
+    return c
+
+
+def _services(plan_svc_overrides=None):
+    tpl = MagicMock()
+    plan = MagicMock()
+    if plan_svc_overrides:
+        for k, v in plan_svc_overrides.items():
+            setattr(plan, k, v)
+    return tpl, plan
+
+
+# ---------------------------------------------------------------------------
+# 单元测试: _clip_to_response
+# ---------------------------------------------------------------------------
+
+
+class TestClipToResponse:
+    """_clip_to_response 纯函数测试 — 覆盖行 53-80"""
+
+    def test_basic_fields(self):
+        from app.api.routes.templates_editor.clips import _clip_to_response
+
+        c = _clip(id="c1", order=3, duration=5.5, text_content="hello")
+        resp = _clip_to_response(c)
+        assert resp.id == "c1"
+        assert resp.order == 3
+        assert resp.duration == 5.5
+        assert resp.text_content == "hello"
+        assert resp.asset_url is None
+
+    def test_enum_clip_type(self):
+        """Enum 值应被 .value 解包"""
+        from app.api.routes.templates_editor.clips import _clip_to_response
+
+        class ClipType(str, Enum):
+            VIDEO = "video"
+            AUDIO = "audio"
+
+        c = _clip(clip_type=ClipType.VIDEO)
+        resp = _clip_to_response(c)
+        assert resp.clip_type == "video"
+
+    def test_plain_string_clip_type(self):
+        """非 Enum 字符串直接用 str()"""
+        from app.api.routes.templates_editor.clips import _clip_to_response
+
+        c = _clip(clip_type="main")
+        resp = _clip_to_response(c)
+        assert resp.clip_type == "main"
+
+    def test_enum_transition_effect(self):
+        from app.api.routes.templates_editor.clips import _clip_to_response
+
+        class Transition(str, Enum):
+            FADE = "fade"
+
+        c = _clip(transition_effect=Transition.FADE)
+        resp = _clip_to_response(c)
+        assert resp.transition_effect == "fade"
+
+    def test_default_transition_when_none(self):
+        """transition_effect 缺失时默认 cut"""
+        from app.api.routes.templates_editor.clips import _clip_to_response
+
+        c = _clip()
+        del c.transition_effect  # 触发 getattr default
+        resp = _clip_to_response(c)
+        assert resp.transition_effect == "cut"
+
+    def test_asset_url_passed(self):
+        """asset_url 参数应透传到响应"""
+        from app.api.routes.templates_editor.clips import _clip_to_response
+
+        c = _clip(asset_id="a1")
+        resp = _clip_to_response(c, asset_url="https://signed-url.example.com/video.mp4")
+        assert resp.asset_url == "https://signed-url.example.com/video.mp4"
+
+    def test_asset_url_none_by_default(self):
+        from app.api.routes.templates_editor.clips import _clip_to_response
+
+        c = _clip()
+        resp = _clip_to_response(c)
+        assert resp.asset_url is None
+
+    def test_datetime_isoformat(self):
+        """datetime 对象应被 isoformat()"""
+        from datetime import datetime
+
+        from app.api.routes.templates_editor.clips import _clip_to_response
+
+        dt = datetime(2026, 8, 17, 12, 0, 0)
+        c = _clip(created_at=dt, updated_at=dt)
+        resp = _clip_to_response(c)
+        assert "2026-08-17" in resp.created_at
+        assert "2026-08-17" in resp.updated_at
+
+    def test_none_datetime_empty_string(self):
+        """None 日期应格式化为空字符串"""
+        from app.api.routes.templates_editor.clips import _clip_to_response
+
+        c = _clip(created_at=None, updated_at=None)
+        resp = _clip_to_response(c)
+        assert resp.created_at == ""
+        assert resp.updated_at == ""
+
+    def test_string_datetime_passthrough(self):
+        """已经是字符串的日期直接 str()"""
+        from app.api.routes.templates_editor.clips import _clip_to_response
+
+        c = _clip(created_at="2026-08-17T00:00:00")
+        resp = _clip_to_response(c)
+        assert resp.created_at == "2026-08-17T00:00:00"
+
+    def test_none_defaults_for_optional_fields(self):
+        """None/缺失字段的默认值"""
+        from app.api.routes.templates_editor.clips import _clip_to_response
+
+        c = _clip(asset_id=None, status=None, template_clip_config_id=None)
+        resp = _clip_to_response(c)
+        assert resp.asset_id == ""
+        assert resp.status == "pending"
+        assert resp.template_clip_config_id == ""
+
+    def test_zero_duration_fallback(self):
+        """duration=0 → playback_speed 默认 1.0"""
+        from app.api.routes.templates_editor.clips import _clip_to_response
+
+        c = _clip(playback_speed=None)
+        resp = _clip_to_response(c)
+        assert resp.playback_speed == 1.0
+
+
+# ---------------------------------------------------------------------------
+# 单元测试: _build_asset_url_map
+# ---------------------------------------------------------------------------
+
+
+class TestBuildAssetUrlMap:
+    """_build_asset_url_map 测试 — 覆盖行 93-118"""
+
+    def test_empty_list(self):
+        """空 asset_ids 直接返回空 dict"""
+        from app.api.routes.templates_editor.clips import _build_asset_url_map
+
+        repo = MagicMock()
+        result = _build_asset_url_map([], repo)
+        assert result == {}
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_storage_service_failure(self, mock_get_storage):
+        """存储服务获取失败时返回全 None"""
+        from app.api.routes.templates_editor.clips import _build_asset_url_map
+
+        mock_get_storage.side_effect = RuntimeError("storage unavailable")
+        repo = MagicMock()
+        result = _build_asset_url_map(["a1", "a2"], repo)
+        assert result == {"a1": None, "a2": None}
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_asset_not_found(self, mock_get_storage):
+        """asset_id 找不到对应素材 → None"""
+        from app.api.routes.templates_editor.clips import _build_asset_url_map
+
+        storage = MagicMock()
+        mock_get_storage.return_value = storage
+        repo = MagicMock()
+        repo.find_by_ids.return_value = []
+
+        result = _build_asset_url_map(["missing-id"], repo)
+        assert result == {"missing-id": None}
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_no_storage_key(self, mock_get_storage):
+        """素材没有 storage_key → None"""
+        from app.api.routes.templates_editor.clips import _build_asset_url_map
+
+        storage = MagicMock()
+        mock_get_storage.return_value = storage
+        repo = MagicMock()
+        asset = MagicMock()
+        asset.id = "a1"
+        asset.storage_key = ""
+        repo.find_by_ids.return_value = [asset]
+
+        result = _build_asset_url_map(["a1"], repo)
+        assert result == {"a1": None}
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_successful_url_generation(self, mock_get_storage):
+        """正常路径：返回签名 URL"""
+        from app.api.routes.templates_editor.clips import _build_asset_url_map
+
+        storage = MagicMock()
+        storage.get_download_url.return_value = "https://cdn.example.com/signed.mp4"
+        mock_get_storage.return_value = storage
+        repo = MagicMock()
+        asset = MagicMock()
+        asset.id = "a1"
+        asset.storage_key = "videos/test.mp4"
+        repo.find_by_ids.return_value = [asset]
+
+        result = _build_asset_url_map(["a1"], repo)
+        assert result == {"a1": "https://cdn.example.com/signed.mp4"}
+        storage.get_download_url.assert_called_once_with("videos/test.mp4", expires_seconds=3600)
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_exception_during_url_generation(self, mock_get_storage):
+        """单个 asset 生成 URL 异常 → None，不影响其他"""
+        from app.api.routes.templates_editor.clips import _build_asset_url_map
+
+        storage = MagicMock()
+        storage.get_download_url.side_effect = [Exception("boom"), "https://ok.com/v2"]
+        mock_get_storage.return_value = storage
+
+        repo = MagicMock()
+        asset1 = MagicMock()
+        asset1.id = "a1"
+        asset1.storage_key = "v1.mp4"
+        asset2 = MagicMock()
+        asset2.id = "a2"
+        asset2.storage_key = "v2.mp4"
+        repo.find_by_ids.return_value = [asset1, asset2]
+
+        result = _build_asset_url_map(["a1", "a2"], repo)
+        assert result["a1"] is None
+        assert result["a2"] == "https://ok.com/v2"
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_skip_empty_asset_id(self, mock_get_storage):
+        """空字符串 asset_id 被跳过"""
+        from app.api.routes.templates_editor.clips import _build_asset_url_map
+
+        storage = MagicMock()
+        mock_get_storage.return_value = storage
+        repo = MagicMock()
+
+        result = _build_asset_url_map(["", "a1"], repo)
+        # "" not in result because it's skipped by `if not aid: continue`
+        assert "" not in result
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_multiple_assets_mixed(self, mock_get_storage):
+        """混合场景：正常+异常+缺失"""
+        from app.api.routes.templates_editor.clips import _build_asset_url_map
+
+        storage = MagicMock()
+        storage.get_download_url.return_value = "https://cdn.com/ok.mp4"
+        mock_get_storage.return_value = storage
+
+        repo = MagicMock()
+        good_asset = MagicMock()
+        good_asset.id = "a1"
+        good_asset.storage_key = "good.mp4"
+        # a1=good, a2=not found, a3=good
+        good_asset2 = MagicMock()
+        good_asset2.id = "a3"
+        good_asset2.storage_key = "good.mp4"
+        repo.find_by_ids.return_value = [good_asset, good_asset2]
+
+        result = _build_asset_url_map(["a1", "a2", "a3"], repo)
+        assert result["a1"] == "https://cdn.com/ok.mp4"
+        assert result["a2"] is None
+        assert result["a3"] == "https://cdn.com/ok.mp4"
+
+
+# ---------------------------------------------------------------------------
+# 集成测试: 路由层 asset_repo 注入
+# ---------------------------------------------------------------------------
+
+
+class TestClipRoutesAssetIntegration:
+    """路由层测试 — 覆盖 asset_url 在 list/detail/split/merge 中的拼接逻辑"""
+
+    def _create_app(self, plan_svc_config=None):
+        from app.api.routes import templates_editor as editor_module
+        from app.dependencies import get_asset_repository
+
+        mock_clip_1 = _clip(id="c1", asset_id="asset-001")
+        mock_clip_2 = _clip(id="c2", asset_id="")
+
+        mock_tpl_svc = MagicMock()
+        mock_plan_svc = MagicMock()
+        mock_plan_svc.list_clips.return_value = [mock_clip_1, mock_clip_2]
+        mock_plan_svc.count_clips.return_value = 2
+        mock_plan_svc.get_clip.return_value = mock_clip_1
+        mock_plan_svc.create_clip.return_value = _clip(id="c-new", asset_id="")
+        mock_plan_svc.update_clip.return_value = _clip(id="c1", duration=15.0)
+        mock_plan_svc.delete_clip.return_value = True
+        mock_plan_svc.split_clip.return_value = {
+            "left_clip": _clip(id="c-left", asset_id="asset-L"),
+            "right_clip": _clip(id="c-right", asset_id="asset-R"),
+        }
+        mock_plan_svc.merge_clips.return_value = _clip(id="c-merged", asset_id="asset-M")
+
+        if plan_svc_config:
+            for k, v in plan_svc_config.items():
+                setattr(mock_plan_svc, k, v)
+
+        def _deps():
+            return mock_tpl_svc, mock_plan_svc
+
+        mock_asset_repo = MagicMock()
+
+        app = FastAPI()
+        app.include_router(
+            editor_module.router,
+            prefix="/api/v1/templates/{template_id}/editor",
+        )
+        app.dependency_overrides[editor_module.get_current_user] = _auth_user
+        app.dependency_overrides[editor_module.get_draft_plan_id] = lambda: TEST_PLAN_ID
+        app.dependency_overrides[editor_module.get_editor_services] = _deps
+        app.dependency_overrides[get_asset_repository] = lambda: mock_asset_repo
+
+        return TestClient(app), mock_plan_svc, mock_asset_repo
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_list_clips_includes_asset_urls(self, mock_get_storage):
+        """GET /clips 应为有 asset_id 的片段返回签名 URL"""
+        storage = MagicMock()
+        storage.get_download_url.return_value = "https://cdn.com/c1.mp4"
+        mock_get_storage.return_value = storage
+
+        client, _, asset_repo = self._create_app()
+        asset = MagicMock()
+        asset.id = "asset-001"
+        asset.storage_key = "videos/c1.mp4"
+        asset_repo.find_by_ids.return_value = [asset]
+
+        resp = client.get(f"/api/v1/templates/{TEST_TEMPLATE_ID}/editor/clips")
+        assert resp.status_code == 200
+        data = resp.json()
+        items = data["items"]
+        assert len(items) == 2
+        # c1 has asset_id → should have url
+        assert items[0]["asset_url"] == "https://cdn.com/c1.mp4"
+        # c2 has empty asset_id → None
+        assert items[1]["asset_url"] is None
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_get_clip_detail_with_asset_url(self, mock_get_storage):
+        """GET /clips/{clip_id} 应返回素材签名 URL"""
+        storage = MagicMock()
+        storage.get_download_url.return_value = "https://cdn.com/detail.mp4"
+        mock_get_storage.return_value = storage
+
+        client, _, asset_repo = self._create_app()
+        asset = MagicMock()
+        asset.id = "asset-001"
+        asset.storage_key = "videos/detail.mp4"
+        asset_repo.find_by_ids.return_value = [asset]
+
+        resp = client.get(f"/api/v1/templates/{TEST_TEMPLATE_ID}/editor/clips/clip-001")
+        assert resp.status_code == 200
+        assert resp.json()["asset_url"] == "https://cdn.com/detail.mp4"
+
+    def test_get_clip_detail_no_asset(self):
+        """片段没有 asset_id 时不应调用 URL 解析"""
+        client, plan_svc, asset_repo = self._create_app()
+        # 返回没有 asset_id 的片段
+        plan_svc.get_clip.return_value = _clip(id="c-no-asset", asset_id="")
+
+        resp = client.get(f"/api/v1/templates/{TEST_TEMPLATE_ID}/editor/clips/c-no-asset")
+        assert resp.status_code == 200
+        assert resp.json()["asset_url"] is None
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_split_clip_returns_asset_urls(self, mock_get_storage):
+        """POST /clips/{clip_id}/split 返回的左右片段应带签名 URL"""
+        storage = MagicMock()
+        storage.get_download_url.side_effect = ["https://cdn.com/L.mp4", "https://cdn.com/R.mp4"]
+        mock_get_storage.return_value = storage
+
+        client, _, asset_repo = self._create_app()
+        asset_l = MagicMock()
+        asset_l.storage_key = "videos/L.mp4"
+        asset_r = MagicMock()
+        asset_r.storage_key = "videos/R.mp4"
+        asset_l.id = "asset-L"
+        asset_r.id = "asset-R"
+        asset_repo.find_by_ids.return_value = [asset_l, asset_r]
+
+        resp = client.post(
+            f"/api/v1/templates/{TEST_TEMPLATE_ID}/editor/clips/clip-001/split",
+            json={"split_time": 5.0},
+        )
+        assert resp.status_code == 200
+        data = resp.json()
+        assert data["left_clip"]["asset_url"] == "https://cdn.com/L.mp4"
+        assert data["right_clip"]["asset_url"] == "https://cdn.com/R.mp4"
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_merge_clips_returns_asset_url(self, mock_get_storage):
+        """POST /clips/merge 返回的合并片段应带签名 URL"""
+        storage = MagicMock()
+        storage.get_download_url.return_value = "https://cdn.com/M.mp4"
+        mock_get_storage.return_value = storage
+
+        client, _, asset_repo = self._create_app()
+        asset = MagicMock()
+        asset.id = "asset-M"
+        asset.storage_key = "videos/M.mp4"
+        asset_repo.find_by_ids.return_value = [asset]
+
+        resp = client.post(
+            f"/api/v1/templates/{TEST_TEMPLATE_ID}/editor/clips/merge",
+            json={"clip_ids": ["c1", "c2"]},
+        )
+        assert resp.status_code == 200
+        data = resp.json()
+        assert data["merged_clip"]["asset_url"] == "https://cdn.com/M.mp4"
+        assert data["deleted_clip_ids"] == ["c1", "c2"]
+
+    def test_merge_clips_not_found(self):
+        """merge 时某片段不存在应返回 404"""
+        client, plan_svc, _ = self._create_app()
+        plan_svc.get_clip.return_value = None
+
+        resp = client.post(
+            f"/api/v1/templates/{TEST_TEMPLATE_ID}/editor/clips/merge",
+            json={"clip_ids": ["nonexistent-1", "nonexistent-2"]},
+        )
+        assert resp.status_code == 404
+
+    def test_delete_clip_success(self):
+        """DELETE /clips/{clip_id} 成功返回 204"""
+        client, _, _ = self._create_app()
+        resp = client.delete(f"/api/v1/templates/{TEST_TEMPLATE_ID}/editor/clips/clip-001")
+        assert resp.status_code == 204
+
+    def test_delete_clip_not_found(self):
+        """DELETE 片段不存在返回 404"""
+        client, plan_svc, _ = self._create_app()
+        plan_svc.delete_clip.return_value = False
+        resp = client.delete(f"/api/v1/templates/{TEST_TEMPLATE_ID}/editor/clips/bad-id")
+        assert resp.status_code == 404

tests/unit/test_cover_url_finalize.py

@@ -0,0 +1,132 @@
+"""Tests for cover_url backfill to GenerationTask.
+
+Verifies _finalize_render_success correctly writes cover_url
+from cover_candidates to gen_task.cover_url.
+"""
+
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+# Add worker app to sys.path
+_WORKER_ROOT = Path(__file__).resolve().parents[2] / "apps" / "worker"
+if str(_WORKER_ROOT) not in sys.path:
+    sys.path.insert(0, str(_WORKER_ROOT))
+
+
+class FakeGenTask:
+    """Simple stand-in for GenerationTask that tracks attribute assignment."""
+
+    def __init__(self):
+        object.__setattr__(self, "_assigned", {})
+        self.id = "task-1"
+        self.status = MagicMock()
+        self.status.value = "running"
+
+    def __setattr__(self, name, value):
+        if not name.startswith("_"):
+            self._assigned[name] = value
+        object.__setattr__(self, name, value)
+
+    def append_log(self, **kwargs):
+        pass
+
+
+def _make_plan():
+    plan = MagicMock()
+    plan.project_id = "proj-1"
+    plan.created_by_user_id = "user-1"
+    plan.config = {"batch_id": "batch-1", "mode": "edit_plan", "title": {"text": "test"}}
+    plan.mark_completed = MagicMock()
+    return plan
+
+
+def _call_finalize(cover_candidates=None, gen_task=None, plan=None):
+    from worker_app.tasks.edit_plan_generation import _finalize_render_success
+
+    plan = plan or _make_plan()
+    gen_task = gen_task or FakeGenTask()
+
+    plan_repo = MagicMock()
+    clip_repo = MagicMock()
+    gen_task_repo = MagicMock()
+    gen_task_repo.get.return_value = gen_task
+    db = MagicMock()
+
+    with patch("worker_app.tasks.edit_plan_generation.create_video_record_and_dedup"):
+        result = _finalize_render_success(
+            plan=plan,
+            plan_repo=plan_repo,
+            clip_repo=clip_repo,
+            gen_task_repo=gen_task_repo,
+            db=db,
+            plan_id="plan-1",
+            output_url="https://oss.example.com/output.mp4",
+            storage_key="rendered/plan-1/task-1.mp4",
+            duration=10.0,
+            file_size=1024,
+            width=1280,
+            height=720,
+            rendered_clip_ids=["clip-1"],
+            failed_clip_ids=[],
+            generation_task_id="task-1",
+            output_path=Path("/tmp/output.mp4"),
+            engine="unified",
+            thumbnail_url="",
+            cover_candidates=cover_candidates,
+        )
+
+    return result, gen_task, gen_task_repo
+
+
+class TestFinalizeCoverUrl:
+
+    def test_cover_url_set_from_image_url(self):
+        """cover_candidates with image_url should set gen_task.cover_url"""
+        candidates = [
+            {"image_url": "https://oss.example.com/cover1.jpg", "frame_time": 1.5},
+            {"image_url": "https://oss.example.com/cover2.jpg", "frame_time": 3.0},
+        ]
+        _, gen_task, gen_task_repo = _call_finalize(cover_candidates=candidates)
+        assert gen_task.cover_url == "https://oss.example.com/cover1.jpg"
+        gen_task_repo.update.assert_called()
+
+    def test_cover_url_fallback_to_url_key(self):
+        """Should fallback to 'url' key when 'image_url' is absent"""
+        candidates = [{"url": "https://oss.example.com/cover_url_key.jpg"}]
+        _, gen_task, _ = _call_finalize(cover_candidates=candidates)
+        assert gen_task.cover_url == "https://oss.example.com/cover_url_key.jpg"
+
+    def test_cover_url_not_set_when_empty_list(self):
+        """Empty cover_candidates should not set cover_url"""
+        _, gen_task, _ = _call_finalize(cover_candidates=[])
+        assert "cover_url" not in gen_task._assigned
+
+    def test_cover_url_not_set_when_none(self):
+        """None cover_candidates should not set cover_url"""
+        _, gen_task, _ = _call_finalize(cover_candidates=None)
+        assert "cover_url" not in gen_task._assigned
+
+    def test_cover_url_not_set_when_url_empty(self):
+        """Empty URL strings in candidates should not set cover_url"""
+        candidates = [{"image_url": "", "url": ""}]
+        _, gen_task, _ = _call_finalize(cover_candidates=candidates)
+        assert "cover_url" not in gen_task._assigned
+
+    def test_no_generation_task_no_crash(self):
+        """Should not crash when gen_task is None"""
+        candidates = [{"image_url": "https://oss.example.com/cover.jpg"}]
+        gen_task_repo = MagicMock()
+        gen_task_repo.get.return_value = None
+        result, _, _ = _call_finalize(cover_candidates=candidates)
+        assert result["status"] == "completed"
+
+    def test_image_url_priority_over_url(self):
+        """image_url should take priority over url key"""
+        candidates = [{"image_url": "https://a.jpg", "url": "https://b.jpg"}]
+        _, gen_task, _ = _call_finalize(cover_candidates=candidates)
+        assert gen_task.cover_url == "https://a.jpg"