style: fix lint issues in batch_thumbnail.py

问题根因：
1. OSS bucket 为私有，upload_to_oss 返回的公开 URL 返回 403
2. API 降级逻辑把视频 .mp4 的 signed URL 当作 thumbnail_url 返回
3. 浏览器 <img> 无法渲染 .mp4 文件，显示黑屏

修复：
- ingest.py: 存储 storage_key 而非公开 URL，API 层统一签名
- assets.py: 移除 .mp4 降级，thumbnail_url 通过 get_download_url 签名
- AssetCard.tsx: 检测视频 URL 后缀 + onError 降级显示类型图标
- 新增 batch_generate_thumbnails Celery 任务，为 2587 个历史视频补生成缩略图
- 同时处理 242 条旧格式公开 URL → storage_key 迁移

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

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

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

@@ -264,6 +264,34 @@ def create_preview_generation_task(
     if not video_ratio and request.template_id:
         video_ratio = _infer_video_ratio_from_template(request.template_id, db, user_id)
 
+    # 根据 video_ratio 计算输出分辨率（默认竖屏 1080x1920）
+    output_width, output_height = 1080, 1920
+    if video_ratio:
+        parts = video_ratio.split(":")
+        if len(parts) == 2:
+            try:
+                w, h = int(parts[0]), int(parts[1])
+                base = 1920
+                if w < h:
+                    # 竖屏
+                    output_width = round(base * w / h)
+                    output_height = base
+                else:
+                    # 横屏
+                    output_width = base
+                    output_height = round(base * h / w)
+                # 对齐到偶数
+                output_width = output_width - output_width % 2
+                output_height = output_height - output_height % 2
+            except (ValueError, ZeroDivisionError):
+                output_width, output_height = 1080, 1920
+    resolution = f"{output_width}x{output_height}"
+
+    logger.info(
+        "[预览生成] 分辨率: video_ratio=%s → %s (%dx%d)",
+        video_ratio, resolution, output_width, output_height,
+    )
+
     # 从模板读取 editing_mode / mode 作为 strategy_id（渲染 pipeline 的 mode 参数）
     strategy_id = _resolve_strategy_id_from_template(request.template_id, db, user_id)
 
@@ -287,12 +315,14 @@ def create_preview_generation_task(
                 asset_select_mode="",
                 batch_id="",
                 video_title=request.video_title,
-                resolution="",
+                resolution=resolution,
                 bgm_config=request.bgm_config or {},
                 auto_retry_enabled=False,
                 auto_retry_max=0,
                 is_preview=True,
                 title_config=title_config,
+                output_width=output_width,
+                output_height=output_height,
             )
         )
     except ValueError as e:

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

@@ -169,8 +169,22 @@ def _writeback_edit_plan_config(
         current_config = plan_model.config if isinstance(plan_model.config, dict) else {}
         merged = dict(current_config)
         merged["generation_task_id"] = task_id
+        
+        # 检查标题是否发生变化，如果变化则清除 cover 字段强制重新生成封面
         if title_config:
+            old_title_config = merged.get("title_config", {}) or {}
+            old_title_text = (old_title_config.get("text") or "").strip()
+            new_title_text = (title_config.get("text") or "").strip()
+            if old_title_text != new_title_text:
+                # 标题变化，清除旧封面
+                if "cover" in merged:
+                    del merged["cover"]
+                    logger.info(
+                        "[生成任务] 标题变化，清除旧封面: plan_id=%s old_title=%s new_title=%s",
+                        plan_id, old_title_text, new_title_text,
+                    )
             merged["title_config"] = title_config
+        
         plan_model.config = merged
         db.commit()
         logger.info(

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

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

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

@@ -167,6 +167,7 @@ class ClipsFromAssetsRequest(BaseModel):
 
     asset_ids: List[str] = Field(..., min_length=1, max_length=200, description="素材 ID 列表，按顺序追加到时间线末尾")
     clip_type: str = Field(default="main", description="片段类型，默认 main")
+    required_clips_count: Optional[int] = Field(default=None, ge=1, le=200, description="要求创建的片段数量；不传则等于素材数量")
 
 
 class ClipsFromAssetsResponse(BaseModel):

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

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

apps/api/app/schemas/tts.py

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

apps/api/app/services/edit_plan_service.py

@@ -400,7 +400,7 @@ class EditPlanService:
                 order = clip_item.get("order") or i
                 clip = EditPlanClip.create(
                     plan_id=plan_id,
-                    clip_type="main",
+                    clip_type=clip_item.get("clip_type", "main"),
                     order=order,
                     asset_id=clip_item.get("asset_id", ""),
                     start_time=clip_item.get("start_time", 0.0),

apps/api/app/services/plan_generator_service.py

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

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

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

apps/web/e2e/test_asset.spec.ts

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

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

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

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

@@ -114,8 +114,6 @@ export interface EditPlanConfig {
   auto_subtitles?: boolean
   /** 是否启用 BGM */
   bgm?: boolean
-  /** 生成数量 */
-  generate_count?: number
   /** 素材模式 */
   material_mode?: string
   /** 前端标题设置（Step4 自动保存，与 title_config 字段分离，不影响后端渲染） */

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

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

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

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

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

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

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

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

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

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

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

@@ -1,38 +1,36 @@
 /**
- * 智能剪辑页面 — 服务器渲染预览架构
+ * 智能剪辑页面 — 前端实时预览架构
  * 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
  *
  * 架构：
- * - Step4+ 右侧预览面板自动创建服务器预览渲染任务（POST /generation/preview）
- * - 轮询完成后播放服务器渲染的真实视频（<video> 标签）
- * - 标题样式编辑时 CSS 层实时叠加预览
- * - 素材/配音/BGM 变更自动重新渲染；标题变更标记 stale
- * - 点"确认生成"时走 confirm 路径，成品就是预览视频本身，100% 一致
+ * - 步骤 4-6 右侧显示 FrontendPreviewPlayer 实时预览
+ * - 步骤 7 右侧内联播放生成的最终视频
+ * - 点"确认生成"时调用 createGenerationTask 创建一次服务器渲染任务
  */
-import React, { useMemo, useCallback } from "react"
-import { Modal, message } from "antd"
+import React, { useMemo, useState, useEffect, useRef } from "react"
+import { message } from "antd"
 import { useNavigate } from "react-router-dom"
 import type { VoiceClone } from "@/api/voice-clone"
+import { useQuery } from "@tanstack/react-query"
 import { useCloneProgress } from "@/hooks/useCloneProgress"
-import { calculateResolution } from "./utils/calculateResolution"
 import CloneModal from "@/components/voice/CloneModal"
 import GenerateHeader from "./components/GenerateHeader"
 import {
   calculateTotalVideoDuration,
   estimateTotalVideoDuration,
 } from "./utils/calculateTotalVideoDuration"
+import FrontendPreviewPlayer from "./components/FrontendPreviewPlayer"
 import GenerateStepsBar from "./components/GenerateStepsBar"
-import GenerateResultPanel from "./components/GenerateResultPanel"
-import PreviewVideoPanel from "./components/PreviewVideoPanel"
 import GenerateStepContent from "./components/GenerateStepContent"
 import GenerateStepActions from "./components/GenerateStepActions"
 import { useGenerateFormState } from "./hooks/useGenerateFormState"
 import { useStepNavigation } from "./hooks/useStepNavigation"
 import { useGenerateVideo } from "./hooks/useGenerateVideo"
 import { usePreviewAssets } from "./hooks/usePreviewAssets"
-import { useServerPreview } from "./hooks/useServerPreview"
 import { useTitleStyleUpdaters } from "./hooks/useStep4Title/useTitleStyleUpdaters"
+import { getAssetsByKind } from "@/api/assets"
+import { previewTts } from "@/api/tts"
 import "./generate.css"
 
 const GeneratePage: React.FC = () => {
@@ -65,8 +63,6 @@ const GeneratePage: React.FC = () => {
     presetVoices,
     cloneModalOpen,
     setCloneModalOpen,
-    generateCount,
-    setGenerateCount,
     videoRatio,
     duration,
     style,
@@ -74,14 +70,12 @@ const GeneratePage: React.FC = () => {
     bgm,
     editPlanId,
     sourceEditPlanId,
-    previewVideo,
-    setPreviewVideo,
-    previewModalOpen,
-    setPreviewModalOpen,
     previewTaskId,
     setPreviewTaskId,
     storedSourceEditPlanId,
     setStoredSourceEditPlanId,
+    serverClips,
+    setServerClips,
   } = formState
 
   /* ── 标题样式回调 ── */
@@ -90,6 +84,53 @@ const GeneratePage: React.FC = () => {
     onTitleSettingsChange: setTitleSettings,
   })
 
+  /* ── 配音素材库（TTS 试听）── */
+  const { data: voiceMaterials = [] } = useQuery({
+    queryKey: ["assets", "voice"],
+    queryFn: () => getAssetsByKind("voice", { limit: 50 }),
+  })
+
+  const [previewVoiceAudioUrl, setPreviewVoiceAudioUrl] = useState<string | null>(null)
+  const ttsAbortRef = useRef<AbortController | null>(null)
+
+  useEffect(() => {
+    const voiceAsset = voiceMaterials.find((m) => m.id === selectedVoice)
+    if (voiceAsset?.file_url) {
+      setPreviewVoiceAudioUrl(voiceAsset.file_url)
+      return
+    }
+
+    const voiceId = selectedClonedVoice || selectedVoice
+    if (!voiceId || !titleSettings.title) {
+      setPreviewVoiceAudioUrl(null)
+      return
+    }
+
+    ttsAbortRef.current?.abort()
+    const controller = new AbortController()
+    ttsAbortRef.current = controller
+    let cancelled = false
+
+    previewTts({ text: titleSettings.title, voice_id: voiceId })
+      .then((res) => {
+        if (!cancelled && res.audio_url) {
+          setPreviewVoiceAudioUrl(res.audio_url)
+        }
+      })
+      .catch((err) => {
+        if (!cancelled) {
+          console.warn("[预览配音生成失败]", err)
+          setPreviewVoiceAudioUrl(null)
+        }
+      })
+
+    return () => {
+      cancelled = true
+      controller.abort()
+    }
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [selectedVoice, selectedClonedVoice, titleSettings.title, voiceMaterials])
+
   /* ── 克隆声音 ── */
   const { clones: clonedVoices, addClone, hasProcessing } = useCloneProgress()
 
@@ -120,10 +161,7 @@ const GeneratePage: React.FC = () => {
     [bgm, currentTemplate],
   )
 
-  /* ── 输出分辨率（根据视频比例计算） ── */
-  const resolution = useMemo(() => calculateResolution(videoRatio), [videoRatio])
-
-  /* ── 加载素材详情（仅用于配音时长校验，不用于播放） ── */
+  /* ── 加载素材详情（供前端预览播放器使用 + 配音时长校验） ── */
   const previewAssetsEnabled = previewAssetIds.length > 0
   const { assets: previewAssets } = usePreviewAssets(previewAssetIds, previewAssetsEnabled)
 
@@ -134,74 +172,7 @@ const GeneratePage: React.FC = () => {
     return estimateTotalVideoDuration(currentTemplate ?? undefined)
   }, [previewAssets, currentTemplate])
 
-  /* ── 解析配音 ID ── */
-  const voiceLibraryId = useMemo(
-    () => (voiceMode === "clone" ? selectedClonedVoice || "" : selectedVoice || ""),
-    [voiceMode, selectedClonedVoice, selectedVoice],
-  )
-
-  /* ── 构建服务器预览请求参数 ── */
-  const buildPreviewRequest = useCallback(() => {
-    return {
-      template_id: selectedTemplate,
-      asset_ids: previewAssetIds,
-      duration: duration || 30,
-      video_ratio: videoRatio,
-      output_width: resolution.width,
-      output_height: resolution.height,
-      ...(voiceLibraryId ? { voice_library_id: voiceLibraryId } : {}),
-      bgm_config: {
-        enabled: bgm !== false,
-        ...(bgmConfig.music_id ? { preset_id: bgmConfig.music_id } : {}),
-      },
-      ...(titleSettings.title
-        ? {
-            title_config: {
-              text: titleSettings.title,
-              font: titleSettings.font,
-              font_size: titleSettings.size,
-              font_color: titleSettings.color,
-              position: titleSettings.position,
-              bold: titleSettings.bold,
-              stroke: titleSettings.stroke,
-              shadow: titleSettings.shadow,
-            },
-          }
-        : {}),
-    }
-  }, [
-    selectedTemplate,
-    previewAssetIds,
-    duration,
-    videoRatio,
-    resolution,
-    voiceLibraryId,
-    bgm,
-    bgmConfig,
-    titleSettings,
-  ])
-
-  /* ── 服务器预览 ── */
-  const serverPreviewEnabled = currentStep >= 4 && !!selectedTemplate && previewAssetIds.length > 0
-
-  const {
-    status: previewStatus,
-    videoUrl: previewVideoUrl,
-    progress: previewProgress,
-    error: previewError,
-    triggerPreview,
-  } = useServerPreview({
-    enabled: serverPreviewEnabled,
-    buildRequest: buildPreviewRequest,
-    onPreviewTaskCreated: (taskId, planId) => {
-      setPreviewTaskId(taskId)
-      if (planId) setStoredSourceEditPlanId(planId)
-    },
-  })
-
   /* ── 步骤导航 ── */
-  // Step5 需要服务器预览完成才能前进
-  const previewReady = previewStatus === "ready"
   const { goNext, goPrev } = useStepNavigation({
     currentStep,
     setCurrentStep,
@@ -210,7 +181,6 @@ const GeneratePage: React.FC = () => {
     selectedMaterials,
     smartSelectedIds,
     titleSettings,
-    previewReady,
   })
 
   /* ── 视频生成核心逻辑 ── */
@@ -240,7 +210,6 @@ const GeneratePage: React.FC = () => {
     duration,
     autoSubtitles,
     bgm,
-    generateCount,
     sourceEditPlanId: storedSourceEditPlanId || sourceEditPlanId,
     previewTaskId,
     bgmConfig,
@@ -250,11 +219,6 @@ const GeneratePage: React.FC = () => {
     },
   })
 
-  /* ── 手动重新预览（标题变更后或失败重试） ── */
-  const handleRetryPreview = useCallback(() => {
-    triggerPreview()
-  }, [triggerPreview])
-
   /* ================================================================
      渲染
      ================================================================ */
@@ -265,7 +229,7 @@ const GeneratePage: React.FC = () => {
 
       <GenerateStepsBar currentStep={currentStep} onStepClick={setCurrentStep} />
 
-      <div className="xx-generate-layout">
+      <div className={`xx-generate-layout${currentStep < 4 ? " full-width" : ""}`}>
         {/* ════ 左侧：表单区 ════ */}
         <div className="xx-generate-form">
           <GenerateStepContent
@@ -301,6 +265,7 @@ const GeneratePage: React.FC = () => {
             selectedVoice={selectedVoice}
             onSelectedVoiceChange={setSelectedVoice}
             totalVideoDuration={totalVideoDuration}
+            onServerClipsChange={setServerClips}
             voiceMode={voiceMode}
             onVoiceModeChange={setVoiceMode}
             selectedClonedVoice={selectedClonedVoice}
@@ -310,8 +275,6 @@ const GeneratePage: React.FC = () => {
             hasProcessing={hasProcessing}
             cloneModalOpen={cloneModalOpen}
             onCloneModalOpenChange={setCloneModalOpen}
-            generateCount={generateCount}
-            onGenerateCountChange={setGenerateCount}
             generating={generating}
             generated={generated}
             generateError={generateError}
@@ -320,8 +283,6 @@ const GeneratePage: React.FC = () => {
             onRetry={handleRetryGenerate}
             onDismissError={handleDismissError}
             presetVoices={presetVoices}
-            previewStatus={previewStatus}
-            onRetryPreview={handleRetryPreview}
           />
 
           <GenerateStepActions
@@ -335,62 +296,57 @@ const GeneratePage: React.FC = () => {
           />
         </div>
 
-        {/* ════ 右侧：预览 + 结果 ════ */}
+        {/* ════ 右侧：步骤 4-6 实时预览，步骤 7 最终视频 ════ */}
         <div className="xx-generate-right-col">
-          {currentStep >= 4 && (
-            <PreviewVideoPanel
-              previewStatus={previewStatus}
-              videoUrl={previewVideoUrl}
-              progress={previewProgress}
-              error={previewError}
-              onRetry={handleRetryPreview}
+          {currentStep >= 4 && currentStep <= 6 && !!currentTemplate && (
+            <FrontendPreviewPlayer
+              assets={previewAssets}
+              template={currentTemplate}
               videoRatio={videoRatio}
-              titleSettings={titleSettings}
-              assetCount={previewAssetIds.length}
+              ready={previewAssets.length > 0}
+              serverClips={serverClips}
+              voiceAudioUrl={previewVoiceAudioUrl || undefined}
+              titleSettings={{
+                title: titleSettings.title,
+                size: titleSettings.size,
+                font: titleSettings.font,
+                color: titleSettings.color,
+                position: titleSettings.position as "top" | "center" | "bottom",
+                bold: titleSettings.bold,
+                italic: titleSettings.italic,
+                stroke: titleSettings.stroke,
+                shadow: titleSettings.shadow,
+              }}
             />
           )}
-          {currentStep >= 6 && (
-            <GenerateResultPanel
-              generated={generated}
-              generating={generating}
-              progress={progress}
-              generateError={generateError}
-              generatedVideos={generatedVideos}
-              onVideoPreview={(video) => {
-                setPreviewVideo(video)
-                setPreviewModalOpen(true)
-              }}
-              onDownload={handleDownload}
-              onShare={handleShare}
-              onGoToLibrary={() => navigate("/app/products")}
-            />
+          {currentStep === 7 && generated && generatedVideos.length > 0 && (
+            <div className="xx-inline-video-player">
+              <video
+                src={generatedVideos[0].download_url || generatedVideos[0].file_url}
+                controls
+                autoPlay
+                style={{ width: "100%", maxHeight: "70vh", objectFit: "contain", borderRadius: 12 }}
+                poster={generatedVideos[0].thumbnail_url || undefined}
+              />
+              <div style={{ display: "flex", gap: 8, marginTop: 12, justifyContent: "center" }}>
+                <button className="xx-btn xx-btn-ghost xx-btn-sm" onClick={handleDownload}>
+                  ⬇️ 下载
+                </button>
+                <button className="xx-btn xx-btn-ghost xx-btn-sm" onClick={handleShare}>
+                  🔗 分享
+                </button>
+                <button
+                  className="xx-btn xx-btn-ghost xx-btn-sm"
+                  onClick={() => navigate("/app/products")}
+                >
+                  📁 前往成片库
+                </button>
+              </div>
+            </div>
           )}
         </div>
       </div>
 
-      {/* 视频预览弹窗 */}
-      <Modal
-        className="xx-preview-modal"
-        open={previewModalOpen}
-        onCancel={() => setPreviewModalOpen(false)}
-        footer={null}
-        width="80vw"
-        centered
-        destroyOnClose
-      >
-        {previewVideo && (
-          <div className="xx-preview-modal-content">
-            <video
-              src={previewVideo.download_url || previewVideo.file_url}
-              controls
-              autoPlay
-              style={{ width: "100%", maxHeight: "70vh", objectFit: "contain" }}
-              poster={previewVideo.thumbnail_url || undefined}
-            />
-          </div>
-        )}
-      </Modal>
-
       {/* 音色克隆弹窗 */}
       <CloneModal
         open={cloneModalOpen}

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

@@ -16,6 +16,7 @@ import {
 } from "@ant-design/icons"
 import type { AssetItem } from "@/api/assets"
 import type { EditingTemplate } from "@/api/editing-planner"
+import type { EditPlanClip } from "@/api/template-editor"
 import { useSegmentScheduler, type PlaybackSegment } from "../hooks/useSegmentScheduler"
 import { useCanvasPlayer } from "../hooks/useCanvasPlayer"
 
@@ -24,6 +25,7 @@ interface FrontendPreviewPlayerProps {
   template: EditingTemplate | null
   videoRatio: string
   ready: boolean
+  serverClips?: EditPlanClip[]
   voiceAudioUrl?: string
   titleSettings?: {
     title: string
@@ -50,9 +52,31 @@ function formatTime(seconds: number): string {
 function buildPlaybackSegments(
   assets: AssetItem[],
   template: EditingTemplate | null,
+  serverClips?: EditPlanClip[],
 ): PlaybackSegment[] {
   if (!assets.length) return []
 
+  // Build asset lookup map
+  const assetMap = new Map(assets.map((a) => [a.id, a]))
+
+  // 优先使用服务端 clips（含随机 start_time 和正确数量），与最终生成结果一致
+  if (serverClips && serverClips.length > 0) {
+    const segments: PlaybackSegment[] = []
+    for (const clip of serverClips) {
+      const asset = assetMap.get(clip.asset_id)
+      if (!asset) continue
+      const assetDuration = asset.duration || asset.metadata?.duration || 30
+      const startTime = clip.start_time || 0
+      const endTime = Math.min(startTime + (clip.duration || assetDuration), assetDuration)
+      const videoUrl = asset.file_url || asset.storage_key
+      segments.push({ assetId: asset.id, videoUrl, startTime, endTime, order: clip.order })
+    }
+    if (segments.length > 0) {
+      return segments.sort((a, b) => a.order - b.order)
+    }
+  }
+
+  // Fallback: 本地构建片段（与旧行为一致）
   const templateSegments = template?.segments || []
   const segments: PlaybackSegment[] = []
 
@@ -76,12 +100,63 @@ function buildPlaybackSegments(
 const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
   assets,
   template,
-  videoRatio: _videoRatio,
+  videoRatio,
   ready,
+  serverClips,
   voiceAudioUrl,
   titleSettings,
 }) => {
-  const segments = useMemo(() => buildPlaybackSegments(assets, template), [assets, template])
+  const segments = useMemo(
+    () => buildPlaybackSegments(assets, template, serverClips),
+    [assets, template, serverClips],
+  )
+
+  // ── ASS 坐标系参数（与后端 ass_subtitle_builder.py 一致） ──
+  const TITLE_MARGIN_TOP = 120
+  const TITLE_MARGIN_BOTTOM = 60
+  const TITLE_MARGIN_SIDE = 40
+  const playRes = (() => {
+    switch (videoRatio) {
+      case "16:9":
+        return { width: 1920, height: 1080 }
+      case "1:1":
+        return { width: 1080, height: 1080 }
+      case "9:16":
+      default:
+        return { width: 1080, height: 1920 }
+    }
+  })()
+  const playerContainerRef = useRef<HTMLDivElement>(null)
+  const [containerHeight, setContainerHeight] = useState(0)
+  useEffect(() => {
+    const el = playerContainerRef.current
+    if (!el) return
+    const ro = new ResizeObserver((entries) => {
+      for (const entry of entries) {
+        const h = entry.contentRect.height
+        if (h > 0) setContainerHeight(h)
+      }
+    })
+    ro.observe(el)
+    const rect = el.getBoundingClientRect()
+    if (rect.height > 0) setContainerHeight(rect.height)
+    return () => ro.disconnect()
+  }, [])
+
+  // 标题字号按容器高度与 PlayResY 的比例缩放
+  const titleFontSizePx =
+    containerHeight > 0
+      ? ((titleSettings?.size ?? 36) / playRes.height) * containerHeight
+      : (titleSettings?.size ?? 36)
+  const titleSidePct = (TITLE_MARGIN_SIDE / playRes.width) * 100
+  const titleTopPct = (TITLE_MARGIN_TOP / playRes.height) * 100
+  const titleBottomPct = (TITLE_MARGIN_BOTTOM / playRes.height) * 100
+  // 描边/阴影也要按缩放比例放大
+  const titleScale = containerHeight > 0 ? containerHeight / playRes.height : 1
+  const titleStrokeWidth = Math.max(1, 2 * titleScale)
+  const titleShadowBlur = 4 * titleScale
+  const titleShadowOffset = 2 * titleScale
+
   // 默认走原生 video 播放（浏览器硬件解码，独立线程，不阻塞 UI）
   // WebCodecs 仅在明确需要时启用（保留代码作为兜底）
   const useWebCodecs = false
@@ -96,7 +171,7 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
         fontSize: titleSettings.size,
         fontFamily: titleSettings.font || "思源黑体",
         color: titleSettings.color || "#ffffff",
-        position: titleSettings.position || "bottom",
+        position: titleSettings.position || "top",
         bold: titleSettings.bold,
         stroke: titleSettings.stroke,
         shadow: titleSettings.shadow,
@@ -277,20 +352,29 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
   if (!ready || !assets.length) {
     return (
       <div
-        className="xx-preview-empty"
         style={{
-          position: "absolute",
-          inset: 0,
+          position: "relative",
+          width: "100%",
+          maxWidth: 280,
+          aspectRatio: "9 / 16",
+          background: "#0a0a0a",
+          borderRadius: 24,
+          overflow: "hidden",
+          boxShadow:
+            "0 4px 6px -1px rgba(0,0,0,0.3), 0 20px 50px -12px rgba(0,0,0,0.5), inset 0 0 0 1px rgba(255,255,255,0.06)",
           display: "flex",
           flexDirection: "column",
           alignItems: "center",
           justifyContent: "center",
-          zIndex: 1,
         }}
       >
-        <SoundOutlined style={{ fontSize: 48, color: "var(--text-tertiary)", marginBottom: 12 }} />
-        <p className="xx-preview-empty-title">准备预览素材...</p>
-        <p className="xx-preview-empty-desc">加载素材后即可预览播放</p>
+        <SoundOutlined style={{ fontSize: 40, color: "rgba(255,255,255,0.3)", marginBottom: 12 }} />
+        <p style={{ color: "rgba(255,255,255,0.6)", fontSize: 14, margin: "0 0 4px" }}>
+          准备预览素材...
+        </p>
+        <p style={{ color: "rgba(255,255,255,0.35)", fontSize: 12, margin: 0 }}>
+          加载素材后即可预览播放
+        </p>
       </div>
     )
   }
@@ -300,31 +384,48 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
     const showDecodeError = forceVideoFallback && canvasState.hasDecodeError
     return (
       <div
-        className="xx-preview-empty"
         style={{
-          position: "absolute",
-          inset: 0,
+          position: "relative",
+          width: "100%",
+          maxWidth: 280,
+          aspectRatio: "9 / 16",
+          background: "#0a0a0a",
+          borderRadius: 24,
+          overflow: "hidden",
+          boxShadow:
+            "0 4px 6px -1px rgba(0,0,0,0.3), 0 20px 50px -12px rgba(0,0,0,0.5), inset 0 0 0 1px rgba(255,255,255,0.06)",
           display: "flex",
           flexDirection: "column",
           alignItems: "center",
           justifyContent: "center",
-          zIndex: 1,
+          padding: 24,
         }}
       >
         {isBuffering ? (
           <>
-            <LoadingOutlined style={{ fontSize: 48, color: "#fff", marginBottom: 12 }} spin />
-            <p style={{ color: "rgba(255,255,255,0.8)" }}>加载中...</p>
+            <LoadingOutlined style={{ fontSize: 40, color: "#fff", marginBottom: 12 }} spin />
+            <p style={{ color: "rgba(255,255,255,0.8)", fontSize: 14, margin: 0 }}>加载中...</p>
           </>
         ) : showDecodeError ? (
           <>
-            <PlayCircleOutlined style={{ fontSize: 48, color: "#ef4444", marginBottom: 12 }} />
-            <p className="xx-preview-empty-title" style={{ color: "rgba(255,255,255,0.9)" }}>
+            <PlayCircleOutlined style={{ fontSize: 40, color: "#ef4444", marginBottom: 12 }} />
+            <p
+              style={{
+                color: "rgba(255,255,255,0.9)",
+                fontSize: 14,
+                margin: "0 0 4px",
+                fontWeight: 500,
+              }}
+            >
               视频解码失败
             </p>
             <p
-              className="xx-preview-empty-desc"
-              style={{ color: "rgba(255,255,255,0.6)", maxWidth: 300, textAlign: "center" }}
+              style={{
+                color: "rgba(255,255,255,0.5)",
+                fontSize: 12,
+                margin: 0,
+                textAlign: "center",
+              }}
             >
               {canvasState.errorMessage || "当前浏览器不支持该视频编码格式，请刷新重试"}
             </p>
@@ -332,10 +433,14 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
         ) : (
           <>
             <PlayCircleOutlined
-              style={{ fontSize: 48, color: "var(--text-tertiary)", marginBottom: 12 }}
+              style={{ fontSize: 40, color: "rgba(255,255,255,0.3)", marginBottom: 12 }}
             />
-            <p className="xx-preview-empty-title">暂无可播放素材</p>
-            <p className="xx-preview-empty-desc">请先在左侧选择素材</p>
+            <p style={{ color: "rgba(255,255,255,0.6)", fontSize: 14, margin: "0 0 4px" }}>
+              暂无可播放素材
+            </p>
+            <p style={{ color: "rgba(255,255,255,0.35)", fontSize: 12, margin: 0 }}>
+              请先在左侧选择素材
+            </p>
           </>
         )}
       </div>
@@ -343,7 +448,20 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
   }
 
   return (
-    <>
+    <div
+      ref={playerContainerRef}
+      style={{
+        position: "relative",
+        width: "100%",
+        maxWidth: 280,
+        aspectRatio: "9 / 16",
+        background: "#0a0a0a",
+        borderRadius: 24,
+        overflow: "hidden",
+        boxShadow:
+          "0 4px 6px -1px rgba(0,0,0,0.3), 0 20px 50px -12px rgba(0,0,0,0.5), inset 0 0 0 1px rgba(255,255,255,0.06)",
+      }}
+    >
       {/* ── Canvas 渲染层（WebCodecs 路径） ── */}
       {effectiveUseWebCodecs && (
         <div
@@ -392,54 +510,121 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
           />
         ))}
 
-      {/* 播放按钮 */}
+      {/* 标题CSS叠加层 — 与后端 ASS 烧录坐标系 1:1 对齐 */}
+      {titleSettings?.title && (
+        <div
+          style={{
+            position: "absolute",
+            inset: 0,
+            zIndex: 5,
+            pointerEvents: "none",
+            overflow: "hidden",
+          }}
+        >
+          <div
+            style={{
+              position: "absolute",
+              left: `${titleSidePct}%`,
+              right: `${titleSidePct}%`,
+              textAlign: "center",
+              ...(titleSettings.position === "top"
+                ? { top: `${titleTopPct}%` }
+                : titleSettings.position === "center"
+                  ? { top: "50%", transform: "translateY(-50%)" }
+                  : { bottom: `${titleBottomPct}%` }),
+            }}
+          >
+            <span
+              style={{
+                fontSize: `${titleFontSizePx}px`,
+                fontFamily: titleSettings.font || "思源黑体",
+                color: titleSettings.color || "#ffffff",
+                fontWeight: titleSettings.bold ? 700 : 400,
+                fontStyle: titleSettings.italic ? "italic" : "normal",
+                lineHeight: 1.05,
+                wordBreak: "break-word",
+                WebkitTextStroke: titleSettings.stroke
+                  ? `${titleStrokeWidth}px #000000`
+                  : undefined,
+                textShadow: titleSettings.shadow
+                  ? `${titleShadowOffset}px ${titleShadowOffset}px ${titleShadowBlur}px rgba(0,0,0,0.8)`
+                  : undefined,
+              }}
+            >
+              {titleSettings.title.split(/[/／]/).map((part, i) => (
+                <span key={i}>
+                  {i > 0 && <br />}
+                  {part}
+                </span>
+              ))}
+            </span>
+          </div>
+        </div>
+      )}
+
+      {/* 中央播放按钮 */}
       {!isPlaying && (
         <button
-          className="xx-preview-play-btn"
           onClick={handleTogglePlay}
           style={{
             position: "absolute",
             top: "50%",
             left: "50%",
             transform: "translate(-50%, -50%)",
-            background: "rgba(0,0,0,0.5)",
-            border: "none",
+            background: "rgba(0,0,0,0.45)",
+            backdropFilter: "blur(12px)",
+            WebkitBackdropFilter: "blur(12px)",
+            border: "1px solid rgba(255,255,255,0.15)",
             borderRadius: "50%",
-            width: 56,
-            height: 56,
+            width: 52,
+            height: 52,
             cursor: "pointer",
             display: "flex",
             alignItems: "center",
             justifyContent: "center",
             color: "#fff",
-            fontSize: 28,
+            fontSize: 26,
             zIndex: 10,
+            transition: "transform 0.2s ease, background 0.2s ease",
+            boxShadow: "0 4px 20px rgba(0,0,0,0.4)",
+          }}
+          onMouseEnter={(e) => {
+            e.currentTarget.style.transform = "translate(-50%, -50%) scale(1.08)"
+            e.currentTarget.style.background = "rgba(0,0,0,0.6)"
+          }}
+          onMouseLeave={(e) => {
+            e.currentTarget.style.transform = "translate(-50%, -50%) scale(1)"
+            e.currentTarget.style.background = "rgba(0,0,0,0.45)"
           }}
         >
           <PlayCircleOutlined />
         </button>
       )}
 
-      {/* 片段指示器 */}
+      {/* 片段指示器 — 右上角胶囊 */}
       <div
         style={{
           position: "absolute",
           top: 8,
-          left: 8,
-          background: "rgba(0,0,0,0.6)",
-          color: "#fff",
-          fontSize: 11,
+          right: 8,
+          background: "rgba(0,0,0,0.45)",
+          backdropFilter: "blur(8px)",
+          WebkitBackdropFilter: "blur(8px)",
+          color: "rgba(255,255,255,0.9)",
+          fontSize: 10,
+          fontWeight: 500,
           padding: "2px 8px",
-          borderRadius: 4,
+          borderRadius: 999,
           zIndex: 10,
+          border: "1px solid rgba(255,255,255,0.1)",
+          letterSpacing: 0.3,
         }}
       >
-        {`片段 ${videoCurrentSegIdx + 1}/${segments.length}`}
+        {`${videoCurrentSegIdx + 1} / ${segments.length}`}
       </div>
 
-      {/* 控制条 */}
+      {/* 控制条 — 手机风格毛玻璃 */}
       <div
-        className="xx-preview-controls"
         style={{
           position: "absolute",
           bottom: 0,
@@ -447,23 +632,36 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
           right: 0,
           display: "flex",
           alignItems: "center",
-          gap: 8,
-          padding: "8px 12px",
-          background: "linear-gradient(transparent, rgba(0,0,0,0.6))",
+          gap: 10,
+          padding: "12px 16px 16px",
+          background: "linear-gradient(transparent, rgba(0,0,0,0.7))",
+          backdropFilter: "blur(4px)",
+          WebkitBackdropFilter: "blur(4px)",
           zIndex: 10,
         }}
       >
         <button
           onClick={handleTogglePlay}
           style={{
-            background: "none",
+            background: "rgba(255,255,255,0.15)",
             border: "none",
             color: "#fff",
-            fontSize: 18,
+            fontSize: 16,
             cursor: "pointer",
-            padding: 4,
+            width: 32,
+            height: 32,
+            borderRadius: "50%",
             display: "flex",
             alignItems: "center",
+            justifyContent: "center",
+            flexShrink: 0,
+            transition: "background 0.15s",
+          }}
+          onMouseEnter={(e) => {
+            e.currentTarget.style.background = "rgba(255,255,255,0.25)"
+          }}
+          onMouseLeave={(e) => {
+            e.currentTarget.style.background = "rgba(255,255,255,0.15)"
           }}
         >
           {isPlaying ? <PauseCircleOutlined /> : <PlayCircleOutlined />}
@@ -471,10 +669,11 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
 
         <span
           style={{
-            fontSize: 12,
-            color: "rgba(255,255,255,0.8)",
-            minWidth: 80,
+            fontSize: 11,
+            color: "rgba(255,255,255,0.85)",
+            minWidth: 72,
             fontVariantNumeric: "tabular-nums",
+            letterSpacing: 0.2,
           }}
         >
           {formatTime(currentTime)} / {formatTime(totalDuration)}
@@ -485,8 +684,8 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
           onMouseDown={handleMouseDown}
           style={{
             flex: 1,
-            height: 4,
-            background: "rgba(255,255,255,0.15)",
+            height: 3,
+            background: "rgba(255,255,255,0.2)",
             borderRadius: 2,
             cursor: "pointer",
             position: "relative",
@@ -496,7 +695,7 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
             style={{
               height: "100%",
               width: `${progressPercent}%`,
-              background: "#3b82f6",
+              background: "#fff",
               borderRadius: 2,
               transition: isDragging ? "none" : "width 0.1s linear",
             }}
@@ -510,15 +709,15 @@ const FrontendPreviewPlayer: React.FC<FrontendPreviewPlayerProps> = ({
               width: 10,
               height: 10,
               borderRadius: "50%",
-              background: "#3b82f6",
-              border: "2px solid #fff",
+              background: "#fff",
+              boxShadow: "0 0 6px rgba(255,255,255,0.5)",
               opacity: isDragging ? 1 : 0,
               transition: "opacity 0.15s",
             }}
           />
         </div>
       </div>
-    </>
+    </div>
   )
 }
 

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

@@ -7,6 +7,7 @@
  */
 import React from "react"
 import type { EditingTemplate } from "@/api/editing-planner"
+import type { EditPlanClip } from "@/api/template-editor"
 import type { PresetVoiceItem } from "@/api/voices"
 import type { VoiceClone } from "@/api/voice-clone"
 import type { CoverConfig } from "../types/cover"
@@ -55,6 +56,7 @@ export interface GenerateStepContentProps {
   selectedVoice: string
   onSelectedVoiceChange: (id: string) => void
   totalVideoDuration?: number
+  onServerClipsChange: (clips: EditPlanClip[]) => void
   voiceMode: "preset" | "custom" | "clone"
   onVoiceModeChange: (mode: "preset" | "custom" | "clone") => void
   selectedClonedVoice: string
@@ -65,8 +67,6 @@ export interface GenerateStepContentProps {
   cloneModalOpen: boolean
   onCloneModalOpenChange: (open: boolean) => void
   /* 生成 */
-  generateCount: number
-  onGenerateCountChange: (n: number) => void
   generating: boolean
   generated: boolean
   generateError: string | null
@@ -84,10 +84,6 @@ export interface GenerateStepContentProps {
   bgm: boolean
   /** BGM 配置（来自模板） */
   bgmConfig?: { enabled: boolean; music_id?: string }
-  /** 服务器预览状态 */
-  previewStatus?: import("../hooks/useServerPreview").ServerPreviewStatus
-  /** 重新预览回调 */
-  onRetryPreview?: () => void
 }
 
 export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) => {
@@ -120,11 +116,10 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
     selectedVoice,
     onSelectedVoiceChange,
     totalVideoDuration,
+    onServerClipsChange,
     voiceMode,
     selectedClonedVoice,
     clonedVoices,
-    generateCount,
-    onGenerateCountChange,
     generating,
     generated,
     generateError,
@@ -137,8 +132,6 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
     onSourceEditPlanIdExtracted,
     bgm,
     bgmConfig,
-    previewStatus,
-    onRetryPreview,
   } = props
 
   /* 当前模板的 segments，传给 Step2 构建 clips */
@@ -165,6 +158,7 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
           onSmartSelectedIdsChange={onSmartSelectedIdsChange}
           selectedTemplate={selectedTemplate}
           templateSegments={templateSegments}
+          onServerClipsChange={onServerClipsChange}
         />
       )
     case 3:
@@ -197,8 +191,6 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
           onApplyPreset={onApplyPreset}
           activePreset={activePreset}
           titlePresets={titlePresets}
-          previewStatus={previewStatus || "idle"}
-          onRetryPreview={onRetryPreview || (() => {})}
         />
       )
     case 6:
@@ -234,8 +226,6 @@ export const GenerateStepContent: React.FC<GenerateStepContentProps> = (props) =
           presetVoices={presetVoices}
           clonedVoices={clonedVoices}
           coverSettings={coverSettings}
-          generateCount={generateCount}
-          onGenerateCountChange={onGenerateCountChange}
           generating={generating}
           generated={generated}
           generateError={generateError}

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

@@ -40,17 +40,33 @@ interface PreviewVideoPanelProps {
 }
 
 /* ── ASS 坐标系参数（与后端 ass_subtitle_builder.py 一致） ── */
-const ASS_VIDEO_HEIGHT = 720
-const ASS_TITLE_MARGIN_TOP = 60
-const ASS_TITLE_MARGIN_BOTTOM = 60
-const ASS_TITLE_MARGIN_SIDE = 40
+const TITLE_MARGIN_TOP = 120
+const TITLE_MARGIN_BOTTOM = 60
+const TITLE_MARGIN_SIDE = 40
 
-function getPositionStyle(position: string): React.CSSProperties {
-  const sidePercent = (ASS_TITLE_MARGIN_SIDE / 1280) * 100
+/** 根据视频比例返回后端实际渲染分辨率（PlayResX × PlayResY） */
+function getResolution(ratio: string): { width: number; height: number } {
+  switch (ratio) {
+    case "16:9":
+      return { width: 1920, height: 1080 }
+    case "1:1":
+      return { width: 1080, height: 1080 }
+    case "9:16":
+    default:
+      return { width: 1080, height: 1920 }
+  }
+}
+
+function getPositionStyle(
+  position: string,
+  playResX: number,
+  playResY: number,
+): React.CSSProperties {
+  const sidePercent = (TITLE_MARGIN_SIDE / playResX) * 100
   switch (position) {
     case "bottom":
       return {
-        bottom: `${(ASS_TITLE_MARGIN_BOTTOM / ASS_VIDEO_HEIGHT) * 100}%`,
+        bottom: `${(TITLE_MARGIN_BOTTOM / playResY) * 100}%`,
         left: `${sidePercent}%`,
         right: `${sidePercent}%`,
         textAlign: "center",
@@ -66,7 +82,7 @@ function getPositionStyle(position: string): React.CSSProperties {
     case "top":
     default:
       return {
-        top: `${(ASS_TITLE_MARGIN_TOP / ASS_VIDEO_HEIGHT) * 100}%`,
+        top: `${(TITLE_MARGIN_TOP / playResY) * 100}%`,
         left: `${sidePercent}%`,
         right: `${sidePercent}%`,
         textAlign: "center",
@@ -74,11 +90,16 @@ function getPositionStyle(position: string): React.CSSProperties {
   }
 }
 
-function buildTitleStyle(settings: TitleSettings, containerHeight: number): React.CSSProperties {
+function buildTitleStyle(
+  settings: TitleSettings,
+  containerHeight: number,
+  playResY: number,
+): React.CSSProperties {
+  // 字号按容器高度与 PlayResY 的比例缩放，不设上限（与后端一致）
   const fontSizePx =
     containerHeight > 0
-      ? (Math.min(settings.size, 96) / ASS_VIDEO_HEIGHT) * containerHeight
-      : (Math.min(settings.size, 96) / ASS_VIDEO_HEIGHT) * 400
+      ? (settings.size / playResY) * containerHeight
+      : (settings.size / playResY) * 400
 
   const base: React.CSSProperties = {
     fontFamily: getFontFamily(settings.font),
@@ -86,12 +107,10 @@ function buildTitleStyle(settings: TitleSettings, containerHeight: number): Reac
     color: settings.color || "#ffffff",
     fontWeight: settings.bold ? 700 : 400,
     fontStyle: settings.italic ? "italic" : "normal",
-    lineHeight: 1.3,
+    lineHeight: 1.05,
     wordBreak: "break-word",
     pointerEvents: "none",
     userSelect: "none",
-    paddingLeft: `${(ASS_TITLE_MARGIN_SIDE / 1280) * 100}%`,
-    paddingRight: `${(ASS_TITLE_MARGIN_SIDE / 1280) * 100}%`,
   }
   if (settings.stroke) base.WebkitTextStroke = "1px #000000"
   if (settings.shadow) base.textShadow = "2px 2px 4px rgba(0,0,0,0.8)"
@@ -99,7 +118,10 @@ function buildTitleStyle(settings: TitleSettings, containerHeight: number): Reac
 }
 
 /** CSS 标题实时预览覆盖层 */
-const TitleOverlay: React.FC<{ titleSettings: TitleSettings }> = ({ titleSettings }) => {
+const TitleOverlay: React.FC<{ titleSettings: TitleSettings; videoRatio: string }> = ({
+  titleSettings,
+  videoRatio,
+}) => {
   const containerRef = useRef<HTMLDivElement>(null)
   const [containerHeight, setContainerHeight] = useState(400)
 
@@ -118,12 +140,15 @@ const TitleOverlay: React.FC<{ titleSettings: TitleSettings }> = ({ titleSetting
     return () => ro.disconnect()
   }, [])
 
+  const { width: playResX, height: playResY } = getResolution(videoRatio)
+
   const positionStyle = useMemo(
-    () => getPositionStyle(titleSettings.position),
-    [titleSettings.position],
+    () => getPositionStyle(titleSettings.position, playResX, playResY),
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+    [titleSettings.position, playResX, playResY],
   )
   const titleStyle = useMemo(
-    () => buildTitleStyle(titleSettings, containerHeight),
+    () => buildTitleStyle(titleSettings, containerHeight, playResY),
     // eslint-disable-next-line react-hooks/exhaustive-deps
     [
       containerHeight,
@@ -134,6 +159,7 @@ const TitleOverlay: React.FC<{ titleSettings: TitleSettings }> = ({ titleSetting
       titleSettings.italic,
       titleSettings.stroke,
       titleSettings.shadow,
+      playResY,
     ],
   )
 
@@ -151,7 +177,7 @@ const TitleOverlay: React.FC<{ titleSettings: TitleSettings }> = ({ titleSetting
       }}
     >
       <div style={{ ...positionStyle, ...titleStyle, position: "absolute" }}>
-        {displayTitle.split("/").map((part, i) => (
+        {displayTitle.split(/[/／]/).map((part, i) => (
           <span key={i}>
             {i > 0 && <br />}
             {part}
@@ -266,7 +292,9 @@ export const PreviewVideoPanel: React.FC<PreviewVideoPanelProps> = ({
         )}
 
         {/* 标题样式实时预览层（仅在有视频时叠加） */}
-        {isReady && titleSettings && <TitleOverlay titleSettings={titleSettings} />}
+        {isReady && titleSettings && (
+          <TitleOverlay titleSettings={titleSettings} videoRatio={videoRatio} />
+        )}
 
         {/* stale 遮罩：配置变更提示 */}
         {isStale && (

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

@@ -3,6 +3,7 @@
  */
 import React from "react"
 import type { TemplateSegment } from "@/api/templates/types"
+import type { EditPlanClip } from "@/api/template-editor"
 import { useStep2Materials } from "../hooks/useStep2Materials"
 import MaterialModeTabs from "./material/MaterialModeTabs"
 import ManualMaterialList from "./material/ManualMaterialList"
@@ -20,6 +21,8 @@ interface Step2MaterialSelectProps {
   selectedTemplate?: string
   /** 当前模板的 segments（用于构建 clips duration） */
   templateSegments?: TemplateSegment[]
+  /** 服务端 clips 创建成功后的回调 */
+  onServerClipsChange?: (clips: EditPlanClip[]) => void
 }
 
 const Step2MaterialSelect: React.FC<Step2MaterialSelectProps> = (props) => {

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

@@ -1,18 +1,15 @@
 /**
  * Step 5 预览设置组件
  *
- * 服务器渲染预览架构：
- * - 进入此步骤时右侧面板自动播放服务器渲染的真实视频
- * - 标题样式可实时调整（CSS 层叠加预览）
- * - 调整标题后点击"重新预览"可刷新服务器渲染结果
- * - 素材/配音/BGM 变更会自动重新渲染
+ * 前端实时预览架构：
+ * - 右侧面板使用 FrontendPreviewPlayer 实时播放素材片段
+ * - 标题样式可实时调整，CSS 层即时叠加预览
+ * - 点"确认生成"时触发一次服务器渲染
  */
 import React from "react"
-import { PlayCircleOutlined, ReloadOutlined } from "@ant-design/icons"
-import { Button } from "antd"
+import { PlayCircleOutlined } from "@ant-design/icons"
 import { POSITION_OPTIONS, FONT_OPTIONS } from "../constants"
 import type { TitleSettings } from "../types"
-import type { ServerPreviewStatus } from "../hooks/useServerPreview"
 import TitleStylePanel from "./title/TitleStylePanel"
 
 interface Step5GeneratePreviewProps {
@@ -27,10 +24,6 @@ interface Step5GeneratePreviewProps {
   onApplyPreset: (presetKey: string) => void
   activePreset: string | null
   titlePresets: { key: string; label: string; previewStyle: React.CSSProperties }[]
-  /** 服务器预览状态 */
-  previewStatus: ServerPreviewStatus
-  /** 重新预览回调 */
-  onRetryPreview: () => void
 }
 
 const Step5GeneratePreview: React.FC<Step5GeneratePreviewProps> = ({
@@ -45,13 +38,7 @@ const Step5GeneratePreview: React.FC<Step5GeneratePreviewProps> = ({
   onApplyPreset,
   activePreset,
   titlePresets,
-  previewStatus,
-  onRetryPreview,
 }) => {
-  const isLoading = previewStatus === "loading"
-  const isStale = previewStatus === "stale"
-  const isFailed = previewStatus === "failed"
-
   return (
     <div className="xx-form-section">
       <h3>🎬 预览设置</h3>
@@ -70,71 +57,10 @@ const Step5GeneratePreview: React.FC<Step5GeneratePreviewProps> = ({
       >
         <PlayCircleOutlined style={{ fontSize: 18, color: "#3b82f6" }} />
         <span style={{ fontSize: 13, color: "var(--text-secondary, #666)" }}>
-          右侧为服务器渲染的真实预览视频，最终成片与预览完全一致
+          右侧为实时预览，选完素材即可播放。确认生成后服务器渲染最终视频
         </span>
       </div>
 
-      {/* 配置变更提示条 */}
-      {isStale && (
-        <div
-          style={{
-            display: "flex",
-            alignItems: "center",
-            justifyContent: "space-between",
-            gap: 8,
-            padding: "10px 16px",
-            background: "rgba(250, 173, 20, 0.1)",
-            borderRadius: 8,
-            marginBottom: 16,
-            border: "1px solid rgba(250, 173, 20, 0.2)",
-          }}
-        >
-          <span style={{ fontSize: 13, color: "#d48806" }}>
-            标题已修改，点击重新预览刷新服务器渲染
-          </span>
-          <Button size="small" type="primary" icon={<ReloadOutlined />} onClick={onRetryPreview}>
-            重新预览
-          </Button>
-        </div>
-      )}
-
-      {isFailed && (
-        <div
-          style={{
-            display: "flex",
-            alignItems: "center",
-            justifyContent: "space-between",
-            gap: 8,
-            padding: "10px 16px",
-            background: "rgba(255, 77, 79, 0.1)",
-            borderRadius: 8,
-            marginBottom: 16,
-            border: "1px solid rgba(255, 77, 79, 0.2)",
-          }}
-        >
-          <span style={{ fontSize: 13, color: "#ff4d4f" }}>预览渲染失败</span>
-          <Button size="small" danger icon={<ReloadOutlined />} onClick={onRetryPreview}>
-            重新预览
-          </Button>
-        </div>
-      )}
-
-      {isLoading && (
-        <div
-          style={{
-            padding: "10px 16px",
-            background: "rgba(82, 196, 26, 0.08)",
-            borderRadius: 8,
-            marginBottom: 16,
-            border: "1px solid rgba(82, 196, 26, 0.15)",
-          }}
-        >
-          <span style={{ fontSize: 13, color: "#389e0d" }}>
-            ⏳ 正在服务器渲染预览视频，请稍候...
-          </span>
-        </div>
-      )}
-
       <TitleStylePanel
         settings={titleSettings}
         onUpdatePosition={onUpdatePosition}

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

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

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

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

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

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

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

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

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

@@ -113,6 +113,14 @@
   align-items: start;
 }
 
+.xx-generate-layout.full-width {
+  grid-template-columns: 1fr;
+}
+
+.xx-generate-layout.full-width .xx-generate-right-col {
+  display: none;
+}
+
 /* ============================================================
    左侧表单区 generate-form
    ============================================================ */
@@ -186,16 +194,16 @@
    ============================================================ */
 .xx-choice-list {
   display: grid;
-  grid-template-columns: repeat(3, 1fr);
-  gap: 12px;
+  grid-template-columns: repeat(6, 1fr);
+  gap: 8px;
 }
 
 .xx-choice-item {
   position: relative;
   background: var(--bg-primary);
   border: 2px solid var(--border-color);
-  border-radius: var(--radius-md);
-  padding: 14px;
+  border-radius: var(--radius-sm);
+  padding: 6px;
   cursor: pointer;
   transition: 0.18s ease;
   text-align: center;
@@ -211,18 +219,20 @@
 }
 
 .xx-choice-thumb {
-  height: 60px;
+  width: 33%;
+  max-width: 52px;
+  height: 24px;
   border-radius: var(--radius-sm);
   display: grid;
   place-items: center;
   color: var(--text-inverse);
-  font-size: 24px;
+  font-size: 12px;
   font-weight: 700;
-  margin-bottom: 10px;
+  margin: 0 auto 4px;
 }
 
 .xx-choice-item h4 {
-  margin: 0 0 4px;
+  margin: 0 0 2px;
   font-size: 13px;
   font-weight: 600;
   color: var(--text-primary);
@@ -230,7 +240,7 @@
 
 .xx-choice-item p {
   margin: 0;
-  font-size: 11px;
+  font-size: 12px;
   color: var(--text-tertiary);
 }
 
@@ -1055,7 +1065,7 @@
   }
 
   .xx-choice-list {
-    grid-template-columns: repeat(2, 1fr);
+    grid-template-columns: repeat(4, 1fr);
   }
 
   .xx-voice-choice-list {
@@ -1280,53 +1290,6 @@
   }
 }
 
-/* ── 生成数量步进器 ── */
-.xx-count-stepper {
-  display: inline-flex;
-  align-items: center;
-  gap: 6px;
-}
-
-.xx-count-stepper-btn {
-  display: inline-flex;
-  align-items: center;
-  justify-content: center;
-  width: 28px;
-  height: 28px;
-  border: 1px solid var(--border-primary, #e2e8f0);
-  border-radius: 8px;
-  background: var(--bg-surface, #fff);
-  color: var(--text-secondary, #64748b);
-  font-size: 13px;
-  cursor: pointer;
-  transition: all 0.15s;
-}
-
-.xx-count-stepper-btn:hover:not(:disabled) {
-  border-color: var(--primary-400, #818cf8);
-  color: var(--primary-600, #4f46e5);
-  background: var(--primary-50, #eef2ff);
-}
-
-.xx-count-stepper-btn:disabled {
-  opacity: 0.35;
-  cursor: not-allowed;
-}
-
-.xx-count-stepper-value {
-  min-width: 24px;
-  text-align: center;
-  font-size: 16px;
-  font-weight: 600;
-  color: var(--text-primary, #1e293b);
-}
-
-.xx-count-stepper-hint {
-  font-size: 12px;
-  color: var(--text-tertiary, #94a3b8);
-  margin-left: 2px;
-}
-
 /* ── 素材选择模式切换 Tab ── */
 .xx-material-mode-tabs {
   display: flex;
@@ -2525,9 +2488,25 @@
 .xx-generate-right-col {
   display: flex;
   flex-direction: column;
+  align-items: center;
   gap: 16px;
 }
 
+/* ── 内联视频播放器（右侧） ── */
+.xx-inline-video-player {
+  width: 100%;
+  max-width: 320px;
+  background: var(--bg-surface, #fff);
+  border: 1px solid var(--border-primary, #e2e8f0);
+  border-radius: 16px;
+  padding: 16px;
+  box-shadow: 0 2px 8px rgba(0, 0, 0, 0.04);
+}
+
+.xx-inline-video-player video {
+  background: #000;
+}
+
 .xx-preview-header {
   display: flex;
   align-items: center;

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

@@ -18,7 +18,6 @@ export interface UseGenerateVideoProps {
   duration: number
   autoSubtitles: boolean
   bgm: boolean
-  generateCount: number
   /** 当前草稿 ID（URL 参数 edit_plan_id，用于后端回写任务关联） */
   sourceEditPlanId?: string | null
   /** 预览任务 ID（由 useStep6Cover 创建后写入，供 confirmGeneration 复用预览产物） */

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

@@ -4,8 +4,8 @@
  */
 import { useState } from "react"
 import { useSearchParams } from "react-router-dom"
-import type { GeneratedVideo } from "@/api/template-editor"
 import type { EditingTemplate } from "@/api/editing-planner"
+import type { EditPlanClip } from "@/api/template-editor"
 import type { CoverConfig } from "../../types/cover"
 import type { PresetVoiceItem } from "@/api/voices"
 import { DEFAULT_COVER_SETTINGS } from "../../constants"
@@ -19,7 +19,7 @@ import { usePersistedState } from "../usePersistedState"
 const DEFAULT_TITLE_SETTINGS: TitleSettings = {
   aiAutoSelect: false,
   title: "",
-  position: "bottom",
+  position: "top",
   font: "思源黑体",
   size: 28,
   bold: true,
@@ -47,6 +47,10 @@ export interface GenerateFormState {
   smartSelectedIds: string[]
   setSmartSelectedIds: (ids: string[]) => void
 
+  /* 服务端片段（/clips/from-assets 创建后获取） */
+  serverClips: EditPlanClip[]
+  setServerClips: (clips: EditPlanClip[]) => void
+
   /* 标题 */
   titleSettings: TitleSettings
   setTitleSettings: (settings: TitleSettings | ((prev: TitleSettings) => TitleSettings)) => void
@@ -68,10 +72,6 @@ export interface GenerateFormState {
   cloneModalOpen: boolean
   setCloneModalOpen: (open: boolean) => void
 
-  /* 生成数量 */
-  generateCount: number
-  setGenerateCount: (n: number) => void
-
   /* 高级设置 */
   videoRatio: string
   duration: number
@@ -91,12 +91,6 @@ export interface GenerateFormState {
    */
   sourceEditPlanId: string | null
 
-  /* 预览弹窗 */
-  previewVideo: GeneratedVideo | null
-  setPreviewVideo: (v: GeneratedVideo | null) => void
-  previewModalOpen: boolean
-  setPreviewModalOpen: (open: boolean) => void
-
   /** 预览任务 ID（由 useStep6Cover 创建后写入，供 useGenerateVideo 复用） */
   previewTaskId: string | null
   setPreviewTaskId: (id: string | null) => void
@@ -127,6 +121,9 @@ export const useGenerateFormState = (): GenerateFormState => {
   const [materialMode, setMaterialMode] = useState<"manual" | "auto">("manual")
   const [smartSelectedIds, setSmartSelectedIds] = useState<string[]>([])
 
+  /* ── 服务端片段（供预览播放器使用）── */
+  const [serverClips, setServerClips] = useState<EditPlanClip[]>([])
+
   /* ── 标题设置 ── */
   const [titleSettings, setTitleSettings] = useState<TitleSettings>(DEFAULT_TITLE_SETTINGS)
 
@@ -155,9 +152,6 @@ export const useGenerateFormState = (): GenerateFormState => {
   /* ── 克隆声音弹窗 ── */
   const [cloneModalOpen, setCloneModalOpen] = useState(false)
 
-  /* ── 生成数量 ── */
-  const [generateCount, setGenerateCount] = useState(1)
-
   /* ── 高级设置（隐藏但保留） ── */
   const [videoRatio] = useState("9:16")
   const [duration] = useState(30)
@@ -165,10 +159,6 @@ export const useGenerateFormState = (): GenerateFormState => {
   const [autoSubtitles] = useState(true)
   const [bgm] = useState(true)
 
-  /* ── 预览弹窗 ── */
-  const [previewVideo, setPreviewVideo] = useState<GeneratedVideo | null>(null)
-  const [previewModalOpen, setPreviewModalOpen] = useState(false)
-
   /* ── 预览任务 ID（useStep6Cover 创建预览时写入，useGenerateVideo 复用） ── */
   // 持久化到 localStorage，key 按 editPlanId/templateId 区分，刷新页面后可恢复
   const previewStorageKey = editPlanId
@@ -214,6 +204,8 @@ export const useGenerateFormState = (): GenerateFormState => {
     setMaterialMode,
     smartSelectedIds,
     setSmartSelectedIds,
+    serverClips,
+    setServerClips,
     titleSettings,
     setTitleSettings,
     coverSettings,
@@ -227,8 +219,6 @@ export const useGenerateFormState = (): GenerateFormState => {
     presetVoices,
     cloneModalOpen,
     setCloneModalOpen,
-    generateCount,
-    setGenerateCount,
     videoRatio,
     duration,
     style,
@@ -237,10 +227,6 @@ export const useGenerateFormState = (): GenerateFormState => {
     editPlanId,
     sourceEditPlanId,
     planConfigStr,
-    previewVideo,
-    setPreviewVideo,
-    previewModalOpen,
-    setPreviewModalOpen,
     previewTaskId,
     setPreviewTaskId,
     storedSourceEditPlanId,

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

@@ -4,9 +4,8 @@
  */
 import { useState, useCallback } from "react"
 import { message } from "antd"
-import type { GeneratedVideo } from "@/api/template-editor"
+import { type GeneratedVideo, getEditPlanClips, createClipsFromAssets } from "@/api/template-editor"
 import { createGenerationTask } from "@/api/tasks/tasks"
-import { confirmGeneration } from "@/api/generation/confirm"
 import type { UseGenerateVideoProps } from "./generate-video/types"
 import { getGenerationPhase } from "./generate-video/phase"
 import { useGenerationPolling } from "./generate-video/useGenerationPolling"
@@ -30,7 +29,6 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
       setGenerating(false)
       setGenerated(true)
       setGeneratedVideos(videos as GeneratedVideo[])
-      // 生成成功后清除持久化的预览状态，避免下次进入复用旧任务
       onGenerationSuccess?.()
     },
     [onGenerationSuccess],
@@ -61,7 +59,6 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
     clearTimer()
 
     try {
-      // 解析分辨率（共享工具函数，与预览 API 一致）
       const { width: outputWidth, height: outputHeight } = calculateResolution(
         props.videoRatio || "9:16",
       )
@@ -69,46 +66,32 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
       const assetIds =
         props.materialMode === "auto" ? props.smartSelectedIds : props.selectedMaterials
 
-      // 封面 URL：优先 AI 生成缩略图，兜底用户上传
-      const coverUrl = props.coverSettings?.thumbnail_url || props.coverSettings?.upload_url || ""
-
-      // ── Step7 确认生成：优先复用预览产物（秒出），fallback 到全量渲染 ──
-      let taskId: string | undefined
-
-      // 主路径：如果有预览任务 ID 且只生成 1 个视频，调用 confirmGeneration 复用预览产物
-      // generateCount > 1 时需要走 createGenerationTask 支持批量生成
-      const isSingleGenerate = !props.generateCount || props.generateCount === 1
-      if (props.previewTaskId && isSingleGenerate) {
+      // from-assets 已由 useStep2Materials 在用户选素材时（debounce 800ms）调用，
+      // 后端已改为异步秒级返回，这里做一次轻量兜底：
+      // 单次查 clips，已有则直接放行；没有则再调一次 from-assets。
+      if (assetIds.length > 0 && selectedTemplate) {
         try {
-          console.log(
-            "[handleGenerate] 尝试 confirmGeneration 复用预览产物, previewTaskId:",
-            props.previewTaskId,
-          )
-          const confirmResp = await confirmGeneration(props.previewTaskId, {
-            output_width: outputWidth,
-            output_height: outputHeight,
-            cover_url: coverUrl,
-            custom_title: props.titleSettings?.title || "",
-          })
-          taskId = confirmResp.items?.[0]?.id
-          if (taskId) {
-            console.log("[handleGenerate] confirmGeneration 成功, taskId:", taskId)
+          const clipList = await getEditPlanClips(selectedTemplate, { limit: 500 })
+          if (clipList.items.length === 0) {
+            // 片段不存在（极端情况：useStep2Materials 的 debounce 还没触发）
+            // 手动补一次 from-assets（后端秒级返回）
+            await createClipsFromAssets(selectedTemplate, assetIds, "main")
           }
-        } catch (confirmErr) {
-          console.warn(
-            "[handleGenerate] confirmGeneration 失败, fallback 到 createGenerationTask:",
-            confirmErr,
-          )
-          // 继续走 fallback 路径
+        } catch {
+          // 查询失败不阻塞，继续生成
         }
       }
 
-      // 解析配音参数：voiceMode=clone 时用 selectedClonedVoice，否则用 selectedVoice（配音素材库 asset ID）
-      const voiceLibraryId =
-        props.voiceMode === "clone" ? props.selectedClonedVoice || "" : props.selectedVoice || ""
+      const hide = message.loading("正在生成预览视频...", 0)
 
-      // Fallback 路径：没有预览任务或 confirmGeneration 失败，创建新的生成任务
-      if (!taskId) {
+      const coverUrl = props.coverSettings?.thumbnail_url || props.coverSettings?.upload_url || ""
+
+      const voiceLibraryId =
+        props.voiceMode === "clone"
+          ? props.selectedClonedVoice || props.selectedVoice || ""
+          : props.selectedVoice || ""
+
+      try {
         const taskResp = await createGenerationTask({
           template_id: selectedTemplate,
           asset_ids: assetIds,
@@ -118,10 +101,8 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
           custom_title: props.titleSettings?.title || "",
           duration: props.duration || undefined,
           video_ratio: props.videoRatio,
-          // 配音：优先用 voice_library_id（配音素材库 asset），兜底 voice_ids
-          ...(voiceLibraryId ? { voice_library_id: voiceLibraryId } : {}),
+          voice_library_id: voiceLibraryId,
           ...(props.selectedVoice && !voiceLibraryId ? { voice_ids: [props.selectedVoice] } : {}),
-          // BGM 配置：受 bgm 开关控制，enabled=false 时也显式传覆盖模板 BGM
           bgm_config: {
             enabled: props.bgm !== false,
             ...(props.bgmConfig?.music_id ? { preset_id: props.bgmConfig.music_id } : {}),
@@ -142,13 +123,17 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
               }
             : {}),
         })
-        taskId = taskResp.items?.[0]?.id
-      }
+        hide()
+        const taskId = taskResp.items?.[0]?.id
 
-      if (!taskId) {
-        throw new Error("创建任务成功但未返回任务 ID，请稍后在任务列表查看")
+        if (!taskId) {
+          throw new Error("创建任务成功但未返回任务 ID，请稍后在任务列表查看")
+        }
+        startPolling(taskId)
+      } catch (err) {
+        hide()
+        throw err
       }
-      startPolling(taskId)
     } catch (err: unknown) {
       console.error("[handleGenerate] 生成失败:", err)
       setGenerating(false)
@@ -160,18 +145,15 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
     }
   }, [props, clearTimer, startPolling, selectedTemplate])
 
-  /* 重新生成（失败后重试） */
   const retry = useCallback(() => {
     setGenerateError(null)
     generate()
   }, [generate])
 
-  /* 清除错误 */
   const dismissError = useCallback(() => {
     setGenerateError(null)
   }, [])
 
-  /* ── 下载视频 ── */
   const download = useCallback(async () => {
     if (!generatedVideos.length) return
     const video = generatedVideos[0]
@@ -192,7 +174,6 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
     }
   }, [generatedVideos])
 
-  /* ── 分享视频 ── */
   const share = useCallback(async () => {
     if (!generatedVideos.length) return
     const video = generatedVideos[0]
@@ -206,19 +187,16 @@ export function useGenerateVideo(props: UseGenerateVideoProps) {
   }, [generatedVideos])
 
   return {
-    // 状态
     generating,
     progress,
     generated,
     generateError,
     generatedVideos,
-    // 操作
     generate,
     retry,
     dismissError,
     download,
     share,
-    // 工具
     getGenerationPhase,
   }
 }

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

@@ -236,6 +236,12 @@ export function useSegmentScheduler(segments: PlaybackSegment[]): SegmentSchedul
       await waitForReady(video)
     }
 
+    // 播放前 seek 到片段起始时间，确保 progress 计算正确
+    const seg = segmentsRef.current[idx]
+    if (seg && Math.abs(video.currentTime - seg.startTime) > 0.1) {
+      video.currentTime = seg.startTime
+    }
+
     try {
       await video.play()
       setIsPlaying(true)

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

@@ -3,11 +3,11 @@
  * 组合素材库加载 + 智能匹配两个子 Hook
  */
 import { useCallback, useEffect, useRef } from "react"
-import type { AssetItem } from "@/api/assets"
+import { message } from "antd"
 import type { TemplateSegment } from "@/api/templates/types"
-import { updateEditPlanClips } from "@/api/template-editor"
+import type { EditPlanClip } from "@/api/template-editor"
+import { updateEditPlanClips, createClipsFromAssets, getEditPlanClips } from "@/api/template-editor"
 import { formatDuration } from "../utils/formatDuration"
-import { buildClipsFromAssets } from "../utils/buildClipsFromAssets"
 import { useMaterialLibrary } from "./step2-materials/useMaterialLibrary"
 import { useSmartMatch } from "./step2-materials/useSmartMatch"
 import { useDraftAutoSave } from "./useDraftAutoSave"
@@ -23,6 +23,8 @@ interface UseStep2MaterialsProps {
   selectedTemplate?: string
   /** 当前模板的 segments（用于构建 clips duration） */
   templateSegments?: TemplateSegment[]
+  /** 服务端 clips 创建成功后的回调，用于通知预览播放器 */
+  onServerClipsChange?: (clips: EditPlanClip[]) => void
 }
 
 export function useStep2Materials({
@@ -34,6 +36,7 @@ export function useStep2Materials({
   onSmartSelectedIdsChange,
   selectedTemplate,
   templateSegments,
+  onServerClipsChange,
 }: UseStep2MaterialsProps) {
   const { libraries, selectedLibraryId, setSelectedLibraryId, materials, materialsLoading } =
     useMaterialLibrary()
@@ -72,23 +75,30 @@ export function useStep2Materials({
     scheduleSave({ asset_ids: ids }, 500)
   }, [selectedTemplate, materialMode, selectedMaterials, smartSelectedIds, scheduleSave])
 
-  /* ── Step2 选择素材后同步写入 edit_plan_clips（防抖 800ms，失败静默） ── */
+  /* ── Step2 选择素材后同步写入 edit_plan_clips（防抖 800ms，失败静默） ──
+   * 调用后端 POST /clips/from-assets，由后端处理：
+   *   - 素材不够时同一素材切多个片段
+   *   - 随机 start_time，不重复
+   *   - required_clips_count 保证片段数与模板 segments 一致
+   * 先 PUT /clips（空数组）清空旧片段，再调用 from-assets 创建新片段
+   */
   const clipsTimerRef = useRef<ReturnType<typeof setTimeout>>()
   const clipsAbortRef = useRef<AbortController | null>(null)
   const templateSegmentsRef = useRef(templateSegments)
   templateSegmentsRef.current = templateSegments
   const selectedTemplateRef = useRef(selectedTemplate)
   selectedTemplateRef.current = selectedTemplate
-  const materialsRef = useRef(materials)
-  materialsRef.current = materials
-  const smartMatchedRef = useRef<AssetItem[]>(smartMatch.smartMatchedResults)
-  smartMatchedRef.current = smartMatch.smartMatchedResults
+  const onServerClipsChangeRef = useRef(onServerClipsChange)
+  onServerClipsChangeRef.current = onServerClipsChange
 
   useEffect(() => {
     const tid = selectedTemplateRef.current
     if (!tid) return
     const ids = materialMode === "auto" ? smartSelectedIds : selectedMaterials
-    if (!ids.length) return
+    if (!ids.length) {
+      onServerClipsChangeRef.current?.([])
+      return
+    }
 
     if (clipsTimerRef.current) clearTimeout(clipsTimerRef.current)
     clipsTimerRef.current = setTimeout(async () => {
@@ -97,20 +107,34 @@ export function useStep2Materials({
       const controller = new AbortController()
       clipsAbortRef.current = controller
 
-      const clips = buildClipsFromAssets({
-        selectedIds: ids,
-        materials: materialsRef.current.items,
-        smartMatchedAssets: smartMatchedRef.current,
-        templateSegments: templateSegmentsRef.current || [],
-      })
+      const segs = templateSegmentsRef.current || []
+      const requiredClipsCount = segs.length > 0 ? segs.length : undefined
 
       try {
-        await updateEditPlanClips(tid, clips, controller.signal)
+        // 1. 清空旧片段
+        await updateEditPlanClips(tid, [], controller.signal)
+        // 2. 调用后端 from-assets 接口创建片段（异步秒级返回，60s 超时仅为兜底）
+        await createClipsFromAssets(tid, ids, "main", requiredClipsCount, {
+          signal: controller.signal,
+        })
+        // 3. 获取服务端生成的 clips（含 start_time/duration），供预览播放器使用
+        const clipList = await getEditPlanClips(tid, { limit: 500 })
+        const readyClips = clipList.items
+          .filter((c) => c.status === "ready")
+          .sort((a, b) => a.order - b.order)
+        onServerClipsChangeRef.current?.(readyClips)
       } catch (err) {
         const name = (err as { name?: string })?.name
-        if (name !== "CanceledError" && name !== "AbortError") {
-          console.warn("[useStep2Materials] 写入 clips 失败:", err)
+        // 用户切换素材导致的主动取消，静默
+        if (name === "CanceledError" || name === "AbortError") return
+        // from-assets 60s 超时（MediaKit 智能选片未完成）
+        const code = (err as { code?: string })?.code
+        if (code === "ECONNABORTED" || /timeout/i.test((err as Error)?.message || "")) {
+          console.warn("[useStep2Materials] 智能选片超时:", err)
+          message.error("智能选片失败，请重试")
+          return
         }
+        console.warn("[useStep2Materials] 写入 clips 失败:", err)
       }
     }, 800)
 

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

@@ -178,13 +178,15 @@ export function useStep6Cover({
           previewParamsRef.current = JSON.stringify({ selectedTemplate, assetIds, titleSettings })
           // 解析配音参数：voiceMode=clone 时用 selectedClonedVoice，否则用 selectedVoice
           const previewVoiceLibraryId =
-            voiceMode === "clone" ? selectedClonedVoice || "" : selectedVoice || ""
+            voiceMode === "clone" ? selectedClonedVoice || selectedVoice || "" : selectedVoice || ""
           const previewResp = await createPreview({
             template_id: selectedTemplate,
             asset_ids: assetIds,
             duration: duration || 30,
-            // 配音：voice_library_id 是配音素材库 asset ID（用户上传的音频或 AI 配音）
-            ...(previewVoiceLibraryId ? { voice_library_id: previewVoiceLibraryId } : {}),
+            // 配音：始终传递 voice_library_id，确保后端能正确接收
+            voice_library_id: previewVoiceLibraryId,
+            // 兜底：如果 voice_library_id 为空但 selectedVoice 有值，也传 voice_ids
+            ...(selectedVoice && !previewVoiceLibraryId ? { voice_ids: [selectedVoice] } : {}),
             // BGM 配置：受 bgm 开关控制
             bgm_config: {
               enabled: bgm !== false,

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

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

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

@@ -3,7 +3,7 @@
  * 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)
  *
- * 服务器预览架构：Step5 需要服务器渲染预览完成才能前进
+ * 前端实时预览架构：Step5 无需等待服务器渲染
  */
 import { message } from "antd"
 import type { TitleSettings } from "../types"
@@ -16,8 +16,6 @@ export interface UseStepNavigationOptions {
   selectedMaterials: string[]
   smartSelectedIds: string[]
   titleSettings: TitleSettings
-  /** 服务器预览是否已完成（ready 状态） */
-  previewReady: boolean
 }
 
 export interface UseStepNavigationReturn {
@@ -34,7 +32,6 @@ export const useStepNavigation = (options: UseStepNavigationOptions): UseStepNav
     selectedMaterials,
     smartSelectedIds,
     titleSettings,
-    previewReady,
   } = options
 
   const goNext = () => {
@@ -54,10 +51,6 @@ export const useStepNavigation = (options: UseStepNavigationOptions): UseStepNav
       message.warning("请选择或输入标题")
       return
     }
-    if (currentStep === 5 && !previewReady) {
-      message.warning("请等待预览视频渲染完成后再继续")
-      return
-    }
     if (currentStep < 7) {
       setCurrentStep((s) => s + 1)
     }

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

@@ -1,69 +0,0 @@
-/**
- * 将选中素材 + 模板 segments 构建为 edit_plan_clips 写入数据。
- *
- * 逻辑必须与 FrontendPreviewPlayer.tsx 中 buildPlaybackSegments 完全一致：
- *   assetDuration = asset.duration || asset.metadata?.duration || 30
- *   tplSeg = templateSegments[i] || lastSegment
- *   segDuration = clamp(assetDuration, tplSeg.duration_min, tplSeg.duration_max)
- *   start_time = 0
- *   // 关键：预览播放器中 endTime = min(startTime + segDuration, assetDuration)
- *   // 因此 clips.duration 也必须用 min(segDuration, assetDuration) 截断，
- *   // 避免素材实际时长比 clamp 后的 segDuration 短时，Worker 尝试读取不存在的片段
- *   duration = min(segDuration, assetDuration)
- *
- * 预览播放器（Canvas 实时预览）直接在内存中构建 segments 播放，不读 edit_plan_clips；
- * 本函数产出的 clips 写入 DB 后由 Worker 渲染。两边用完全相同的时长计算，
- * 保证用户在编辑过程中看到的预览与最终生成视频一致。
- */
-import type { AssetItem } from "@/api/assets"
-import type { TemplateSegment } from "@/api/templates/types"
-import type { EditPlanClipInput } from "@/api/template-editor"
-
-interface BuildClipsOptions {
-  /** 选中的素材 ID 列表（按选择顺序） */
-  selectedIds: string[]
-  /** 已加载的素材列表（用于查 duration） */
-  materials: AssetItem[]
-  /** 智能匹配返回的素材（auto 模式下可能不在 materials 列表中） */
-  smartMatchedAssets?: AssetItem[]
-  /** 模板 segments */
-  templateSegments?: TemplateSegment[]
-}
-
-export function buildClipsFromAssets({
-  selectedIds,
-  materials,
-  smartMatchedAssets = [],
-  templateSegments = [],
-}: BuildClipsOptions): EditPlanClipInput[] {
-  if (!selectedIds.length) return []
-
-  // 合并两个素材来源，建立 id → asset 索引
-  const assetMap = new Map<string, AssetItem>()
-  for (const a of materials) assetMap.set(a.id, a)
-  for (const a of smartMatchedAssets) assetMap.set(a.id, a)
-
-  const lastSeg = templateSegments[templateSegments.length - 1]
-
-  return selectedIds.map((assetId, i) => {
-    const asset = assetMap.get(assetId)
-    const assetDuration = asset?.duration || asset?.metadata?.duration || 30
-
-    const tplSeg = templateSegments[i] || lastSeg
-    const segDuration = tplSeg
-      ? Math.min(tplSeg.duration_max, Math.max(tplSeg.duration_min, assetDuration))
-      : Math.min(assetDuration, 10)
-
-    // 与 FrontendPreviewPlayer.buildPlaybackSegments 中
-    //   endTime = Math.min(startTime + segDuration, assetDuration)
-    // 保持一致：duration 不能超过素材实际时长
-    const duration = Math.min(segDuration, assetDuration)
-
-    return {
-      asset_id: assetId,
-      start_time: 0,
-      duration,
-      order: i,
-    }
-  })
-}

apps/worker/video_processing/render_adapter.py

@@ -575,12 +575,13 @@ class RenderAdapter:
 
         self._report_progress(progress_cb, 90.0, "生成封面缩略图")
 
-        # 6. 生成封面缩略图
+        # 6. 生成封面缩略图（结果通过 RenderAdapterResult.thumbnail_url 返回给调用方）
         thumbnail_url = ""
         try:
             from video_processing.thumbnail_generator import generate_and_upload_thumbnail
 
-            thumb_storage_key = f"rendered/{plan_id}/thumbnail.jpg"
+            # job_id 是 _do_render 方法的参数（参见方法签名）
+            thumb_storage_key = f"rendered/{plan_id}/thumbnails/{job_id}.jpg"
             thumbnail_url = generate_and_upload_thumbnail(str(result.output_path), thumb_storage_key)
         except Exception as thumb_err:
             logger.warning(
@@ -598,7 +599,7 @@ class RenderAdapter:
             # 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=""
+                str(result.output_path), plan_id, task_id=job_id, num_frames=3, title_text=""
             )
             if cover_candidates:
                 logger.info(

apps/worker/video_processing/render_audio.py

@@ -512,6 +512,9 @@ def mix_with_independent_audio(
                 clip_filters.append("asetpts=PTS-STARTPTS")
             else:
                 clip_filters.append("asetpts=PTS-STARTPTS")
+            vol = _clip_volume(clip)
+            if abs(vol - 1.0) >= 1e-6:
+                clip_filters.append(f"volume={vol:.4f}")
             # aformat 归一化：concat/amix 前统一音频参数，否则不同采样率/声道会失败
             clip_filters.append(AFORMAT)
             filter_parts.append(f"[{input_idx}:a]{','.join(clip_filters)}[ma{input_idx}]")

apps/worker/video_processing/thumbnail_generator.py

@@ -204,10 +204,81 @@ def generate_and_upload_thumbnail(
         Path(tmp.name).unlink(missing_ok=True)
 
 
+def _extract_frames_via_mediakit(
+    video_path: str,
+    plan_id: str,
+    num_frames: int,
+) -> list[dict] | None:
+    """使用 MediaKit 智能抽帧 API 提取封面帧。
+
+    Args:
+        video_path: 本地视频文件路径
+        plan_id: 编辑计划 ID
+        num_frames: 需要的帧数
+
+    Returns:
+        帧列表 [{"image_url": str, "timestamp": float}, ...]，失败返回 None
+    """
+    import uuid
+
+    from video_processing.oss_helpers import upload_to_oss
+
+    from packages.shared.mediakit_client import get_mediakit_client
+
+    client = get_mediakit_client()
+    if not client.is_available:
+        logger.info("[thumbnail] MediaKit 未配置，跳过智能抽帧")
+        return None
+
+    # 1. 上传视频到 OSS 获取 URL
+    try:
+        video_storage_key = f"temp/{plan_id}/{uuid.uuid4().hex[:8]}_{Path(video_path).name}"
+        video_url = upload_to_oss(video_path, video_storage_key)
+        if not video_url:
+            logger.warning("[thumbnail] 视频上传 OSS 失败，无法使用 MediaKit")
+            return None
+        logger.info("[thumbnail] 视频已上传 OSS: %s", video_url[:80])
+    except Exception as e:
+        logger.warning("[thumbnail] 视频上传 OSS 异常: %s，降级到 ffmpeg", e)
+        return None
+
+    # 2. 调用 MediaKit 智能抽帧
+    try:
+        frames = client.extract_frames(
+            video_url=video_url,
+            strategy="SceneChange",
+            max_frames=num_frames * 2,  # 多取一些帧供选择
+        )
+        if not frames:
+            logger.warning("[thumbnail] MediaKit 抽帧返回空，降级到 ffmpeg")
+            return None
+
+        # 选取最均匀的 num_frames 个帧
+        if len(frames) > num_frames:
+            step = len(frames) // num_frames
+            frames = [frames[i * step] for i in range(num_frames)]
+
+        logger.info("[thumbnail] MediaKit 抽帧成功: %d 帧", len(frames))
+        return frames
+
+    except Exception as e:
+        logger.warning("[thumbnail] MediaKit 抽帧异常: %s，降级到 ffmpeg", e)
+        return None
+    finally:
+        # 清理临时视频文件
+        try:
+            from video_processing.oss_helpers import delete_from_oss
+
+            delete_from_oss(video_storage_key)
+        except Exception:
+            pass
+
+
 def extract_and_upload_cover_frames(
     video_path: str,
     plan_id: str,
     *,
+    task_id: str = "",
     num_frames: int = 3,
     title_text: str = "",
     title_color: str = "#ffffff",
@@ -216,9 +287,12 @@ def extract_and_upload_cover_frames(
 ) -> list[dict]:
     """从视频中抽取多帧作为封面候选，上传到 OSS。
 
+    优先使用 MediaKit 智能抽帧，失败时降级到 ffmpeg 直接抽帧。
+
     Args:
         video_path: 视频文件路径
         plan_id: 编辑计划 ID（用于生成 storage key）
+        task_id: 任务 ID（用于生成独立的 storage key，避免标题变更时封面冲突）
         num_frames: 抽取帧数（默认 3）
         title_text: 标题文字；非空时用 Pillow 叠加到每帧。
             从已渲染视频抽帧时通常传空（标题已烧录）；从源素材抽帧时传标题。
@@ -229,6 +303,7 @@ def extract_and_upload_cover_frames(
     Returns:
         封面候选列表，每项包含 {"url": str, "position": float}
     """
+    import httpx
     from video_processing.ffmpeg_utils import probe_duration
     from video_processing.oss_helpers import upload_to_oss
 
@@ -238,6 +313,51 @@ def extract_and_upload_cover_frames(
         duration = 0.0
 
     candidates: list[dict] = []
+
+    # 优先尝试 MediaKit 智能抽帧
+    mediakit_frames = _extract_frames_via_mediakit(video_path, plan_id, num_frames)
+    if mediakit_frames:
+        for i, frame in enumerate(mediakit_frames):
+            frame_url = frame.get("image_url")
+            if not frame_url:
+                continue
+            tmp = tempfile.NamedTemporaryFile(suffix=".jpg", delete=False)
+            tmp.close()
+            try:
+                # 下载 MediaKit 返回的帧图
+                resp = httpx.get(frame_url, timeout=30, follow_redirects=True)
+                resp.raise_for_status()
+                with open(tmp.name, "wb") as f:
+                    f.write(resp.content)
+
+                # 叠加标题文字（如需要）
+                if title_text and title_text.strip():
+                    apply_title_overlay(
+                        tmp.name,
+                        title_text,
+                        color=title_color,
+                        position=title_position,
+                        font_size=title_font_size,
+                    )
+
+                storage_key = f"covers/{plan_id}/{task_id}/mediakit_frame_{i}.jpg"
+                url = upload_to_oss(tmp.name, storage_key)
+                if url:
+                    seek_time = frame.get("timestamp", 0.0)
+                    candidates.append({"url": url, "position": round(seek_time, 2)})
+            except Exception as e:
+                logger.warning("[thumbnail] MediaKit 帧 %d 处理失败: %s", i, e)
+            finally:
+                Path(tmp.name).unlink(missing_ok=True)
+
+        if len(candidates) >= num_frames:
+            logger.info("[thumbnail] MediaKit 智能抽帧完成: %d 帧", len(candidates))
+            return candidates[:num_frames]
+
+        logger.warning("[thumbnail] MediaKit 抽帧不足 %d 帧，降级到 ffmpeg", num_frames)
+
+    # Fallback: ffmpeg 直接抽帧
+    logger.info("[thumbnail] 使用 ffmpeg 抽帧")
     # 均匀分布抽帧点：从 10% 到 90%
     for i in range(num_frames):
         ratio = 0.1 + 0.8 * i / max(num_frames - 1, 1)
@@ -259,7 +379,7 @@ def extract_and_upload_cover_frames(
                     position=title_position,
                     font_size=title_font_size,
                 )
-            storage_key = f"covers/{plan_id}/frame_{i}.jpg"
+            storage_key = f"covers/{plan_id}/{task_id}/frame_{i}.jpg"
             url = upload_to_oss(frame_path, storage_key)
             if url:
                 seek_time = max(0.5, duration * ratio) if duration > 0 else 0.0

apps/worker/video_processing/unified_render_service.py

@@ -824,6 +824,17 @@ class UnifiedRenderService:
                 )
                 audio_layer.clips.append(vo_clip)
 
+            # replace 模式：静音原视频音轨（main + broll 图层）
+            if tts_config.overlap_mode == "replace":
+                for layer in layers:
+                    if layer.role in ("main", "broll"):
+                        for clip in layer.clips:
+                            clip.config["volume"] = 0
+                logger.info(
+                    "TTS replace 模式：已静音原视频音轨: plan_id=%s",
+                    self.plan.id,
+                )
+
             logger.info(
                 "TTS 配音已添加: plan_id=%s voice_id=%s segments=%d total_%.2fs",
                 self.plan.id,
@@ -889,6 +900,16 @@ class UnifiedRenderService:
             )
             audio_layer.clips.append(vo_clip)
 
+            # 配音素材库默认替换原音：静音原视频音轨（main + broll 图层）
+            for layer in layers:
+                if layer.role in ("main", "broll"):
+                    for clip in layer.clips:
+                        clip.config["volume"] = 0
+            logger.info(
+                "配音素材库：已静音原视频音轨: plan_id=%s",
+                self.plan.id,
+            )
+
             logger.info(
                 "配音素材库音频已添加到 audio 图层: plan_id=%s duration=%.2fs",
                 self.plan.id,

apps/worker/worker_app/tasks/__init__.py

@@ -49,10 +49,15 @@ def __getattr__(name: str):
         from .ai_tasks import run_generate_cover
 
         return run_generate_cover
+    elif name == "batch_generate_thumbnails":
+        from .batch_thumbnail import batch_generate_thumbnails
+
+        return batch_generate_thumbnails
     raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
 
 
 __all__ = [
+    "batch_generate_thumbnails",
     "classify_asset",
     "generate_video",
     "healthcheck",

apps/worker/worker_app/tasks/batch_thumbnail.py

@@ -0,0 +1,165 @@
+"""批量修复素材缩略图 — 为历史视频素材生成缩略图。
+
+使用方式：
+  从管理接口或 shell 触发：
+    celery_app.send_task("worker.batch_generate_thumbnails")
+
+逻辑：
+  1. 查询所有 file_type=video 且 thumbnail_url 为空或为旧格式公开 URL 的素材
+  2. 逐个：下载视频 → 抽第一帧 → 上传 OSS → 更新 thumbnail_url 为 storage_key
+  3. 每处理 50 条 commit 一次，失败单条跳过不阻塞
+"""
+
+from __future__ import annotations
+
+import tempfile
+from pathlib import Path
+
+from celery.utils.log import get_task_logger
+from worker_app.celery_app import celery_app
+from worker_app.db import SessionLocal
+
+logger = get_task_logger(__name__)
+
+# 旧格式 URL 前缀（ingest 旧代码生成的公开 URL），需替换为 storage_key
+_OLD_URL_PREFIX = "https://xiaoxia-autocut.oss-cn-hangzhou.aliyuncs.com/"
+# 从公开 URL 中提取 storage_key 时，去掉域名前缀即可
+_DOMAIN_PREFIXES = [
+    "https://xiaoxia-autocut.oss-cn-hangzhou.aliyuncs.com/",
+    "http://xiaoxia-autocut.oss-cn-hangzhou.aliyuncs.com/",
+]
+
+
+def _url_to_storage_key(url: str) -> str | None:
+    """尝试将旧格式公开 URL 转回 storage_key。"""
+    for prefix in _DOMAIN_PREFIXES:
+        if url.startswith(prefix):
+            return url[len(prefix) :]
+    return None
+
+
+@celery_app.task(name="worker.batch_generate_thumbnails")
+def batch_generate_thumbnails() -> dict:
+    """为所有缺少缩略图的视频素材批量生成缩略图。
+
+    Returns:
+        dict: {total, success, skipped, failed, converted_legacy}
+    """
+    from video_processing.oss_helpers import download_asset, upload_to_oss
+    from video_processing.thumbnail_generator import extract_first_frame
+
+    from packages.adapters.sqlalchemy_impl.models import AssetModel
+
+    db = SessionLocal()
+    stats = {"total": 0, "success": 0, "skipped": 0, "failed": 0, "converted_legacy": 0}
+
+    try:
+        # 查询所有视频素材中缩略图缺失的
+        assets = (
+            db.query(AssetModel)
+            .filter(
+                AssetModel.file_type == "video",
+                AssetModel.status != "deleted",
+            )
+            .all()
+        )
+
+        # 筛选需要处理的：thumbnail_url 为空 或 旧格式公开 URL
+        to_process = []
+        for asset in assets:
+            thumb = asset.thumbnail_url or ""
+            if not thumb:
+                to_process.append((asset, None))  # (asset, None=需要新生成)
+            elif thumb.startswith("http"):
+                # 旧格式公开 URL → 尝试转为 storage_key
+                sk = _url_to_storage_key(thumb)
+                if sk:
+                    to_process.append((asset, sk))  # 已有文件，只需改 DB
+                else:
+                    # 非预期 URL 格式，跳过
+                    stats["skipped"] += 1
+            # else: 已经是 storage_key 格式，跳过
+
+        stats["total"] = len(to_process)
+        logger.info(
+            "批量缩略图修复启动: total=%d (new=%d, legacy_convert=%d)",
+            stats["total"],
+            sum(1 for _, sk in to_process if sk is None),
+            sum(1 for _, sk in to_process if sk is not None),
+        )
+
+        batch_count = 0
+        for asset, existing_key in to_process:
+            try:
+                if existing_key is not None:
+                    # 旧 URL → storage_key，只需更新 DB
+                    asset.thumbnail_url = existing_key
+                    stats["converted_legacy"] += 1
+                    stats["success"] += 1
+                else:
+                    # 需要新生成缩略图
+                    if not asset.storage_key:
+                        stats["skipped"] += 1
+                        continue
+
+                    suffix = Path(asset.storage_key).suffix or ".mp4"
+                    local_file = None
+                    frame_path = None
+                    try:
+                        # 下载视频
+                        with tempfile.NamedTemporaryFile(suffix=suffix, delete=False) as tmp:
+                            local_file = Path(tmp.name)
+
+                        if not download_asset(asset.storage_key, local_file):
+                            logger.warning("下载失败: asset_id=%s key=%s", asset.id, asset.storage_key[:60])
+                            stats["failed"] += 1
+                            continue
+
+                        # 抽帧
+                        frame_path = extract_first_frame(str(local_file), width=640)
+
+                        # 上传
+                        thumb_key = f"assets/{asset.project_id}/thumbnails/{asset.id}.jpg"
+                        upload_ok = upload_to_oss(frame_path, thumb_key)
+                        if upload_ok:
+                            asset.thumbnail_url = thumb_key
+                            stats["success"] += 1
+                        else:
+                            logger.warning("上传失败: asset_id=%s", asset.id)
+                            stats["failed"] += 1
+                    finally:
+                        if local_file and local_file.exists():
+                            try:
+                                local_file.unlink(missing_ok=True)
+                            except OSError:
+                                pass
+                        if frame_path and Path(frame_path).exists():
+                            try:
+                                Path(frame_path).unlink(missing_ok=True)
+                            except OSError:
+                                pass
+
+                batch_count += 1
+                if batch_count % 50 == 0:
+                    db.commit()
+                    logger.info("批量缩略图进度: %d/%d", batch_count, stats["total"])
+
+            except Exception as e:
+                logger.warning("单条处理失败: asset_id=%s error=%s", asset.id, e)
+                stats["failed"] += 1
+                try:
+                    db.rollback()
+                except Exception:
+                    pass
+
+        # 最后提交
+        db.commit()
+        logger.info("批量缩略图修复完成: %s", stats)
+        return stats
+
+    except Exception as e:
+        logger.error("批量缩略图修复异常: %s", e)
+        db.rollback()
+        return {**stats, "error": str(e)}
+    finally:
+        db.close()

apps/worker/worker_app/tasks/generation.py

@@ -389,6 +389,7 @@ def _upload_and_record(
     editing_mode,
     user_id: str = "",
     video_name: str = "",
+    thumbnail_url: str = "",
 ) -> tuple[str, float, int, int]:
     """上传 OSS、创建视频记录并查重。
 
@@ -447,6 +448,7 @@ def _upload_and_record(
             mode=editing_mode.value,
             session=dedup_session,
             name=video_name,
+            thumbnail_url=thumbnail_url,
         )
     finally:
         dedup_session.close()
@@ -535,11 +537,11 @@ def _render_from_edit_plan(
     task_id: str,
     source_edit_plan_id: str,
     task_info: dict,
-) -> tuple[Path, float, list[dict] | None, str | None, str | None]:
+) -> tuple[Path, float, list[dict] | None, str | None, str | None, str]:
     """从 EditPlan 数据库记录直接渲染（不再内存重建clips）。
 
     Returns:
-        (output_path, render_duration, cover_candidates, voiceover_path, temp_dir)
+        (output_path, render_duration, cover_candidates, voiceover_path, temp_dir, thumbnail_url)
     """
     from video_processing.render_adapter import RenderAdapter
     from worker_app.db import SessionLocal
@@ -580,7 +582,14 @@ def _render_from_edit_plan(
         cover_candidates = getattr(result, "cover_candidates", None)
         render_temp_dir = getattr(result, "temp_dir", None)
 
-        return output_path, result.duration, cover_candidates, voiceover_path, render_temp_dir
+        return (
+            output_path,
+            result.duration,
+            cover_candidates,
+            voiceover_path,
+            render_temp_dir,
+            result.thumbnail_url or "",
+        )
     finally:
         db.close()
 
@@ -692,7 +701,7 @@ def generate_video(self, task_id: str) -> dict:
                     gen_task.append_log("渲染模式", "从草稿数据渲染（与预览一致）")
                     _flush_logs(task_id, gen_task)
 
-                output_path, render_duration, cover_candidates, voiceover_tmp_path, render_temp_dir = (
+                output_path, render_duration, cover_candidates, voiceover_tmp_path, render_temp_dir, thumbnail_url = (
                     _render_from_edit_plan(
                         task_id=task_id,
                         source_edit_plan_id=source_edit_plan_id,
@@ -717,6 +726,7 @@ def generate_video(self, task_id: str) -> dict:
                         editing_mode=editing_mode,
                         user_id=user_id,
                         video_name=task_info.get("video_title", ""),
+                        thumbnail_url=thumbnail_url,
                     )
 
                     if gen_task:

apps/worker/worker_app/tasks/ingest.py

@@ -214,20 +214,26 @@ def ingest_asset(job_id: str) -> dict:
 
                     frame_path = extract_first_frame(str(local_file), width=640)
                     thumb_storage_key = f"assets/{job.project_id}/thumbnails/{job_id}.jpg"
-                    try:
-                        thumbnail_url = upload_to_oss(frame_path, thumb_storage_key)
-                    finally:
-                        if frame_path:
-                            try:
-                                Path(frame_path).unlink(missing_ok=True)
-                            except Exception:
-                                pass
-                    if thumbnail_url:
+                    upload_ok = upload_to_oss(frame_path, thumb_storage_key)
+                    if upload_ok:
+                        # 存储 storage_key（非公开 URL），API 层通过 get_download_url 生成签名 URL
+                        thumbnail_url = thumb_storage_key
                         logger.info(
-                            "素材缩略图生成成功: job_id=%s url=%s",
+                            "素材缩略图生成成功: job_id=%s key=%s",
                             job_id,
-                            thumbnail_url[:80],
+                            thumb_storage_key[:80],
                         )
+                    else:
+                        logger.warning(
+                            "素材缩略图上传 OSS 失败: job_id=%s",
+                            job_id,
+                        )
+                    # frame_path 是临时文件，及时清理
+                    if frame_path:
+                        try:
+                            Path(frame_path).unlink(missing_ok=True)
+                        except Exception:
+                            pass
                 except Exception as thumb_err:
                     logger.warning(
                         "素材缩略图生成失败（不影响主流程）: job_id=%s error=%s",

infra/docker/api-base.Dockerfile

@@ -12,14 +12,22 @@ RUN sed -i 's|deb.debian.org|mirrors.aliyun.com|g' /etc/apt/sources.list.d/debia
     sed -i 's|deb.debian.org|mirrors.aliyun.com|g' /etc/apt/sources.list 2>/dev/null || true
 
 # 预装系统依赖（gcc 编译 psycopg/pg 扩展，libpq-dev 编译期，libpq5 运行期，ffmpeg 封面取帧）
+# 字体修复：fonts-noto-cjk 包的 .ttc 文件混入了 Mono 变体，导致 Bold 匹配到等宽字体
+# 解决方案：删除有问题的 .ttc，使用仓库内预下载的 Noto Sans SC Variable Font（不含 Mono）
 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/*
+    && rm -rf /var/lib/apt/lists/* \
+    # 删除有问题的 .ttc 文件（包含 Mono 变体）
+    && rm -f /usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc \
+    && rm -f /usr/share/fonts/opentype/noto/NotoSansCJK-Bold.ttc
+
+# 复制预下载的 Noto Sans SC Variable Font（包含所有字重，不含 Mono 变体）
+COPY infra/fonts/NotoSansSC-VF.ttf /usr/share/fonts/opentype/noto/NotoSansSC-VF.ttf
+RUN fc-cache -fv
 
 # 创建虚拟环境
 RUN python -m venv /opt/venv
@@ -43,3 +51,5 @@ RUN find /opt/venv -type d -name "__pycache__" -exec rm -rf {} + 2>/dev/null; \
 RUN rm -f /tmp/requirements-base.txt /tmp/requirements.txt
 
 ENV PYTHONPATH=/app
+
+# Rebuild trigger: ACR push retry 20260826

infra/docker/worker-base.Dockerfile

@@ -12,6 +12,8 @@ RUN sed -i 's|deb.debian.org|mirrors.aliyun.com|g' /etc/apt/sources.list.d/debia
     sed -i 's|deb.debian.org|mirrors.aliyun.com|g' /etc/apt/sources.list 2>/dev/null || true
 
 # 预装系统依赖（编译工具 + 运行时 + CJK 字体用于 ASS 字幕渲染）
+# 字体修复：fonts-noto-cjk 包的 .ttc 文件混入了 Mono 变体，导致 Bold 匹配到等宽字体
+# 解决方案：删除有问题的 .ttc，使用仓库内预下载的 Noto Sans SC Variable Font（不含 Mono）
 RUN apt-get update && apt-get install -y --no-install-recommends \
     gcc \
     g++ \
@@ -20,8 +22,14 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
     ffmpeg \
     libglib2.0-0 \
     fonts-noto-cjk \
-    && fc-cache -fv \
-    && rm -rf /var/lib/apt/lists/*
+    && rm -rf /var/lib/apt/lists/* \
+    # 删除有问题的 .ttc 文件（包含 Mono 变体）
+    && rm -f /usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc \
+    && rm -f /usr/share/fonts/opentype/noto/NotoSansCJK-Bold.ttc
+
+# 复制预下载的 Noto Sans SC Variable Font（包含所有字重，不含 Mono 变体）
+COPY infra/fonts/NotoSansSC-VF.ttf /usr/share/fonts/opentype/noto/NotoSansSC-VF.ttf
+RUN fc-cache -fv
 
 # 创建虚拟环境
 RUN python -m venv /opt/venv
@@ -47,3 +55,5 @@ RUN find /opt/venv -type d -name "__pycache__" -exec rm -rf {} + 2>/dev/null; \
 RUN rm -f /tmp/requirements-base.txt /tmp/requirements.txt /tmp/requirements-worker.txt
 
 ENV PYTHONPATH=/app:/app/packages
+
+# Rebuild trigger: ACR push retry 20260826

packages/domain/ass_subtitle_builder.py

@@ -20,10 +20,32 @@ logger = logging.getLogger(__name__)
 # ── 常量 ──────────────────────────────────────────────────────────────────────
 
 # Title/Subtitle 默认边距（像素）
-TITLE_MARGIN_TOP = 60
+TITLE_MARGIN_TOP = 120
 TITLE_MARGIN_BOTTOM = 60
 TITLE_MARGIN_SIDE = 40
 
+# 字体名称映射：前端中文字体名 → 服务器实际注册名（ffmpeg/ASS 通过注册名匹配字体）
+FONT_NAME_MAP: dict[str, str] = {
+    "思源黑体": "Noto Sans SC",
+    "思源宋体": "Noto Serif CJK SC",
+    "苹方": "Noto Sans SC",
+    "PingFang": "Noto Sans SC",
+    "微软雅黑": "Noto Sans SC",
+    "楷体": "Noto Serif CJK SC",
+    "华康俪金黑": "Noto Sans SC",
+}
+
+# ASS Fontsize 是字体 em-square 高度（含 Latin 升降部留白），
+# 中文字符实际只占声明字号的约 65%~75%；浏览器 CSS font-size 让中文字符占满声明高度。
+# 为让成片中文字高与前端 CSS 预览一致，写入 ASS 时对字号乘以补偿系数。
+# font_size=89 → ASS Fontsize=round(89*1.35)=120，实际中文字高约 78~85px。
+ASS_FONTSIZE_COMPENSATION = 1.35
+
+
+def _compensate_ass_fontsize(font_size: int) -> int:
+    """将 CSS 语义字号换算为 ASS Fontsize，补偿中文字符在 em-square 中的留白。"""
+    return max(1, round(font_size * ASS_FONTSIZE_COMPENSATION))
+
 
 # ── 颜色转换 ──────────────────────────────────────────────────────────────────
 
@@ -117,14 +139,20 @@ def build_ass_style(
     bold_val = -1 if bold else 0
     italic_val = -1 if italic else 0
 
+    # 字体名称映射：前端中文字体名 → 服务器注册名，未命中则原样使用
+    actual_font = FONT_NAME_MAP.get(font_name, font_name)
+
     # BackColour 用于阴影（BorderStyle=1 时 outline + shadow）
     back_color = primary_color
 
     # Shadow 深度：shadow_offset[1] 作为纵向偏移
     shadow_depth = shadow_offset[1] if shadow_blur > 0 else 0
 
+    # 写入 ASS Style 时对字号做补偿，使成片中文字高与前端 CSS 预览一致
+    ass_font_size = _compensate_ass_fontsize(font_size)
+
     return (
-        f"Style: {style_name},{font_name},{font_size},{primary_color},"
+        f"Style: {style_name},{actual_font},{ass_font_size},{primary_color},"
         f"&H000000FF,{outline_color},{back_color},"
         f"{bold_val},{italic_val},0,0,100,100,0,0,"
         f"1,{outline_width},{shadow_depth},{alignment},"
@@ -147,6 +175,8 @@ def escape_ass_text(text: str) -> str:
     Returns:
         转义后的 ASS 文本
     """
+    # 用户手动换行符（半角/全角斜杠）转为 ASS 硬换行（在自动换行之前优先处理）
+    text = text.replace("／", "\\N").replace("/", "\\N")
     # 将实际换行转为 ASS 硬换行
     text = text.replace("\r\n", "\\N").replace("\n", "\\N").replace("\r", "\\N")
     # 转义大括号（ASS 用它做样式覆盖标签）
@@ -175,7 +205,6 @@ def format_ass_time(seconds: float) -> str:
 # ── 完整 ASS 内容生成 ─────────────────────────────────────────────────────────
 
 
-
 def _wrap_title_text(
     text: str,
     video_width: int,
@@ -195,26 +224,36 @@ def _wrap_title_text(
     if available_width <= 0:
         return text
 
-    lines: list[str] = []
-    current_line = ""
-    current_width = 0.0
+    # 换行计算使用原始 font_size，与 CSS 预览一致；1.35x 补偿仅用于 ASS Fontsize 渲染
 
-    for ch in text:
-        # CJK 字符按全角估算，其他按半角
-        char_width = float(font_size) if ord(ch) > 0x2E80 else font_size * 0.55
 
-        if current_width + char_width > available_width and current_line:
+    # 先按已有 \N 分段，每段独立自动换行，最后用 \N 拼回
+    segments = text.split("\\N")
+    wrapped_segments: list[str] = []
+
+    for seg in segments:
+        lines: list[str] = []
+        current_line = ""
+        current_width = 0.0
+
+        for ch in seg:
+            # CJK 字符按全角估算，其他按半角
+            char_width = float(font_size) if ord(ch) > 0x2E80 else font_size * 0.55
+
+            if current_width + char_width > available_width and current_line:
+                lines.append(current_line)
+                current_line = ch
+                current_width = char_width
+            else:
+                current_line += ch
+                current_width += char_width
+
+        if current_line:
             lines.append(current_line)
-            current_line = ch
-            current_width = char_width
-        else:
-            current_line += ch
-            current_width += char_width
 
-    if current_line:
-        lines.append(current_line)
+        wrapped_segments.append("\\N".join(lines))
 
-    return "\\N".join(lines)
+    return "\\N".join(wrapped_segments)
 
 
 def build_ass_content(
@@ -247,26 +286,40 @@ def build_ass_content(
     title_config = title_config or {}
     subtitle_config = subtitle_config or {}
 
+    # ── 字段名归一化：前端传 font_size/font_color，内部用 size/color ──
+    if "font_size" in title_config and "size" not in title_config:
+        title_config["size"] = title_config["font_size"]
+    if "font_color" in title_config and "color" not in title_config:
+        title_config["color"] = title_config["font_color"]
+
     # ── 兼容前端简化格式：stroke/shadow 为 boolean 时，转换为标准 dict ──
     # 前端 TitleSettings 发送 stroke=true/false, shadow=true/false
     # 后端 build_ass_style 期望 stroke={enabled, color, width}, shadow={enabled, blur, offset_x, offset_y}
     if title_config:
         _stroke_val = title_config.get("stroke")
         if isinstance(_stroke_val, bool):
-            title_config["stroke"] = {
-                "enabled": _stroke_val,
-                "color": "#000000",
-                "width": 2,
-            } if _stroke_val else {"enabled": False}
+            title_config["stroke"] = (
+                {
+                    "enabled": _stroke_val,
+                    "color": "#000000",
+                    "width": 2,
+                }
+                if _stroke_val
+                else {"enabled": False}
+            )
         _shadow_val = title_config.get("shadow")
         if isinstance(_shadow_val, bool):
-            title_config["shadow"] = {
-                "enabled": _shadow_val,
-                "color": "#000000",
-                "blur": 4,
-                "offset_x": 2,
-                "offset_y": 2,
-            } if _shadow_val else {"enabled": False}
+            title_config["shadow"] = (
+                {
+                    "enabled": _shadow_val,
+                    "color": "#000000",
+                    "blur": 4,
+                    "offset_x": 2,
+                    "offset_y": 2,
+                }
+                if _shadow_val
+                else {"enabled": False}
+            )
 
     title_enabled = title_config.get("enabled", True) and bool(title_text.strip())
     subtitle_enabled = subtitle_config.get("enabled", True) and bool(subtitle_text.strip())
@@ -296,7 +349,7 @@ def build_ass_content(
             build_ass_style(
                 "TitleStyle",
                 font_name=title_config.get("font", "思源黑体"),
-                font_size=min(int(title_config.get("size", 36)), 36),
+                font_size=int(title_config.get("size", 36)),
                 primary_color=title_color,
                 outline_color=stroke_color,
                 outline_width=stroke_width,
@@ -313,7 +366,7 @@ def build_ass_content(
 
         # 根据视频宽度和字号自动换行标题，防止超出画面
         # 先 escape 特殊字符，再插入换行符 \N，避免顺序颠倒导致 \N 被转义
-        title_font_size = min(int(title_config.get("size", 36)), 36)
+        title_font_size = int(title_config.get("size", 36))
         safe_title_text_raw = escape_ass_text(title_text)
         safe_title_text = _wrap_title_text(safe_title_text_raw, video_width, title_font_size)
 

packages/domain/plan_generator_utils.py

@@ -77,12 +77,18 @@ def _distribute_one_take(
     asset_durations: dict[str, float] | None = None,
 ) -> None:
     """ONE_TAKE: 素材按顺序依次分配给 main 类型 clips."""
+    used_segments: dict[str, list[tuple[float, float]]] = {}
     main_clips = [c for c in clips if c.clip_type == ClipType.MAIN.value]
     for i, clip in enumerate(main_clips):
         if i < len(asset_ids):
             asset_id = asset_ids[i]
-            start_time = _calc_random_start_time(asset_id, clip.duration, asset_durations)
+            start_time = _calc_random_start_time(asset_id, clip.duration, asset_durations, used_segments)
             clip.assign_asset(asset_id, start_time=start_time)
+            # Record used segment
+            if start_time is not None and asset_durations is not None:
+                if asset_id not in used_segments:
+                    used_segments[asset_id] = []
+                used_segments[asset_id].append((start_time, start_time + clip.duration))
 
 
 def _distribute_pip(
@@ -91,12 +97,18 @@ def _distribute_pip(
     asset_durations: dict[str, float] | None = None,
 ) -> None:
     """PIP: 第1个素材→main(全屏背景)，其余→overlay clips."""
+    used_segments: dict[str, list[tuple[float, float]]] = {}
     # 第1个素材 → main clip
     main_clips = [c for c in clips if c.clip_type == ClipType.MAIN.value]
     if main_clips and asset_ids:
         asset_id = asset_ids[0]
-        start_time = _calc_random_start_time(asset_id, main_clips[0].duration, asset_durations)
+        start_time = _calc_random_start_time(asset_id, main_clips[0].duration, asset_durations, used_segments)
         main_clips[0].assign_asset(asset_id, start_time=start_time)
+        # Record used segment
+        if start_time is not None and asset_durations is not None:
+            if asset_id not in used_segments:
+                used_segments[asset_id] = []
+            used_segments[asset_id].append((start_time, start_time + main_clips[0].duration))
 
     # 其余素材 → overlay clips
     overlay_clips = [c for c in clips if c.clip_type == "overlay"]
@@ -104,8 +116,13 @@ def _distribute_pip(
     for i, clip in enumerate(overlay_clips):
         if i < len(remaining):
             asset_id = remaining[i]
-            start_time = _calc_random_start_time(asset_id, clip.duration, asset_durations)
+            start_time = _calc_random_start_time(asset_id, clip.duration, asset_durations, used_segments)
             clip.assign_asset(asset_id, start_time=start_time)
+            # Record used segment
+            if start_time is not None and asset_durations is not None:
+                if asset_id not in used_segments:
+                    used_segments[asset_id] = []
+                used_segments[asset_id].append((start_time, start_time + clip.duration))
 
 
 def _distribute_voice_over(
@@ -114,12 +131,18 @@ def _distribute_voice_over(
     asset_durations: dict[str, float] | None = None,
 ) -> None:
     """VOICE_OVER: 素材→main clips (B-roll)."""
+    used_segments: dict[str, list[tuple[float, float]]] = {}
     main_clips = [c for c in clips if c.clip_type == ClipType.MAIN.value]
     for i, clip in enumerate(main_clips):
         if i < len(asset_ids):
             asset_id = asset_ids[i]
-            start_time = _calc_random_start_time(asset_id, clip.duration, asset_durations)
+            start_time = _calc_random_start_time(asset_id, clip.duration, asset_durations, used_segments)
             clip.assign_asset(asset_id, start_time=start_time)
+            # Record used segment
+            if start_time is not None and asset_durations is not None:
+                if asset_id not in used_segments:
+                    used_segments[asset_id] = []
+                used_segments[asset_id].append((start_time, start_time + clip.duration))
 
 
 def _distribute_voice_pip(
@@ -128,6 +151,7 @@ def _distribute_voice_pip(
     asset_durations: dict[str, float] | None = None,
 ) -> None:
     """VOICE_PIP: 第1个→background, 第2个→corner_voice, 其余→b_roll."""
+    used_segments: dict[str, list[tuple[float, float]]] = {}
     bg_clips = [c for c in clips if c.clip_type == "background"]
     voice_clips = [c for c in clips if c.clip_type == "corner_voice"]
     broll_clips = [c for c in clips if c.clip_type == "b_roll"]
@@ -137,15 +161,25 @@ def _distribute_voice_pip(
     # 第1个 → background
     if idx < len(asset_ids) and bg_clips:
         asset_id = asset_ids[idx]
-        start_time = _calc_random_start_time(asset_id, bg_clips[0].duration, asset_durations)
+        start_time = _calc_random_start_time(asset_id, bg_clips[0].duration, asset_durations, used_segments)
         bg_clips[0].assign_asset(asset_id, start_time=start_time)
+        # Record used segment
+        if start_time is not None and asset_durations is not None:
+            if asset_id not in used_segments:
+                used_segments[asset_id] = []
+            used_segments[asset_id].append((start_time, start_time + bg_clips[0].duration))
         idx += 1
 
     # 第2个 → corner_voice
     if idx < len(asset_ids) and voice_clips:
         asset_id = asset_ids[idx]
-        start_time = _calc_random_start_time(asset_id, voice_clips[0].duration, asset_durations)
+        start_time = _calc_random_start_time(asset_id, voice_clips[0].duration, asset_durations, used_segments)
         voice_clips[0].assign_asset(asset_id, start_time=start_time)
+        # Record used segment
+        if start_time is not None and asset_durations is not None:
+            if asset_id not in used_segments:
+                used_segments[asset_id] = []
+            used_segments[asset_id].append((start_time, start_time + voice_clips[0].duration))
         idx += 1
 
     # 剩余 → b_roll clips
@@ -153,8 +187,13 @@ def _distribute_voice_pip(
     for i, clip in enumerate(broll_clips):
         if i < len(remaining):
             asset_id = remaining[i]
-            start_time = _calc_random_start_time(asset_id, clip.duration, asset_durations)
+            start_time = _calc_random_start_time(asset_id, clip.duration, asset_durations, used_segments)
             clip.assign_asset(asset_id, start_time=start_time)
+            # Record used segment
+            if start_time is not None and asset_durations is not None:
+                if asset_id not in used_segments:
+                    used_segments[asset_id] = []
+                used_segments[asset_id].append((start_time, start_time + clip.duration))
 
 
 # ── 随机 start_time 计算 ────────────────────────────────────────────────────
@@ -164,16 +203,19 @@ def _calc_random_start_time(
     asset_id: str,
     clip_duration: float,
     asset_durations: dict[str, float] | None,
+    used_segments: dict[str, list[tuple[float, float]]] | None = None,
 ) -> float | None:
-    """计算随机 start_time.
+    """计算随机 start_time，避开已使用的时间段.
 
     在素材总时长范围内随机取点，确保 clip_duration 不超出素材边界。
     如果 asset_durations 为 None 或素材不在其中，返回 None（使用默认 0.0）。
+    如果提供了 used_segments，会避开已使用的时间区间。
 
     Args:
         asset_id: 素材 ID
         clip_duration: 片段时长（秒）
         asset_durations: 素材 ID -> 时长映射
+        used_segments: {asset_id: [(start1, end1), (start2, end2), ...]} 已使用的时间段
 
     Returns:
         随机 start_time 或 None
@@ -190,7 +232,41 @@ def _calc_random_start_time(
     if max_start <= 0:
         return 0.0
 
-    return random.uniform(0.0, max_start)
+    # 如果没有已使用段，直接随机
+    if not used_segments or asset_id not in used_segments:
+        return random.uniform(0.0, max_start)
+
+    # 尝试找到一个不与已使用段重叠的起始点
+    used = sorted(used_segments[asset_id])
+    max_attempts = 100
+
+    for _ in range(max_attempts):
+        candidate = random.uniform(0.0, max_start)
+        candidate_end = candidate + clip_duration
+
+        # 检查是否与任何已使用段重叠
+        overlap = False
+        for seg_start, seg_end in used:
+            # 两个区间 [a, b] 和 [c, d] 重叠的条件: a < d and c < b
+            if candidate < seg_end and seg_start < candidate_end:
+                overlap = True
+                break
+
+        if not overlap:
+            return candidate
+
+    # 如果尝试多次仍找不到，缩短时长使用素材末尾
+    # 找到最后一个已使用段之后的可用空间
+    last_used_end = 0.0
+    for _seg_start, seg_end in used:
+        last_used_end = max(last_used_end, seg_end)
+
+    if last_used_end < total_duration:
+        # 返回从最后使用点开始的位置
+        return min(last_used_end, max_start)
+
+    # 实在没有空间，返回0（可能会重叠，但至少能执行）
+    return 0.0
 
 
 # ── clip_type 映射 ────────────────────────────────────────────────────────

packages/shared/title_overlay.py

@@ -15,6 +15,7 @@ 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/NotoSansSC-VF.ttf",
     "/usr/share/fonts/opentype/noto/NotoSansCJK-Bold.ttc",
     "/usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc",
     "/usr/share/fonts/truetype/noto/NotoSansCJK-Bold.ttc",

scripts/ci/acr_cleanup.py

@@ -398,7 +398,6 @@ def cleanup_repo(repo, keep_count, dry_run, protected_tags, pr_sha=None, pr_open
             # 注意：pr tag是pr-<sha>，sha可能属于某个PR
             # 简化策略：收集所有打开PR的commit sha，在白名单里的保留
             print("    模式: Gitea PR状态检查")
-            open_pr_shas = set()
             # 这里做了简化：因为每个PR都查commits太慢，我们用另一种方式
             # 对于PR tag，先尝试匹配PR编号（如果tag名里有编号），否则按时间
             # 实际pr-<sha>没法直接知道PR编号，所以降级为按时间+打开PR的head sha白名单

scripts/ci/auto_fix_formatting.py

@@ -232,7 +232,7 @@ def main():
         # 初始化git仓库
         ensure_git_repo_for_push(api_url, repo, token, branch_name)
         head_branch = branch_name
-        fix_mode = "auto_fix_and_push"
+        # fix_mode removed: all PRs auto-fix (2026-08-26)
 
     # ====== PR事件处理 ======
     elif event_name == "pull_request":
@@ -251,7 +251,7 @@ def main():
 
         # 防循环检测：检查最新commit是否已经是格式修复commit
         # 修复commit message 带 [skip ci-format-check] 标记，检测到则跳过
-        head_branch_tmp = pr_info.get("head", {}).get("ref", "")
+        _head_branch_tmp = pr_info.get("head", {}).get("ref", "")  # noqa: F841
         skip_marker = "[skip ci-format-check]"
         try:
             commits_url = f"{api_url}/repos/{repo}/pulls/{pr_number}/commits?limit=3"
@@ -268,7 +268,7 @@ def main():
 
         # 所有PR都自动修复格式（不再区分人/Agent）
         print("检测到格式问题，将自动修复并推送回分支")
-        fix_mode = "auto_fix_and_push"
+        # fix_mode removed: all PRs auto-fix (2026-08-26)
 
         print("=== 检测到代码格式问题，尝试自动修复 ===")
         print(f"PR #{pr_number}")

tests/unit/domain/test_ass_subtitle_builder.py

@@ -15,6 +15,7 @@ from packages.domain.ass_subtitle_builder import (
     TITLE_MARGIN_BOTTOM,
     TITLE_MARGIN_SIDE,
     TITLE_MARGIN_TOP,
+    _wrap_title_text,
     build_ass_content,
     build_ass_style,
     escape_ass_text,
@@ -105,9 +106,9 @@ class TestBuildAssStyle:
 
     def test_contains_font_size(self):
         result = build_ass_style("S1", font_size=36)
-        # Style行格式：Name, Fontname, Fontsize, ...
+        # Style行格式：Name, Fontname, Fontsize, ...（36*1.35=48.6→49）
         parts = result.split(",")
-        assert parts[2] == "36"
+        assert parts[2] == "49"
 
     def test_bold_true(self):
         result = build_ass_style("S1", bold=True)
@@ -218,6 +219,14 @@ class TestEscapeAssText:
     def test_chinese_text(self):
         assert escape_ass_text("你好世界") == "你好世界"
 
+    def test_slash_converted_to_newline(self):
+        """半角斜杠 / 应转为 ASS 硬换行。"""
+        assert escape_ass_text("标题一/标题二") == "标题一\\N标题二"
+
+    def test_fullwidth_slash_converted_to_newline(self):
+        """全角斜杠 ／ 应转为 ASS 硬换行。"""
+        assert escape_ass_text("标题一／标题二") == "标题一\\N标题二"
+
     def test_backslash_n_in_input(self):
         # 文本里本身有 \n 字符串（不是换行符）
         result = escape_ass_text("\\n")
@@ -408,11 +417,53 @@ class TestBuildAssContent:
             title_text="T",
             title_config={"size": 72},
         )
-        # 在TitleStyle行里查找字体大小
+        # 在TitleStyle行里查找字体大小（字号上限已移除，72应原样保留）
         for line in result.split("\n"):
             if line.startswith("Style: TitleStyle"):
                 parts = line.split(",")
-                assert parts[2] == "36"
+                assert parts[2] == "97"  # 72*1.35=97.2→97
+                break
+
+    def test_title_font_size_frontend_field_alias(self):
+        """前端传 font_size 应归一化为内部 size 字段。"""
+        result = build_ass_content(
+            video_width=1920,
+            video_height=1080,
+            video_duration=5.0,
+            title_text="T",
+            title_config={"font_size": 48},
+        )
+        for line in result.split("\n"):
+            if line.startswith("Style: TitleStyle"):
+                parts = line.split(",")
+                assert parts[2] == "65"  # 48*1.35=64.8→65
+                break
+
+    def test_title_font_color_frontend_field_alias(self):
+        """前端传 font_color 应归一化为内部 color 字段。"""
+        result = build_ass_content(
+            video_width=1920,
+            video_height=1080,
+            video_duration=5.0,
+            title_text="T",
+            title_config={"font_color": "#FF0000"},
+        )
+        # 红色 → &H0000FF
+        assert "&H0000FF" in result
+
+    def test_title_size_takes_precedence_over_font_size(self):
+        """同时传 size 和 font_size 时，size 优先。"""
+        result = build_ass_content(
+            video_width=1920,
+            video_height=1080,
+            video_duration=5.0,
+            title_text="T",
+            title_config={"size": 56, "font_size": 28},
+        )
+        for line in result.split("\n"):
+            if line.startswith("Style: TitleStyle"):
+                parts = line.split(",")
+                assert parts[2] == "76"  # 56*1.35=75.6→76
                 break
 
     def test_title_bold(self):
@@ -529,3 +580,44 @@ class TestConstants:
         assert isinstance(TITLE_MARGIN_TOP, int)
         assert isinstance(TITLE_MARGIN_BOTTOM, int)
         assert isinstance(TITLE_MARGIN_SIDE, int)
+
+# ============================================================
+# _wrap_title_text 换行逻辑验证
+# ============================================================
+
+
+class TestWrapTitleTextFontSizeConsistency:
+    """验证换行计算使用原始 font_size，与 CSS 预览一致。"""
+
+    def test_font_size_113_uses_original_not_compensated(self):
+        """font_size=113 时，每行应容纳8个字（113px字宽），而非6个字（153px字宽）。"""
+        # 标题"永康拾掇脚阔头"共7个字
+        # 可用宽度 = 1080 - 60 - 60 = 960px
+        # 用 font_size=113：char_width=113，960/113 ≈ 8.5，每行8个字
+        # 7个字 < 8个字，应该在一行内
+        title = "永康拾掇脚阔头"
+        result = _wrap_title_text(title, video_width=1080, font_size=113, margin_l=60, margin_r=60)
+        # 不应该有换行
+        assert "\\N" not in result
+        assert result == title
+
+    def test_long_title_wraps_correctly(self):
+        """长标题应该按 font_size 字宽正确换行。"""
+        # 16个中文字，每行8个字，应该换行为2行
+        title = "永康拾掇脚阔头来一个笑一个哈哈哈"
+        result = _wrap_title_text(title, video_width=1080, font_size=113, margin_l=60, margin_r=60)
+        # 应该有一个换行
+        assert result.count("\\N") == 1
+        # 每行8个字
+        lines = result.split("\\N")
+        assert len(lines) == 2
+        assert len(lines[0]) == 8
+        assert len(lines[1]) == 8
+
+    def test_mixed_cjk_and_ascii(self):
+        """混合中英文时，英文按半角宽度计算。"""
+        # "测试test" = 2个中文(2*113=226) + 4个英文(4*113*0.55=248.6) = 474.6px
+        title = "测试test"
+        result = _wrap_title_text(title, video_width=1080, font_size=113, margin_l=60, margin_r=60)
+        # 总宽度474.6px < 960px，应该在一行内
+        assert "\\N" not in result

tests/unit/test_ass_subtitle_builder.py

@@ -79,12 +79,12 @@ class TestBuildAssStyle:
     def test_minimal_style(self):
         result = build_ass_style("TestStyle")
         assert result.startswith("Style: TestStyle,")
-        assert "思源黑体" in result
-        assert ",48," in result
+        assert "Noto Sans SC" in result
+        assert ",65," in result  # 48*1.35=64.8→65
 
     def test_custom_font_size(self):
         result = build_ass_style("Title", font_size=64)
-        assert ",64," in result
+        assert ",86," in result  # 64*1.35=86.4→86
 
     def test_bold_enabled(self):
         result = build_ass_style("BoldStyle", bold=True)
@@ -152,6 +152,26 @@ class TestBuildAssStyle:
         # Style: 行有 23 个字段（去掉 "Style: " 前缀后）
         assert len(parts) == 23
 
+    def test_font_name_mapping_siyuan(self):
+        """思源黑体 → Noto Sans SC"""
+        result = build_ass_style("Test", font_name="思源黑体")
+        assert "Noto Sans SC" in result
+
+    def test_font_name_mapping_apple(self):
+        """苹方 → Noto Sans SC"""
+        result = build_ass_style("Test", font_name="苹方")
+        assert "Noto Sans SC" in result
+
+    def test_font_name_mapping_msyh(self):
+        """微软雅黑 → Noto Sans SC"""
+        result = build_ass_style("Test", font_name="微软雅黑")
+        assert "Noto Sans SC" in result
+
+    def test_font_name_mapping_unknown_passthrough(self):
+        """未映射字体原样使用"""
+        result = build_ass_style("Test", font_name="CustomFont")
+        assert "CustomFont" in result
+
 
 # ── 文本转义 ──────────────────────────────────────────────────────────────────
 
@@ -190,6 +210,19 @@ class TestEscapeAssText:
     def test_chinese_text(self):
         assert escape_ass_text("你好世界") == "你好世界"
 
+    def test_slash_converted_to_newline(self):
+        """半角斜杠 / 应转为 ASS 硬换行。"""
+        assert escape_ass_text("第一行/第二行") == "第一行\\N第二行"
+
+    def test_fullwidth_slash_converted_to_newline(self):
+        """全角斜杠 ／ 应转为 ASS 硬换行。"""
+        assert escape_ass_text("第一行／第二行") == "第一行\\N第二行"
+
+    def test_mixed_slashes_and_newlines(self):
+        """斜杠和换行符都应转为硬换行。"""
+        result = escape_ass_text("A/B\nC／D")
+        assert result == "A\\NB\\NC\\ND"
+
 
 # ── 时间格式化 ────────────────────────────────────────────────────────────────
 
@@ -440,7 +473,7 @@ class TestBuildAssContent:
 
 class TestConstants:
     def test_title_margin_top(self):
-        assert TITLE_MARGIN_TOP == 60
+        assert TITLE_MARGIN_TOP == 120
 
     def test_title_margin_bottom(self):
         assert TITLE_MARGIN_BOTTOM == 60
@@ -448,6 +481,7 @@ class TestConstants:
     def test_title_margin_side(self):
         assert TITLE_MARGIN_SIDE == 40
 
+
 # ── 标题自动换行 ──────────────────────────────────────────────────────────────
 
 
@@ -491,6 +525,34 @@ class TestWrapTitleText:
         """字号为0时直接返回原文。"""
         assert _wrap_title_text("测试", 1080, 0) == "测试"
 
+    def test_preserves_explicit_newline(self):
+        """已有的 \\N 换行标记应保留，不被当普通字符算宽度。"""
+        text = "第一行\\N第二行"
+        result = _wrap_title_text(text, video_width=1080, font_size=48)
+        assert result == text
+
+    def test_explicit_newline_each_segment_wraps_independently(self):
+        """\\N 分段后，每段各自自动换行。"""
+        # 480px 宽，48px 字号，可用 400px，每段约8个中文字
+        text = "这是第一段很长很长很长的内容\\N这是第二段也很长很长的内容"
+        result = _wrap_title_text(text, video_width=480, font_size=48)
+        # 应该有多个 \N：用户手动的 + 自动换行的
+        assert "\\N" in result
+        segments = result.split("\\N")
+        # 至少3行（两段都需要换行）
+        assert len(segments) >= 3
+        # 验证包含两段的文字
+        joined = result.replace("\\N", "")
+        assert "第一段" in joined
+        assert "第二段" in joined
+
+    def test_multiple_explicit_newlines(self):
+        """多个 \\N 分段都应保留。"""
+        text = "A\\NB\\NC"
+        result = _wrap_title_text(text, video_width=1080, font_size=48)
+        assert result == text
+        assert result.count("\\N") == 2
+
     def test_build_ass_content_integration(self):
         """集成测试：build_ass_content 中的标题应该自动换行。"""
         long_title = "这是一段非常长的标题文字用于测试自动换行功能是否正常工作"
@@ -508,3 +570,35 @@ class TestWrapTitleText:
                 break
         else:
             pytest.fail("未找到 TitleStyle Dialogue 行")
+
+
+class TestFontsizeCompensation:
+    """ASS Fontsize 补偿系数（CSS 字号 → ASS em-square 字号）。"""
+
+    def test_default_48_compensated_to_65(self):
+        result = build_ass_style("S")
+        parts = result.split(",")
+        assert parts[2] == "65"  # round(48*1.35)=65
+
+    def test_89_compensated_to_120(self):
+        """实测对齐点：font_size=89 → ASS Fontsize=120。"""
+        result = build_ass_style("S", font_size=89)
+        parts = result.split(",")
+        assert parts[2] == "120"
+
+    def test_subtitle_also_compensated(self):
+        content = build_ass_content(
+            video_width=1080,
+            video_height=1920,
+            video_duration=5.0,
+            subtitle_text="字幕",
+            subtitle_config={"size": 24},
+        )
+        sub_line = [line for line in content.splitlines() if line.startswith("Style: SubtitleStyle")][0]
+        fields = [f.strip() for f in sub_line.split(",")]
+        assert fields[2] == "32"  # round(24*1.35)=32
+
+    def test_minimum_fontsize_at_least_one(self):
+        result = build_ass_style("S", font_size=0)
+        parts = result.split(",")
+        assert int(parts[2]) >= 1

tests/unit/test_editor_clips_random_start.py

@@ -0,0 +1,486 @@
+"""测试编辑器 from-assets 端点：按模板segment创建片段 + 事务性替换 + 随机起始.
+
+覆盖：
+- 片段数量 = segment 数量（required_clips_count 被忽略）
+- 素材不足时同一素材轮询切多个片段
+- 随机 start_time + used_segments 去重
+- 素材时长不足时 clip duration 缩短
+- 素材时长为 0 时抛 400
+- 使用 replace_all_clips_transactional 原子性替换
+- order 从 0 开始
+- start_time=None 时抛出 400
+- mark_clips_ready 在事务方法内部完成
+"""
+
+from __future__ import annotations
+
+import os
+import sys
+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 HTTPException
+
+TEST_PLAN_ID = "plan-draft-001"
+TEST_USER_ID = "user-001"
+
+# 默认测试用 segments：4 个片段，每个 3~5 秒
+DEFAULT_SEGMENTS = [(0, 3.0, 5.0), (1, 3.0, 5.0), (2, 3.0, 5.0), (3, 3.0, 5.0)]
+
+
+def _segments(count: int, dur_min: float = 3.0, dur_max: float = 5.0):
+    return [(i, dur_min, dur_max) for i in range(count)]
+
+
+def _patch_segments(segments=None):
+    return patch(
+        "app.api.routes.templates_editor.clips._get_template_segments",
+        return_value=segments if segments is not None else DEFAULT_SEGMENTS,
+    )
+
+
+def _make_auth_user():
+    auth = MagicMock()
+    auth.user.id = TEST_USER_ID
+    auth.user.email = "test@example.com"
+    auth.user.display_name = "测试用户"
+    auth.user_id = TEST_USER_ID
+    return auth
+
+
+def _make_mock_asset(asset_id, duration):
+    asset = MagicMock()
+    asset.id = asset_id
+    asset.duration = duration
+    return asset
+
+
+def _make_plan_svc(replace_return_count=None):
+    svc = MagicMock()
+    # replace_all_clips_transactional 返回创建的片段数量
+    if replace_return_count is not None:
+        svc.replace_all_clips_transactional = MagicMock(return_value=replace_return_count)
+    else:
+        svc.replace_all_clips_transactional = MagicMock(return_value=0)
+    return svc
+
+
+def _get_clips_data_from_call(mock_plan_svc):
+    """从 replace_all_clips_transactional 的调用中获取 clips_data。"""
+    assert mock_plan_svc.replace_all_clips_transactional.called, "replace_all_clips_transactional 未被调用"
+    call_args = mock_plan_svc.replace_all_clips_transactional.call_args
+    # call_args = ((plan_id, clips_data), kwargs)
+    if len(call_args.args) >= 2:
+        return call_args.args[1]
+    return call_args.kwargs.get("clips_data", [])
+
+
+class TestEditorClipsBySegments:
+    """测试按 segment 数量创建片段 + 素材轮询。"""
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_creates_clips_matching_segment_count(self, mock_storage):
+        """4 个 segment 即使只有2个素材也创建4个片段，required_clips_count 被忽略。"""
+        from app.api.routes.templates_editor.clips import (
+            create_clips_from_assets_editor,
+        )
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        mock_plan_svc = _make_plan_svc(replace_return_count=4)
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get = MagicMock(side_effect=lambda aid: _make_mock_asset(aid, {"a1": 30.0, "a2": 20.0}[aid]))
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1", "a2"], required_clips_count=2)
+
+        with _patch_segments(DEFAULT_SEGMENTS):
+            result = create_clips_from_assets_editor(
+                template_id="tpl-001",
+                body=body,
+                background_tasks=MagicMock(),
+                plan_id=TEST_PLAN_ID,
+                services=(MagicMock(), mock_plan_svc),
+                asset_repo=mock_asset_repo,
+                db=MagicMock(),
+                current_user=_make_auth_user(),
+            )
+
+        assert result.created_count == 4
+        clips_data = _get_clips_data_from_call(mock_plan_svc)
+        assert len(clips_data) == 4
+
+        # 验证轮询分配：a1, a2, a1, a2
+        assert clips_data[0]["asset_id"] == "a1"
+        assert clips_data[1]["asset_id"] == "a2"
+        assert clips_data[2]["asset_id"] == "a1"
+        assert clips_data[3]["asset_id"] == "a2"
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_orders_start_at_zero(self, mock_storage):
+        """片段 order 从 0 开始递增。"""
+        from app.api.routes.templates_editor.clips import (
+            create_clips_from_assets_editor,
+        )
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        mock_plan_svc = _make_plan_svc(replace_return_count=3)
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get = MagicMock(return_value=_make_mock_asset("a1", 30.0))
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1"], required_clips_count=3)
+
+        with _patch_segments(_segments(3)):
+            create_clips_from_assets_editor(
+                template_id="tpl-001",
+                body=body,
+                background_tasks=MagicMock(),
+                plan_id=TEST_PLAN_ID,
+                services=(MagicMock(), mock_plan_svc),
+                asset_repo=mock_asset_repo,
+                db=MagicMock(),
+                current_user=_make_auth_user(),
+            )
+
+        clips_data = _get_clips_data_from_call(mock_plan_svc)
+        assert clips_data[0]["order"] == 0
+        assert clips_data[1]["order"] == 1
+        assert clips_data[2]["order"] == 2
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_uses_transactional_replace(self, mock_storage):
+        """使用 replace_all_clips_transactional 而不是分别 delete + create。"""
+        from app.api.routes.templates_editor.clips import (
+            create_clips_from_assets_editor,
+        )
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        mock_plan_svc = _make_plan_svc(replace_return_count=2)
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get = MagicMock(return_value=_make_mock_asset("a1", 30.0))
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1"])
+
+        with _patch_segments(_segments(2)):
+            create_clips_from_assets_editor(
+                template_id="tpl-001",
+                body=body,
+                background_tasks=MagicMock(),
+                plan_id=TEST_PLAN_ID,
+                services=(MagicMock(), mock_plan_svc),
+                asset_repo=mock_asset_repo,
+                db=MagicMock(),
+                current_user=_make_auth_user(),
+            )
+
+        # 必须调用事务方法
+        mock_plan_svc.replace_all_clips_transactional.assert_called_once()
+        # 不应调用单独的 delete 或 create
+        assert not hasattr(mock_plan_svc, "create_clip") or not mock_plan_svc.create_clip.called
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_no_segments_raises_400(self, mock_storage):
+        """模板没有 segment 配置时返回 400。"""
+        from app.api.routes.templates_editor.clips import (
+            create_clips_from_assets_editor,
+        )
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        mock_plan_svc = _make_plan_svc()
+        mock_asset_repo = MagicMock()
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1"])
+
+        with _patch_segments([]):
+            with pytest.raises(HTTPException) as exc_info:
+                create_clips_from_assets_editor(
+                    template_id="tpl-001",
+                    body=body,
+                    background_tasks=MagicMock(),
+                    plan_id=TEST_PLAN_ID,
+                    services=(MagicMock(), mock_plan_svc),
+                    asset_repo=mock_asset_repo,
+                    db=MagicMock(),
+                    current_user=_make_auth_user(),
+                )
+
+        assert exc_info.value.status_code == 400
+        assert "片段配置" in exc_info.value.detail
+        # 不应调用替换方法
+        mock_plan_svc.replace_all_clips_transactional.assert_not_called()
+
+
+class TestEditorClipsDurationAndStartTime:
+    """测试素材时长获取、clip duration 缩短、start_time 传入。"""
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_clip_duration_shortened_for_short_assets(self, mock_storage):
+        """素材只有 3s 时 clip duration 缩短到不超过 3.0。"""
+        from app.api.routes.templates_editor.clips import (
+            create_clips_from_assets_editor,
+        )
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        mock_plan_svc = _make_plan_svc(replace_return_count=1)
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get = MagicMock(return_value=_make_mock_asset("short", 3.0))
+
+        body = ClipsFromAssetsRequest(asset_ids=["short"])
+
+        with _patch_segments(_segments(1, dur_min=5.0, dur_max=10.0)):
+            create_clips_from_assets_editor(
+                template_id="tpl-001",
+                body=body,
+                background_tasks=MagicMock(),
+                plan_id=TEST_PLAN_ID,
+                services=(MagicMock(), mock_plan_svc),
+                asset_repo=mock_asset_repo,
+                db=MagicMock(),
+                current_user=_make_auth_user(),
+            )
+
+        clips_data = _get_clips_data_from_call(mock_plan_svc)
+        assert clips_data[0]["duration"] <= 3.0
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_zero_duration_asset_raises_400(self, mock_storage):
+        """素材时长为 0 时应抛出 400，而不是创建无效片段。"""
+        from app.api.routes.templates_editor.clips import (
+            create_clips_from_assets_editor,
+        )
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        mock_plan_svc = _make_plan_svc()
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get = MagicMock(return_value=_make_mock_asset("bad", 0.0))
+
+        body = ClipsFromAssetsRequest(asset_ids=["bad"])
+
+        with _patch_segments(_segments(1)):
+            with pytest.raises(HTTPException) as exc_info:
+                create_clips_from_assets_editor(
+                    template_id="tpl-001",
+                    body=body,
+                    background_tasks=MagicMock(),
+                    plan_id=TEST_PLAN_ID,
+                    services=(MagicMock(), mock_plan_svc),
+                    asset_repo=mock_asset_repo,
+                    db=MagicMock(),
+                    current_user=_make_auth_user(),
+                )
+
+        assert exc_info.value.status_code == 400
+        assert "时长" in exc_info.value.detail
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_missing_duration_asset_raises_400(self, mock_storage):
+        """素材时长缺失（asset_repo.get 返回 None）时抛出 400。"""
+        from app.api.routes.templates_editor.clips import (
+            create_clips_from_assets_editor,
+        )
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        mock_plan_svc = _make_plan_svc()
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get = MagicMock(return_value=None)
+
+        body = ClipsFromAssetsRequest(asset_ids=["missing"])
+
+        with _patch_segments(_segments(1)):
+            with pytest.raises(HTTPException) as exc_info:
+                create_clips_from_assets_editor(
+                    template_id="tpl-001",
+                    body=body,
+                    background_tasks=MagicMock(),
+                    plan_id=TEST_PLAN_ID,
+                    services=(MagicMock(), mock_plan_svc),
+                    asset_repo=mock_asset_repo,
+                    db=MagicMock(),
+                    current_user=_make_auth_user(),
+                )
+
+        assert exc_info.value.status_code == 400
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_start_time_passed_to_create(self, mock_storage):
+        """_calc_random_start_time 返回值被传入 clips_data。"""
+        from app.api.routes.templates_editor.clips import (
+            create_clips_from_assets_editor,
+        )
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        mock_plan_svc = _make_plan_svc(replace_return_count=2)
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get = MagicMock(side_effect=lambda aid: _make_mock_asset(aid, 30.0))
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1", "a2"], required_clips_count=2)
+
+        with (
+            _patch_segments(_segments(2)),
+            patch(
+                "app.api.routes.templates_editor.clips._calc_random_start_time",
+                side_effect=[12.5, 18.0],
+            ) as mock_calc,
+        ):
+            create_clips_from_assets_editor(
+                template_id="tpl-001",
+                body=body,
+                background_tasks=MagicMock(),
+                plan_id=TEST_PLAN_ID,
+                services=(MagicMock(), mock_plan_svc),
+                asset_repo=mock_asset_repo,
+                db=MagicMock(),
+                current_user=_make_auth_user(),
+            )
+
+        assert mock_calc.call_count == 2
+        clips_data = _get_clips_data_from_call(mock_plan_svc)
+        assert clips_data[0]["start_time"] == 12.5
+        assert clips_data[1]["start_time"] == 18.0
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_asset_durations_deduped(self, mock_storage):
+        """asset_ids 有重复时只查询一次素材时长。"""
+        from app.api.routes.templates_editor.clips import (
+            create_clips_from_assets_editor,
+        )
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        mock_plan_svc = _make_plan_svc(replace_return_count=3)
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get = MagicMock(return_value=_make_mock_asset("a1", 30.0))
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1", "a1", "a1"])
+
+        with _patch_segments(_segments(3)):
+            create_clips_from_assets_editor(
+                template_id="tpl-001",
+                body=body,
+                background_tasks=MagicMock(),
+                plan_id=TEST_PLAN_ID,
+                services=(MagicMock(), mock_plan_svc),
+                asset_repo=mock_asset_repo,
+                db=MagicMock(),
+                current_user=_make_auth_user(),
+            )
+
+        # 去重后只调用 1 次获取素材时长
+        assert mock_asset_repo.get.call_count == 1
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_used_segments_maintained_across_clips(self, mock_storage):
+        """同一素材切多个片段时，used_segments 应被维护。"""
+        from app.api.routes.templates_editor.clips import (
+            create_clips_from_assets_editor,
+        )
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        mock_plan_svc = _make_plan_svc(replace_return_count=3)
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get = MagicMock(return_value=_make_mock_asset("a1", 30.0))
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1"], required_clips_count=3)
+
+        captured_used_segments = []
+
+        def fake_calc(asset_id, clip_duration, asset_durations, used_segments):
+            captured_used_segments.append({aid: list(segs) for aid, segs in (used_segments or {}).items()})
+            return (len(captured_used_segments) - 1) * 5.0
+
+        with (
+            _patch_segments(_segments(3)),
+            patch(
+                "app.api.routes.templates_editor.clips._calc_random_start_time",
+                side_effect=fake_calc,
+            ),
+        ):
+            create_clips_from_assets_editor(
+                template_id="tpl-001",
+                body=body,
+                background_tasks=MagicMock(),
+                plan_id=TEST_PLAN_ID,
+                services=(MagicMock(), mock_plan_svc),
+                asset_repo=mock_asset_repo,
+                db=MagicMock(),
+                current_user=_make_auth_user(),
+            )
+
+        # 第一次没有已使用时间段
+        assert captured_used_segments[0] == {}
+        # 第二次有第一次的记录
+        assert len(captured_used_segments[1]["a1"]) == 1
+        # 第三次有前两次的记录
+        assert len(captured_used_segments[2]["a1"]) == 2
+
+
+class TestEditorClipsErrorHandling:
+    """测试异常处理。"""
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_none_start_time_raises_400(self, mock_storage):
+        """_calc_random_start_time 返回 None 时应抛出 HTTPException 400。"""
+        from app.api.routes.templates_editor.clips import (
+            create_clips_from_assets_editor,
+        )
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        mock_plan_svc = _make_plan_svc()
+        mock_asset_repo = MagicMock()
+        # 素材有 duration 但 random 返回 None
+        mock_asset_repo.get = MagicMock(return_value=_make_mock_asset("a1", 30.0))
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1"])
+
+        with (
+            _patch_segments(_segments(1)),
+            patch(
+                "app.api.routes.templates_editor.clips._calc_random_start_time",
+                return_value=None,
+            ),
+        ):
+            with pytest.raises(HTTPException) as exc_info:
+                create_clips_from_assets_editor(
+                    template_id="tpl-001",
+                    body=body,
+                    background_tasks=MagicMock(),
+                    plan_id=TEST_PLAN_ID,
+                    services=(MagicMock(), mock_plan_svc),
+                    asset_repo=mock_asset_repo,
+                    db=MagicMock(),
+                    current_user=_make_auth_user(),
+                )
+
+        assert exc_info.value.status_code == 400
+        assert "时长" in exc_info.value.detail
+
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_transactional_replace_exception_propagates(self, mock_storage):
+        """replace_all_clips_transactional 抛异常时应向上传播（事务已回滚）。"""
+        from app.api.routes.templates_editor.clips import (
+            create_clips_from_assets_editor,
+        )
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        mock_plan_svc = _make_plan_svc()
+        mock_plan_svc.replace_all_clips_transactional = MagicMock(side_effect=RuntimeError("DB connection lost"))
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get = MagicMock(return_value=_make_mock_asset("a1", 30.0))
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1"])
+
+        with _patch_segments(_segments(1)):
+            with pytest.raises(RuntimeError, match="DB connection lost"):
+                create_clips_from_assets_editor(
+                    template_id="tpl-001",
+                    body=body,
+                    background_tasks=MagicMock(),
+                    plan_id=TEST_PLAN_ID,
+                    services=(MagicMock(), mock_plan_svc),
+                    asset_repo=mock_asset_repo,
+                    db=MagicMock(),
+                    current_user=_make_auth_user(),
+                )

tests/unit/test_generated_video_creation_logic.py

@@ -321,6 +321,27 @@ class TestThumbnailInDedupHelpers:
         video_processing.dedup = mock_dedup
         video_processing.thumbnail_generator = mock_thumb
 
+    @classmethod
+    def teardown_class(cls):
+        """恢复 setup_class 中替换的模块，避免污染后续测试。"""
+        import sys
+
+        import video_processing
+
+        # 从 sys.modules 移除 mock 模块
+        for mod_name in ("video_processing.thumbnail_generator", "video_processing.dedup"):
+            sys.modules.pop(mod_name, None)
+
+        # 重新导入真实模块以恢复 sys.modules
+        try:
+            import video_processing.thumbnail_generator  # noqa: F401
+        except Exception:
+            pass
+        try:
+            import video_processing.dedup  # noqa: F401
+        except Exception:
+            pass
+
     def test_pre_generated_thumbnail_url_is_reused(self):
         """传入 thumbnail_url 时直接复用，统一封面管道不再自动生成缩略图。"""
         from video_processing.dedup_helpers import create_video_record_and_dedup

tests/unit/test_generation_preview.py

@@ -1206,7 +1206,10 @@ class TestPreviewRouteAutoInfersVideoRatio:
         # Verify the resolution passed to CreateGenerationTaskCommand
         call_args = MockUC.return_value.execute.call_args
         cmd = call_args[0][0]
-        assert cmd.resolution == "", f"Expected empty resolution, got {cmd.resolution}"
+        # video_ratio inferred from pip → 9:16 → resolution=1080x1920
+        assert cmd.resolution == "1080x1920", f"Expected 1080x1920, got {cmd.resolution}"
+        assert cmd.output_width == 1080, f"Expected output_width=1080, got {cmd.output_width}"
+        assert cmd.output_height == 1920, f"Expected output_height=1920, got {cmd.output_height}"
 
 
 # ═══════════════════════════════════════════════════════════════════════════════

tests/unit/test_mediakit_smart_clips.py

@@ -0,0 +1,657 @@
+"""测试 MediaKit 智能选片 + from-assets 按模板 segment 创建片段。
+
+覆盖：
+- _recommended_time_conflicts 冲突检测
+- _get_mediakit_recommendations 解析与降级
+- _get_template_segments 查询模板片段配置
+- from-assets 端点：按模板 segment 数量和时长创建片段
+- from-assets 端点：事务性原子替换
+- from-assets 端点：素材轮询分配
+- from-assets 端点：MediaKit 推荐时间首片段使用
+- from-assets 端点：同一素材多片段时后续用随机
+- from-assets 端点：无 segment 配置时报错
+- from-assets 端点：素材时长为0时报400
+"""
+
+from __future__ import annotations
+
+import os
+import sys
+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
+
+# ── _recommended_time_conflicts 单元测试 ─────────────────────────────────────
+
+
+class TestRecommendedTimeConflicts:
+    """测试推荐时间与已使用时间段的冲突检测。"""
+
+    def test_no_conflict_when_empty(self):
+        from app.api.routes.templates_editor.clips import _recommended_time_conflicts
+
+        assert _recommended_time_conflicts(5.0, 5.0, []) is False
+
+    def test_no_conflict_when_before(self):
+        from app.api.routes.templates_editor.clips import _recommended_time_conflicts
+
+        assert _recommended_time_conflicts(5.0, 5.0, [(15.0, 20.0)]) is False
+
+    def test_no_conflict_when_after(self):
+        from app.api.routes.templates_editor.clips import _recommended_time_conflicts
+
+        assert _recommended_time_conflicts(20.0, 5.0, [(0.0, 10.0)]) is False
+
+    def test_conflict_overlap_start(self):
+        from app.api.routes.templates_editor.clips import _recommended_time_conflicts
+
+        # 推荐 [5, 10]，已用 [0, 7]
+        assert _recommended_time_conflicts(5.0, 5.0, [(0.0, 7.0)]) is True
+
+    def test_conflict_overlap_end(self):
+        from app.api.routes.templates_editor.clips import _recommended_time_conflicts
+
+        # 推荐 [5, 10]，已用 [8, 15]
+        assert _recommended_time_conflicts(5.0, 5.0, [(8.0, 15.0)]) is True
+
+    def test_conflict_contained(self):
+        from app.api.routes.templates_editor.clips import _recommended_time_conflicts
+
+        # 推荐 [5, 10]，已用 [0, 20]
+        assert _recommended_time_conflicts(5.0, 5.0, [(0.0, 20.0)]) is True
+
+    def test_conflict_exact_boundary_no_overlap(self):
+        from app.api.routes.templates_editor.clips import _recommended_time_conflicts
+
+        # 推荐 [10, 15]，已用 [0, 10] — 边界相接不算冲突
+        assert _recommended_time_conflicts(10.0, 5.0, [(0.0, 10.0)]) is False
+
+    def test_conflict_multiple_used(self):
+        from app.api.routes.templates_editor.clips import _recommended_time_conflicts
+
+        used = [(0.0, 5.0), (10.0, 15.0), (20.0, 25.0)]
+        # 推荐 [6, 11] 与 [10, 15] 冲突
+        assert _recommended_time_conflicts(6.0, 5.0, used) is True
+        # 推荐 [15, 20] 不冲突
+        assert _recommended_time_conflicts(15.0, 5.0, used) is False
+
+
+# ── _get_mediakit_recommendations 单元测试 ──────────────────────────────────
+
+
+class TestGetMediakitRecommendations:
+    """测试 MediaKit 推荐结果解析和降级。"""
+
+    @patch("app.api.routes.templates_editor.clips.get_mediakit_client")
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_returns_parsed_recommendations(self, mock_storage, mock_client_fn):
+        from app.api.routes.templates_editor.clips import _get_mediakit_recommendations
+
+        mock_client = MagicMock()
+        mock_client.is_available = True
+        mock_client.analyze_videos.return_value = [
+            '[{"asset_id": "a1", "recommended_start_time": 12.5, "reason": "动作场景"}]'
+        ]
+        mock_client_fn.return_value = mock_client
+
+        mock_storage_svc = MagicMock()
+        mock_storage_svc.get_download_url.return_value = "https://example.com/v.mp4"
+        mock_storage.return_value = mock_storage_svc
+
+        asset = MagicMock()
+        asset.storage_key = "v.mp4"
+        asset.mime_type = "video/mp4"
+
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get.return_value = asset
+
+        result = _get_mediakit_recommendations(["a1"], mock_asset_repo)
+        assert result == {"a1": 12.5}
+
+    @patch("app.api.routes.templates_editor.clips.get_mediakit_client")
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_client_not_available_returns_empty(self, mock_storage, mock_client_fn):
+        from app.api.routes.templates_editor.clips import _get_mediakit_recommendations
+
+        mock_client = MagicMock()
+        mock_client.is_available = False
+        mock_client_fn.return_value = mock_client
+
+        result = _get_mediakit_recommendations(["a1"], MagicMock())
+        assert result == {}
+
+    @patch("app.api.routes.templates_editor.clips.get_mediakit_client")
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_empty_contents_returns_empty(self, mock_storage, mock_client_fn):
+        from app.api.routes.templates_editor.clips import _get_mediakit_recommendations
+
+        mock_client = MagicMock()
+        mock_client.is_available = True
+        mock_client.analyze_videos.return_value = []
+        mock_client_fn.return_value = mock_client
+
+        mock_storage_svc = MagicMock()
+        mock_storage_svc.get_download_url.return_value = "https://example.com/v.mp4"
+        mock_storage.return_value = mock_storage_svc
+
+        asset = MagicMock()
+        asset.storage_key = "v.mp4"
+        asset.mime_type = "video/mp4"
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get.return_value = asset
+
+        result = _get_mediakit_recommendations(["a1"], mock_asset_repo)
+        assert result == {}
+
+    @patch("app.api.routes.templates_editor.clips.get_mediakit_client")
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_unparseable_response_returns_empty(self, mock_storage, mock_client_fn):
+        from app.api.routes.templates_editor.clips import _get_mediakit_recommendations
+
+        mock_client = MagicMock()
+        mock_client.is_available = True
+        mock_client.analyze_videos.return_value = ["这是一段自然语言描述，没有JSON"]
+        mock_client_fn.return_value = mock_client
+
+        mock_storage_svc = MagicMock()
+        mock_storage_svc.get_download_url.return_value = "https://example.com/v.mp4"
+        mock_storage.return_value = mock_storage_svc
+
+        asset = MagicMock()
+        asset.storage_key = "v.mp4"
+        asset.mime_type = "video/mp4"
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get.return_value = asset
+
+        result = _get_mediakit_recommendations(["a1"], mock_asset_repo)
+        assert result == {}
+
+    @patch("app.api.routes.templates_editor.clips.get_mediakit_client")
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_storage_failure_returns_empty(self, mock_storage, mock_client_fn):
+        from app.api.routes.templates_editor.clips import _get_mediakit_recommendations
+
+        mock_client = MagicMock()
+        mock_client.is_available = True
+        mock_client_fn.return_value = mock_client
+        mock_storage.side_effect = RuntimeError("storage unavailable")
+
+        result = _get_mediakit_recommendations(["a1"], MagicMock())
+        assert result == {}
+
+    @patch("app.api.routes.templates_editor.clips.get_mediakit_client")
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_exception_returns_empty(self, mock_storage, mock_client_fn):
+        from app.api.routes.templates_editor.clips import _get_mediakit_recommendations
+
+        mock_client_fn.side_effect = RuntimeError("unexpected error")
+
+        result = _get_mediakit_recommendations(["a1"], MagicMock())
+        assert result == {}
+
+    @patch("app.api.routes.templates_editor.clips.get_mediakit_client")
+    @patch("app.api.routes.templates_editor.clips.get_storage_service")
+    def test_skips_non_video_assets(self, mock_storage, mock_client_fn):
+        from app.api.routes.templates_editor.clips import _get_mediakit_recommendations
+
+        mock_client = MagicMock()
+        mock_client.is_available = True
+        mock_client_fn.return_value = mock_client
+
+        mock_asset_repo = MagicMock()
+        mock_asset = MagicMock()
+        mock_asset.storage_key = "images/test.jpg"
+        mock_asset.mime_type = "image/jpeg"
+        mock_asset_repo.get.return_value = mock_asset
+
+        result = _get_mediakit_recommendations(["a1"], mock_asset_repo)
+        assert result == {}
+        mock_client.analyze_videos.assert_not_called()
+
+
+# ── _get_template_segments 单元测试 ─────────────────────────────────────────
+
+
+class TestGetTemplateSegments:
+    """测试模板片段配置查询。"""
+
+    def test_returns_segments_from_new_template_system(self):
+        """新模板系统（clip_configs）有数据时优先使用。"""
+        from app.api.routes.templates_editor.clips import _get_template_segments
+
+        mock_tpl_svc = MagicMock()
+        cc1 = MagicMock()
+        cc1.order = 0
+        cc1.min_duration = 3.0
+        cc1.max_duration = 5.0
+        cc2 = MagicMock()
+        cc2.order = 1
+        cc2.min_duration = 4.0
+        cc2.max_duration = 8.0
+        mock_tpl_svc.list_clip_configs.return_value = [cc2, cc1]  # 乱序返回
+
+        result = _get_template_segments("tmpl-1", mock_tpl_svc, MagicMock())
+        assert len(result) == 2
+        assert result[0] == (0, 3.0, 5.0)
+        assert result[1] == (1, 4.0, 8.0)
+
+    def test_falls_back_to_old_template_segments(self):
+        """新模板系统无数据时回退到旧系统。"""
+        from app.api.routes.templates_editor.clips import _get_template_segments
+
+        mock_tpl_svc = MagicMock()
+        mock_tpl_svc.list_clip_configs.return_value = []
+
+        with patch("app.api.routes.templates_editor.clips.SQLAlchemyTemplateRepository") as MockRepo:
+            mock_repo = MagicMock()
+            seg1 = MagicMock()
+            seg1.segment_order = 0
+            seg1.duration_min = 2.0
+            seg1.duration_max = 4.0
+            mock_repo.list_segments.return_value = [seg1]
+            MockRepo.return_value = mock_repo
+
+            result = _get_template_segments("tmpl-1", mock_tpl_svc, MagicMock())
+            assert len(result) == 1
+            assert result[0] == (0, 2.0, 4.0)
+
+    def test_returns_empty_when_no_segments(self):
+        """两套系统都没有片段配置时返回空列表。"""
+        from app.api.routes.templates_editor.clips import _get_template_segments
+
+        mock_tpl_svc = MagicMock()
+        mock_tpl_svc.list_clip_configs.return_value = []
+
+        with patch("app.api.routes.templates_editor.clips.SQLAlchemyTemplateRepository") as MockRepo:
+            mock_repo = MagicMock()
+            mock_repo.list_segments.return_value = []
+            MockRepo.return_value = mock_repo
+
+            result = _get_template_segments("tmpl-1", mock_tpl_svc, MagicMock())
+            assert result == []
+
+    def test_new_system_exception_falls_back(self):
+        """新模板系统异常时回退到旧系统。"""
+        from app.api.routes.templates_editor.clips import _get_template_segments
+
+        mock_tpl_svc = MagicMock()
+        mock_tpl_svc.list_clip_configs.side_effect = RuntimeError("db error")
+
+        with patch("app.api.routes.templates_editor.clips.SQLAlchemyTemplateRepository") as MockRepo:
+            mock_repo = MagicMock()
+            seg = MagicMock()
+            seg.segment_order = 0
+            seg.duration_min = 1.0
+            seg.duration_max = 3.0
+            mock_repo.list_segments.return_value = [seg]
+            MockRepo.return_value = mock_repo
+
+            result = _get_template_segments("tmpl-1", mock_tpl_svc, MagicMock())
+            assert len(result) == 1
+
+
+# ── from-assets 端点集成测试 ────────────────────────────────────────────────
+
+
+def _make_auth_user():
+    auth = MagicMock()
+    auth.user.id = "user-001"
+    auth.user.email = "test@example.com"
+    auth.user.display_name = "test"
+    auth.user_id = "user-001"
+    return auth
+
+
+def _make_clip_config(order, min_dur, max_dur):
+    cc = MagicMock()
+    cc.order = order
+    cc.min_duration = min_dur
+    cc.max_duration = max_dur
+    return cc
+
+
+def _make_plan_svc(replace_return_count=None):
+    svc = MagicMock()
+    svc.get_plan_or_raise = MagicMock()
+    if replace_return_count is not None:
+        svc.replace_all_clips_transactional = MagicMock(return_value=replace_return_count)
+    else:
+        svc.replace_all_clips_transactional = MagicMock(return_value=0)
+    return svc
+
+
+def _make_tpl_svc_with_segments(segments):
+    """segments: list of (order, min_dur, max_dur)"""
+    svc = MagicMock()
+    clip_configs = [_make_clip_config(o, mn, mx) for o, mn, mx in segments]
+    svc.list_clip_configs.return_value = clip_configs
+    return svc
+
+
+def _make_rich_asset(asset_id, duration, storage_key="v.mp4", mime="video/mp4"):
+    asset = MagicMock()
+    asset.id = asset_id
+    asset.duration = duration
+    asset.storage_key = storage_key
+    asset.mime_type = mime
+    return asset
+
+
+def _get_clips_data(mock_plan_svc):
+    """从 replace_all_clips_transactional 调用中提取 clips_data。"""
+    call_args = mock_plan_svc.replace_all_clips_transactional.call_args
+    if len(call_args.args) >= 2:
+        return call_args.args[1]
+    return call_args.kwargs.get("clips_data", [])
+
+
+class TestFromAssetsByTemplateSegments:
+    """测试 from-assets 按模板 segment 创建片段（V2 事务性替换）。"""
+
+    def test_creates_clips_matching_segment_count(self):
+        """片段数量 = segment 数量，忽略 required_clips_count。"""
+        from app.api.routes.templates_editor.clips import create_clips_from_assets_editor
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        segments = [(0, 3.0, 5.0), (1, 4.0, 8.0), (2, 2.0, 6.0), (3, 5.0, 10.0)]
+        mock_tpl_svc = _make_tpl_svc_with_segments(segments)
+        mock_plan_svc = _make_plan_svc(replace_return_count=4)
+
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get.return_value = _make_rich_asset("a1", 60.0)
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1"], required_clips_count=2)
+        result = create_clips_from_assets_editor(
+            template_id="tmpl-1",
+            body=body,
+            background_tasks=MagicMock(),
+            plan_id="plan-1",
+            services=(mock_tpl_svc, mock_plan_svc),
+            asset_repo=mock_asset_repo,
+            db=MagicMock(),
+            current_user=_make_auth_user(),
+        )
+
+        assert result.created_count == 4
+        clips_data = _get_clips_data(mock_plan_svc)
+        assert len(clips_data) == 4
+
+    def test_uses_transactional_replace(self):
+        """使用 replace_all_clips_transactional 而不是分别 delete + create。"""
+        from app.api.routes.templates_editor.clips import create_clips_from_assets_editor
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        segments = [(0, 3.0, 5.0), (1, 4.0, 8.0)]
+        mock_tpl_svc = _make_tpl_svc_with_segments(segments)
+        mock_plan_svc = _make_plan_svc(replace_return_count=2)
+
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get.return_value = _make_rich_asset("a1", 60.0)
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1"])
+        create_clips_from_assets_editor(
+            template_id="tmpl-1",
+            body=body,
+            background_tasks=MagicMock(),
+            plan_id="plan-1",
+            services=(mock_tpl_svc, mock_plan_svc),
+            asset_repo=mock_asset_repo,
+            db=MagicMock(),
+            current_user=_make_auth_user(),
+        )
+
+        mock_plan_svc.replace_all_clips_transactional.assert_called_once()
+        assert not mock_plan_svc.delete_all_clips.called
+        assert not mock_plan_svc.create_clip.called
+
+    def test_duration_within_segment_range(self):
+        """每个片段时长在 segment 的 min~max 范围内。"""
+        from app.api.routes.templates_editor.clips import create_clips_from_assets_editor
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        segments = [(0, 3.0, 5.0), (1, 4.0, 8.0)]
+        mock_tpl_svc = _make_tpl_svc_with_segments(segments)
+        mock_plan_svc = _make_plan_svc(replace_return_count=2)
+
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get.return_value = _make_rich_asset("a1", 60.0)
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1"])
+        create_clips_from_assets_editor(
+            template_id="tmpl-1",
+            body=body,
+            background_tasks=MagicMock(),
+            plan_id="plan-1",
+            services=(mock_tpl_svc, mock_plan_svc),
+            asset_repo=mock_asset_repo,
+            db=MagicMock(),
+            current_user=_make_auth_user(),
+        )
+
+        clips_data = _get_clips_data(mock_plan_svc)
+        assert 3.0 <= clips_data[0]["duration"] <= 5.0
+        assert 4.0 <= clips_data[1]["duration"] <= 8.0
+
+    def test_assets_round_robin_assignment(self):
+        """素材按片段顺序轮询分配。"""
+        from app.api.routes.templates_editor.clips import create_clips_from_assets_editor
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        segments = [(0, 3.0, 5.0), (1, 3.0, 5.0), (2, 3.0, 5.0), (3, 3.0, 5.0)]
+        mock_tpl_svc = _make_tpl_svc_with_segments(segments)
+        mock_plan_svc = _make_plan_svc(replace_return_count=4)
+
+        mock_asset_repo = MagicMock()
+
+        def get_asset(aid):
+            return _make_rich_asset(aid, 60.0)
+
+        mock_asset_repo.get.side_effect = get_asset
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1", "a2"])
+        create_clips_from_assets_editor(
+            template_id="tmpl-1",
+            body=body,
+            background_tasks=MagicMock(),
+            plan_id="plan-1",
+            services=(mock_tpl_svc, mock_plan_svc),
+            asset_repo=mock_asset_repo,
+            db=MagicMock(),
+            current_user=_make_auth_user(),
+        )
+
+        clips_data = _get_clips_data(mock_plan_svc)
+        asset_ids = [c["asset_id"] for c in clips_data]
+        assert asset_ids == ["a1", "a2", "a1", "a2"]
+
+    def test_orders_start_from_zero(self):
+        """片段 order 从 0 开始递增。"""
+        from app.api.routes.templates_editor.clips import create_clips_from_assets_editor
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        segments = [(0, 3.0, 5.0), (1, 3.0, 5.0), (2, 3.0, 5.0)]
+        mock_tpl_svc = _make_tpl_svc_with_segments(segments)
+        mock_plan_svc = _make_plan_svc(replace_return_count=3)
+
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get.return_value = _make_rich_asset("a1", 60.0)
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1"])
+        create_clips_from_assets_editor(
+            template_id="tmpl-1",
+            body=body,
+            background_tasks=MagicMock(),
+            plan_id="plan-1",
+            services=(mock_tpl_svc, mock_plan_svc),
+            asset_repo=mock_asset_repo,
+            db=MagicMock(),
+            current_user=_make_auth_user(),
+        )
+
+        clips_data = _get_clips_data(mock_plan_svc)
+        orders = [c["order"] for c in clips_data]
+        assert orders == [0, 1, 2]
+
+    def test_no_segments_raises_400(self):
+        """模板没有 segment 配置时返回 400。"""
+        from app.api.routes.templates_editor.clips import create_clips_from_assets_editor
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+        from fastapi import HTTPException
+
+        mock_tpl_svc = MagicMock()
+        mock_tpl_svc.list_clip_configs.return_value = []
+        mock_plan_svc = _make_plan_svc()
+
+        with patch("app.api.routes.templates_editor.clips.SQLAlchemyTemplateRepository") as MockRepo:
+            mock_repo = MagicMock()
+            mock_repo.list_segments.return_value = []
+            MockRepo.return_value = mock_repo
+
+            body = ClipsFromAssetsRequest(asset_ids=["a1"])
+            with pytest.raises(HTTPException) as exc_info:
+                create_clips_from_assets_editor(
+                    template_id="tmpl-1",
+                    body=body,
+                    background_tasks=MagicMock(),
+                    plan_id="plan-1",
+                    services=(mock_tpl_svc, mock_plan_svc),
+                    asset_repo=MagicMock(),
+                    db=MagicMock(),
+                    current_user=_make_auth_user(),
+                )
+            assert exc_info.value.status_code == 400
+
+        mock_plan_svc.replace_all_clips_transactional.assert_not_called()
+
+    def test_duration_capped_by_asset_duration(self):
+        """素材时长不足时 clip duration 被缩短。"""
+        from app.api.routes.templates_editor.clips import create_clips_from_assets_editor
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+
+        segments = [(0, 10.0, 20.0)]
+        mock_tpl_svc = _make_tpl_svc_with_segments(segments)
+        mock_plan_svc = _make_plan_svc(replace_return_count=1)
+
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get.return_value = _make_rich_asset("a1", 5.0)
+
+        body = ClipsFromAssetsRequest(asset_ids=["a1"])
+        create_clips_from_assets_editor(
+            template_id="tmpl-1",
+            body=body,
+            background_tasks=MagicMock(),
+            plan_id="plan-1",
+            services=(mock_tpl_svc, mock_plan_svc),
+            asset_repo=mock_asset_repo,
+            db=MagicMock(),
+            current_user=_make_auth_user(),
+        )
+
+        clips_data = _get_clips_data(mock_plan_svc)
+        assert clips_data[0]["duration"] <= 5.0
+
+    def test_zero_duration_asset_raises_400(self):
+        """素材时长为 0 时抛出 400。"""
+        from app.api.routes.templates_editor.clips import create_clips_from_assets_editor
+        from app.api.routes.templates_editor.schemas import ClipsFromAssetsRequest
+        from fastapi import HTTPException
+
+        segments = [(0, 3.0, 5.0)]
+        mock_tpl_svc = _make_tpl_svc_with_segments(segments)
+        mock_plan_svc = _make_plan_svc()
+
+        mock_asset_repo = MagicMock()
+        mock_asset_repo.get.return_value = _make_rich_asset("bad", 0.0)
+
+        body = ClipsFromAssetsRequest(asset_ids=["bad"])
+        with pytest.raises(HTTPException) as exc_info:
+            create_clips_from_assets_editor(
+                template_id="tmpl-1",
+                body=body,
+                background_tasks=MagicMock(),
+                plan_id="plan-1",
+                services=(mock_tpl_svc, mock_plan_svc),
+                asset_repo=mock_asset_repo,
+                db=MagicMock(),
+                current_user=_make_auth_user(),
+            )
+
+        assert exc_info.value.status_code == 400
+        mock_plan_svc.replace_all_clips_transactional.assert_not_called()
+
+    def test_empty_asset_ids_raises_400(self):
+        """asset_ids 为空列表时返回 400（defense-in-depth，schema 层也有 min_length=1）。"""
+        from app.api.routes.templates_editor.clips import create_clips_from_assets_editor
+        from fastapi import HTTPException
+
+        segments = [(0, 3.0, 5.0)]
+        mock_tpl_svc = _make_tpl_svc_with_segments(segments)
+        mock_plan_svc = _make_plan_svc()
+
+        # 用 MagicMock 模拟 body，绕过 Pydantic schema 的 min_length 校验
+        mock_body = MagicMock()
+        mock_body.asset_ids = []
+        mock_body.required_clips_count = None
+
+        with pytest.raises(HTTPException) as exc_info:
+            create_clips_from_assets_editor(
+                template_id="tmpl-1",
+                body=mock_body,
+                background_tasks=MagicMock(),
+                plan_id="plan-1",
+                services=(mock_tpl_svc, mock_plan_svc),
+                asset_repo=MagicMock(),
+                db=MagicMock(),
+                current_user=_make_auth_user(),
+            )
+        assert exc_info.value.status_code == 400
+        assert "素材" in exc_info.value.detail
+        mock_plan_svc.replace_all_clips_transactional.assert_not_called()
+
+
+# ── _safe_segment_duration 单元测试 ─────────────────────────────────────────
+
+
+class TestSafeSegmentDuration:
+    """测试片段时长安全转换。"""
+
+    def test_normal_float(self):
+        from app.api.routes.templates_editor.clips import _safe_segment_duration
+
+        assert _safe_segment_duration(3.5, 1.0) == 3.5
+
+    def test_none_returns_default(self):
+        from app.api.routes.templates_editor.clips import _safe_segment_duration
+
+        assert _safe_segment_duration(None, 5.0) == 5.0
+
+    def test_string_number(self):
+        from app.api.routes.templates_editor.clips import _safe_segment_duration
+
+        assert _safe_segment_duration("4.2", 1.0) == 4.2
+
+    def test_invalid_string_returns_default(self):
+        from app.api.routes.templates_editor.clips import _safe_segment_duration
+
+        assert _safe_segment_duration("abc", 5.0) == 5.0
+
+    def test_negative_returns_default(self):
+        from app.api.routes.templates_editor.clips import _safe_segment_duration
+
+        assert _safe_segment_duration(-1.0, 5.0) == 5.0
+
+    def test_zero_returns_default(self):
+        from app.api.routes.templates_editor.clips import _safe_segment_duration
+
+        assert _safe_segment_duration(0, 5.0) == 5.0
+
+    def test_integer_value(self):
+        from app.api.routes.templates_editor.clips import _safe_segment_duration
+
+        assert _safe_segment_duration(10, 1.0) == 10.0

tests/unit/test_plan_generator.py

@@ -886,3 +886,120 @@ class TestTemplateConfigPropagation:
         assert clip_cfg.get("speed_ratio") == 1.2
         assert clip_cfg.get("name") == "开场"
         assert clip_cfg.get("custom_field") == "value"
+
+
+class TestAssetDurationsAlwaysFetched:
+    """验证 asset_durations 不再受 random_preview 条件限制。
+
+    修复前：asset_durations 仅在 random_preview=True 时传入 distribute_assets
+    修复后：只要 _asset_repo 存在，就始终获取 asset_durations
+    """
+
+    def _make_service_with_asset_repo(self):
+        """创建带 mock asset_repo 的 PlanGeneratorService"""
+        from apps.api.app.services.plan_generator_service import PlanGeneratorService
+
+        plan_repo = StubEditPlanRepository()
+        clip_repo = StubEditPlanClipRepository()
+
+        # mock asset_repo: 返回带 duration 的素材
+        asset_repo = MagicMock()
+
+        def fake_get(asset_id):
+            mock_asset = MagicMock()
+            mock_asset.duration = 30.0  # 每个素材 30 秒
+            return mock_asset
+
+        asset_repo.get = MagicMock(side_effect=fake_get)
+
+        with (
+            patch(
+                "apps.api.app.services.plan_generator_service.SQLAlchemyEditPlanRepository",
+                return_value=plan_repo,
+            ),
+            patch(
+                "apps.api.app.services.plan_generator_service.SQLAlchemyEditPlanClipRepository",
+                return_value=clip_repo,
+            ),
+        ):
+            db = MagicMock()
+            svc = PlanGeneratorService(db, asset_repo=asset_repo)
+            svc._plan_repo = plan_repo
+            svc._clip_repo = clip_repo
+
+        return svc, asset_repo
+
+    def test_asset_durations_fetched_without_random_preview(self):
+        """random_preview=False 时也应获取 asset_durations"""
+        svc, asset_repo = self._make_service_with_asset_repo()
+
+        template = _make_template("one_take")
+        clip_configs = _make_clip_configs(
+            template_id=template.id,
+            specs=[{"clip_type": ClipType.MAIN, "order": 0, "min_duration": 3.0, "max_duration": 5.0}],
+        )
+
+        # patch distribute_assets 以捕获传入的参数
+        with patch("apps.api.app.services.plan_generator_service.distribute_assets") as mock_distribute:
+            svc.generate_from_template(
+                template=template,
+                clip_configs=clip_configs,
+                asset_ids=["a1", "a2"],
+                random_preview=False,  # 关键：非随机预览模式
+            )
+
+            # 验证 asset_durations 被传入（不是 None）
+            mock_distribute.assert_called_once()
+            call_kwargs = mock_distribute.call_args
+            asset_durations = call_kwargs.kwargs.get("asset_durations", call_kwargs[1].get("asset_durations"))
+            assert asset_durations is not None, "asset_durations should be fetched even when random_preview=False"
+            assert "a1" in asset_durations
+            assert "a2" in asset_durations
+            assert asset_durations["a1"] == 30.0
+
+        # 验证 asset_repo.get 被调用（说明 _fetch_asset_durations 执行了）
+        assert asset_repo.get.call_count >= 2
+
+    def test_asset_durations_fetched_with_random_preview(self):
+        """random_preview=True 时仍正常获取 asset_durations（行为不变）"""
+        svc, asset_repo = self._make_service_with_asset_repo()
+
+        template = _make_template("one_take")
+        clip_configs = _make_clip_configs(
+            template_id=template.id,
+            specs=[{"clip_type": ClipType.MAIN, "order": 0, "min_duration": 3.0, "max_duration": 5.0}],
+        )
+
+        with patch("apps.api.app.services.plan_generator_service.distribute_assets") as mock_distribute:
+            svc.generate_from_template(
+                template=template,
+                clip_configs=clip_configs,
+                asset_ids=["a1"],
+                random_preview=True,
+            )
+
+            call_kwargs = mock_distribute.call_args
+            asset_durations = call_kwargs.kwargs.get("asset_durations", call_kwargs[1].get("asset_durations"))
+            assert asset_durations is not None
+            assert "a1" in asset_durations
+
+    def test_no_asset_repo_means_no_durations(self):
+        """_asset_repo 为 None 时 asset_durations 应为 None"""
+        svc, _, _ = _make_generator()  # 默认不带 asset_repo
+
+        template = _make_template("one_take")
+        clip_configs = _make_clip_configs(
+            template_id=template.id,
+            specs=[{"clip_type": ClipType.MAIN, "order": 0, "min_duration": 3.0, "max_duration": 5.0}],
+        )
+
+        with patch("apps.api.app.services.plan_generator_service.distribute_assets") as mock_distribute:
+            svc.generate_from_template(
+                template=template,
+                clip_configs=clip_configs,
+                asset_ids=["a1"],
+            )
+
+            call_kwargs = mock_distribute.call_args
+            asset_durations = call_kwargs.kwargs.get("asset_durations", call_kwargs[1].get("asset_durations"))
+            assert asset_durations is None

tests/unit/test_plan_generator_utils.py

@@ -10,6 +10,7 @@ from packages.domain.edit_plan_clip import EditPlanClip
 from packages.domain.editing_mode import EditingMode
 from packages.domain.plan_generator_utils import (
     DEFAULT_CLIP_DURATION,
+    _calc_random_start_time,
     create_clips_from_configs,
     distribute_assets,
     generate_default_clips,
@@ -647,3 +648,132 @@ class TestConstants:
     def test_default_duration_value(self):
         """默认片段时长应为 5 秒."""
         assert DEFAULT_CLIP_DURATION == 5.0
+
+
+# ── 随机 start_time 与去重测试 ─────────────────────────────────────────
+
+
+class TestCalcRandomStartTime:
+    """_calc_random_start_time 函数测试."""
+
+    def test_no_durations_returns_none(self):
+        """asset_durations 为 None 时返回 None."""
+        result = _calc_random_start_time("a1", 5.0, None)
+        assert result is None
+
+    def test_asset_not_in_durations_returns_none(self):
+        """素材不在 durations 映射中时返回 None."""
+        result = _calc_random_start_time("a_missing", 5.0, {"a1": 60.0})
+        assert result is None
+
+    def test_zero_duration_returns_none(self):
+        """素材时长为 0 时返回 None."""
+        result = _calc_random_start_time("a1", 5.0, {"a1": 0.0})
+        assert result is None
+
+    def test_clip_longer_than_asset_returns_zero(self):
+        """片段时长 >= 素材时长时返回 0.0."""
+        result = _calc_random_start_time("a1", 60.0, {"a1": 30.0})
+        assert result == 0.0
+
+    def test_returns_value_in_range(self):
+        """返回值在 [0, max_start] 范围内."""
+        for _ in range(50):
+            result = _calc_random_start_time("a1", 5.0, {"a1": 30.0})
+            assert result is not None
+            assert 0.0 <= result <= 25.0
+
+    def test_avoids_used_segments(self):
+        """生成的 start_time 不与已使用段重叠."""
+        used = {"a1": [(10.0, 15.0), (20.0, 25.0)]}
+        for _ in range(100):
+            result = _calc_random_start_time("a1", 3.0, {"a1": 30.0}, used)
+            assert result is not None
+            end = result + 3.0
+            # 不应与 [10,15) 重叠
+            assert not (result < 15.0 and 10.0 < end)
+            # 不应与 [20,25) 重叠
+            assert not (result < 25.0 and 20.0 < end)
+
+    def test_fallback_when_no_space(self):
+        """所有空间都被占用时，回退到最后已使用段之后或 0."""
+        used = {"a1": [(0.0, 28.0)]}
+        result = _calc_random_start_time("a1", 5.0, {"a1": 30.0}, used)
+        assert result is not None
+        # 应该返回 last_used_end (28.0) 但受限于 max_start (25.0)，取 min
+        assert result == 25.0
+
+    def test_no_used_segments_for_asset(self):
+        """used_segments 中没有当前 asset_id 时正常随机."""
+        used = {"a2": [(10.0, 15.0)]}
+        for _ in range(20):
+            result = _calc_random_start_time("a1", 5.0, {"a1": 30.0}, used)
+            assert result is not None
+            assert 0.0 <= result <= 25.0
+
+
+class TestDistributeAssetsOverlapAvoidance:
+    """测试 distribute_assets 中同一素材被多次使用时的去重逻辑."""
+
+    def test_same_asset_reused_no_overlap(self):
+        """同一素材分配给多个 clip 时，start_time 不重叠."""
+        # 2 个 clip 共享同一素材，素材时长足够
+        clips = [
+            EditPlanClip.create(plan_id="p1", clip_type=ClipType.MAIN.value, order=0, duration=5.0),
+            EditPlanClip.create(plan_id="p1", clip_type=ClipType.MAIN.value, order=1, duration=5.0),
+        ]
+        asset_ids = ["a1", "a1"]  # 同一素材用两次
+        durations = {"a1": 60.0}
+
+        distribute_assets(clips, asset_ids, EditingMode.ONE_TAKE.value, asset_durations=durations)
+
+        # 两个 clip 都应有 start_time
+        assert clips[0].start_time is not None
+        assert clips[1].start_time is not None
+        # 两段不应重叠
+        s1, e1 = clips[0].start_time, clips[0].start_time + 5.0
+        s2, e2 = clips[1].start_time, clips[1].start_time + 5.0
+        assert not (s1 < e2 and s2 < e1), f"Segments overlap: [{s1},{e1}) and [{s2},{e2})"
+
+    def test_different_assets_independent(self):
+        """不同素材各自独立随机，互不影响."""
+        clips = [
+            EditPlanClip.create(plan_id="p1", clip_type=ClipType.MAIN.value, order=0, duration=5.0),
+            EditPlanClip.create(plan_id="p1", clip_type=ClipType.MAIN.value, order=1, duration=5.0),
+        ]
+        asset_ids = ["a1", "a2"]
+        durations = {"a1": 60.0, "a2": 60.0}
+
+        distribute_assets(clips, asset_ids, EditingMode.ONE_TAKE.value, asset_durations=durations)
+
+        assert clips[0].asset_id == "a1"
+        assert clips[1].asset_id == "a2"
+        assert clips[0].start_time is not None
+        assert clips[1].start_time is not None
+
+    def test_no_durations_still_works(self):
+        """不传 asset_durations 时仍正常分配（start_time 为 None）."""
+        clips = [
+            EditPlanClip.create(plan_id="p1", clip_type=ClipType.MAIN.value, order=0, duration=5.0),
+        ]
+        distribute_assets(clips, ["a1"], EditingMode.ONE_TAKE.value)
+        assert clips[0].asset_id == "a1"
+        # start_time 默认为 0.0（模型默认值）
+        assert clips[0].start_time == 0.0
+
+    def test_voice_pip_overlap_avoidance(self):
+        """VOICE_PIP 模式下同一素材多次使用时避免重叠."""
+        clips = [
+            EditPlanClip.create(plan_id="p1", clip_type="background", order=0, duration=5.0),
+            EditPlanClip.create(plan_id="p1", clip_type="b_roll", order=1, duration=5.0),
+        ]
+        asset_ids = ["a1", "a1"]  # 同一素材用两次
+        durations = {"a1": 60.0}
+
+        distribute_assets(clips, asset_ids, EditingMode.VOICE_PIP.value, asset_durations=durations)
+
+        assert clips[0].start_time is not None
+        assert clips[1].start_time is not None
+        s1, e1 = clips[0].start_time, clips[0].start_time + 5.0
+        s2, e2 = clips[1].start_time, clips[1].start_time + 5.0
+        assert not (s1 < e2 and s2 < e1), f"Segments overlap: [{s1},{e1}) and [{s2},{e2})"

tests/unit/test_render_audio_pure.py

@@ -11,6 +11,7 @@ from video_processing.render_audio import (
     RenderContext,
     clip_effective_duration,
     clip_has_audio,
+    mix_with_independent_audio,
 )
 from video_processing.unified_render_service import ResolvedClip
 
@@ -153,3 +154,93 @@ class TestRenderContext:
         """音频缓存初始为空."""
         ctx = _make_ctx()
         assert ctx._audio_cache == {}
+
+
+class TestMixWithIndependentAudioVolume:
+    """测试 mix_with_independent_audio 中主视频 clip 音量滤镜是否生效。"""
+
+    def _make_clip(self, volume=None, clip_id="c1", duration=5.0):
+        """创建测试用 ResolvedClip。"""
+        config = {}
+        if volume is not None:
+            config["volume"] = volume
+        return ResolvedClip(
+            clip_id=clip_id,
+            asset_id=f"a_{clip_id}",
+            local_path=Path(f"/tmp/{clip_id}.mp4"),
+            clip_type="main",
+            order=0,
+            duration=duration,
+            actual_duration=duration,
+            config=config,
+        )
+
+    def test_main_clip_volume_zero_applied_in_filter(self):
+        """volume=0 的主视频 clip 应在 filter_complex 中包含 volume=0.0000。"""
+        from unittest.mock import patch
+
+        ctx = _make_ctx()
+        main_clip = self._make_clip(volume=0, clip_id="m1")
+        captured_cmd = {}
+
+        def fake_run_ffmpeg(cmd):
+            captured_cmd["cmd"] = cmd
+
+        with patch("video_processing.render_audio.run_ffmpeg", side_effect=fake_run_ffmpeg):
+            mix_with_independent_audio(
+                ctx=ctx,
+                main_clips=[main_clip],
+                audio_clips=[],
+                output_path=Path("/tmp/out.m4a"),
+                video_duration=5.0,
+            )
+
+        filter_complex = captured_cmd["cmd"][captured_cmd["cmd"].index("-filter_complex") + 1]
+        assert "volume=0.0000" in filter_complex, f"volume filter missing in: {filter_complex}"
+
+    def test_main_clip_default_volume_no_filter(self):
+        """默认 volume=1.0 时不应添加 volume 滤镜。"""
+        from unittest.mock import patch
+
+        ctx = _make_ctx()
+        main_clip = self._make_clip(clip_id="m1")  # no volume set
+        captured_cmd = {}
+
+        def fake_run_ffmpeg(cmd):
+            captured_cmd["cmd"] = cmd
+
+        with patch("video_processing.render_audio.run_ffmpeg", side_effect=fake_run_ffmpeg):
+            mix_with_independent_audio(
+                ctx=ctx,
+                main_clips=[main_clip],
+                audio_clips=[],
+                output_path=Path("/tmp/out.m4a"),
+                video_duration=5.0,
+            )
+
+        filter_complex = captured_cmd["cmd"][captured_cmd["cmd"].index("-filter_complex") + 1]
+        # 默认音量不应出现 volume= 滤镜
+        assert "volume=" not in filter_complex, f"unexpected volume filter in: {filter_complex}"
+
+    def test_main_clip_partial_volume_applied(self):
+        """volume=0.5 的主视频 clip 应包含 volume=0.5000。"""
+        from unittest.mock import patch
+
+        ctx = _make_ctx()
+        main_clip = self._make_clip(volume=0.5, clip_id="m1")
+        captured_cmd = {}
+
+        def fake_run_ffmpeg(cmd):
+            captured_cmd["cmd"] = cmd
+
+        with patch("video_processing.render_audio.run_ffmpeg", side_effect=fake_run_ffmpeg):
+            mix_with_independent_audio(
+                ctx=ctx,
+                main_clips=[main_clip],
+                audio_clips=[],
+                output_path=Path("/tmp/out.m4a"),
+                video_duration=5.0,
+            )
+
+        filter_complex = captured_cmd["cmd"][captured_cmd["cmd"].index("-filter_complex") + 1]
+        assert "volume=0.5000" in filter_complex, f"volume filter missing in: {filter_complex}"

tests/unit/test_render_subtitles.py

@@ -80,13 +80,13 @@ class TestBuildAssStyle:
     def test_basic_style(self):
         style = _build_ass_style("Default")
         assert style.startswith("Style: Default,")
-        assert "思源黑体" in style
-        assert "48" in style  # font_size
+        assert "Noto Sans SC" in style
+        assert "65" in style  # font_size 48*1.35=65
 
     def test_custom_font(self):
         style = _build_ass_style("Custom", font_name="Arial", font_size=32)
         assert "Arial" in style
-        assert ",32," in style
+        assert ",43," in style  # font_size 32*1.35=43
 
     def test_bold(self):
         style = _build_ass_style("Bold", bold=True)

tests/unit/test_render_subtitles_pure.py

@@ -84,8 +84,8 @@ class TestBuildAssStyle:
         """基本样式行包含关键字段."""
         line = _build_ass_style("Default")
         assert line.startswith("Style: Default,")
-        assert "思源黑体" in line
-        assert "48" in line  # font_size
+        assert "Noto Sans SC" in line
+        assert "65" in line  # font_size 48*1.35=65
 
     def test_bold_enabled(self):
         """加粗时Bold=-1."""
@@ -112,7 +112,7 @@ class TestBuildAssStyle:
         """自定义字号."""
         line = _build_ass_style("Big", font_size=72)
         parts = line.split(",")
-        assert parts[2] == "72"  # Fontsize
+        assert parts[2] == "97"  # Fontsize 72*1.35=97
 
     def test_custom_alignment(self):
         """自定义对齐方式."""

tests/unit/test_thumbnail_generator.py

@@ -86,3 +86,138 @@ class TestFormatSeekTime:
         sec_parts = parts[2].split(".")
         assert len(sec_parts) == 2
         assert len(sec_parts[1]) == 2  # 两位小数
+
+
+# ── MediaKit 智能抽帧集成测试 ──────────────────────────────────────────
+
+
+class TestExtractFramesViaMediakit:
+    """_extract_frames_via_mediakit 函数测试."""
+
+    def test_mediakit_not_configured_returns_none(self, tmp_path, monkeypatch):
+        """MediaKit 未配置时返回 None."""
+        from video_processing.thumbnail_generator import _extract_frames_via_mediakit
+
+        # Mock get_mediakit_client 返回不可用客户端
+        class FakeClient:
+            is_available = False
+
+        monkeypatch.setattr(
+            "packages.shared.mediakit_client.get_mediakit_client",
+            lambda: FakeClient(),
+        )
+
+        video_file = tmp_path / "test.mp4"
+        video_file.write_bytes(b"fake video")
+
+        result = _extract_frames_via_mediakit(str(video_file), "plan1", 3)
+        assert result is None
+
+    def test_mediakit_success_returns_frames(self, tmp_path, monkeypatch):
+        """MediaKit 成功时返回帧列表."""
+        from video_processing.thumbnail_generator import _extract_frames_via_mediakit
+
+        class FakeClient:
+            is_available = True
+
+            def extract_frames(self, video_url, strategy, max_frames):
+                return [
+                    {"image_url": "https://example.com/frame1.jpg", "timestamp": 1.5},
+                    {"image_url": "https://example.com/frame2.jpg", "timestamp": 3.2},
+                ]
+
+        monkeypatch.setattr(
+            "packages.shared.mediakit_client.get_mediakit_client",
+            lambda: FakeClient(),
+        )
+        # Mock upload_to_oss
+        monkeypatch.setattr(
+            "video_processing.oss_helpers.upload_to_oss",
+            lambda path, key: f"https://oss.example.com/{key}",
+        )
+        # Mock httpx.get for downloading frame
+        import httpx
+
+        class FakeResponse:
+            status_code = 200
+            content = b"fake image data"
+
+            def raise_for_status(self):
+                pass
+
+        monkeypatch.setattr("httpx.get", lambda url, **kw: FakeResponse())
+
+        video_file = tmp_path / "test.mp4"
+        video_file.write_bytes(b"fake video")
+
+        result = _extract_frames_via_mediakit(str(video_file), "plan1", 2)
+        assert result is not None
+        assert len(result) == 2
+        assert result[0]["timestamp"] == 1.5
+        assert result[1]["timestamp"] == 3.2
+
+    def test_mediakit_failure_returns_none(self, tmp_path, monkeypatch):
+        """MediaKit 调用失败时返回 None."""
+        from video_processing.thumbnail_generator import _extract_frames_via_mediakit
+
+        class FakeClient:
+            is_available = True
+
+            def extract_frames(self, video_url, strategy, max_frames):
+                return None
+
+        monkeypatch.setattr(
+            "packages.shared.mediakit_client.get_mediakit_client",
+            lambda: FakeClient(),
+        )
+        monkeypatch.setattr(
+            "video_processing.oss_helpers.upload_to_oss",
+            lambda path, key: f"https://oss.example.com/{key}",
+        )
+
+        video_file = tmp_path / "test.mp4"
+        video_file.write_bytes(b"fake video")
+
+        result = _extract_frames_via_mediakit(str(video_file), "plan1", 3)
+        assert result is None
+
+
+class TestExtractAndUploadCoverFramesFallback:
+    """extract_and_upload_cover_frames 降级逻辑测试."""
+
+    def test_fallback_to_ffmpeg_when_mediakit_fails(self, tmp_path, monkeypatch):
+        """MediaKit 失败时降级到 ffmpeg 抽帧."""
+        from video_processing.thumbnail_generator import extract_and_upload_cover_frames
+
+        # Mock MediaKit 返回 None（未配置或失败）
+        class FakeClient:
+            is_available = False
+
+        monkeypatch.setattr(
+            "packages.shared.mediakit_client.get_mediakit_client",
+            lambda: FakeClient(),
+        )
+
+        # Mock ffmpeg 抽帧
+        monkeypatch.setattr(
+            "video_processing.thumbnail_generator.extract_first_frame",
+            lambda video_path, output_path, **kw: output_path,
+        )
+        # Mock upload
+        monkeypatch.setattr(
+            "video_processing.oss_helpers.upload_to_oss",
+            lambda path, key: f"https://oss.example.com/{key}",
+        )
+        # Mock probe_duration
+        monkeypatch.setattr(
+            "video_processing.ffmpeg_utils.probe_duration",
+            lambda path: 60.0,
+        )
+
+        video_file = tmp_path / "test.mp4"
+        video_file.write_bytes(b"fake video")
+
+        result = extract_and_upload_cover_frames(str(video_file), "plan1", num_frames=2)
+        assert len(result) == 2
+        assert all("url" in item for item in result)
+        assert all("position" in item for item in result)

tests/unit/test_title_render_consistency.py

@@ -1,7 +1,7 @@
 """标题渲染前后端一致性测试。
 
 验证 build_ass_content 生成的 ASS 样式参数与前端 drawTitleOnCanvas.ts 一致：
-- 字号上限 36px
+- 字号不再设置上限，由前端/调用方控制
 - 描边宽度 2px
 - 阴影 blur=4, offset=2
 - boolean stroke/shadow 自动转换
@@ -10,14 +10,16 @@
 import sys
 from pathlib import Path
 
+import pytest
+
 # Ensure packages is importable
 sys.path.insert(0, str(Path(__file__).resolve().parents[2] / "packages"))
 
 from domain.ass_subtitle_builder import build_ass_content, build_ass_style
 
 
-class TestFontSizeCap:
-    """字号上限应与前端 Math.min(settings.size, 36) 一致。"""
+class TestFontSize:
+    """字号处理：默认值与保留逻辑，不再做上限截断。"""
 
     def test_default_font_size_is_36(self):
         """无 size 字段时，默认字号应为 36。"""
@@ -29,7 +31,7 @@ class TestFontSizeCap:
             title_text="测试标题",
             title_config=config,
         )
-        assert ",36," in content, f"默认字号应为36，实际内容: {content}"
+        assert ",49," in content, f"默认字号36应补偿为49(36*1.35)，实际内容: {content}"
 
     def test_size_32_preserved(self):
         """size=32 应原样使用。"""
@@ -41,10 +43,10 @@ class TestFontSizeCap:
             title_text="测试标题",
             title_config=config,
         )
-        assert ",32," in content
+        assert ",43," in content  # 32*1.35=43
 
-    def test_size_60_capped_at_36(self):
-        """size=60 应被 cap 到 36。"""
+    def test_size_60_preserved(self):
+        """size=60 应原样保留（字号上限已移除）。"""
         config = {"size": 60}
         content = build_ass_content(
             video_width=1080,
@@ -53,14 +55,40 @@ class TestFontSizeCap:
             title_text="测试标题",
             title_config=config,
         )
-        # 解析 Style 行的 Fontsize 字段（第3个字段，索引2）
         style_line = [line for line in content.splitlines() if line.startswith("Style: TitleStyle")][0]
         fields = [f.strip() for f in style_line.split(",")]
         font_size = int(fields[2])
-        assert font_size == 36, f"字号60应被cap到36, 实际={font_size}"
+        assert font_size == 81, f"字号60应补偿为81(60*1.35), 实际={font_size}"
+
+    def test_font_size_alias_normalized(self):
+        """前端传 font_size 应归一化为 size。"""
+        config = {"font_size": 52}
+        content = build_ass_content(
+            video_width=1080,
+            video_height=1920,
+            video_duration=10.0,
+            title_text="测试标题",
+            title_config=config,
+        )
+        style_line = [line for line in content.splitlines() if line.startswith("Style: TitleStyle")][0]
+        fields = [f.strip() for f in style_line.split(",")]
+        assert fields[2] == "70", f"font_size=52 应补偿为70(52*1.35), 实际={fields[2]}"
+
+    def test_font_color_alias_normalized(self):
+        """前端传 font_color 应归一化为 color。"""
+        config = {"font_color": "#00FF00"}
+        content = build_ass_content(
+            video_width=1080,
+            video_height=1920,
+            video_duration=10.0,
+            title_text="测试标题",
+            title_config=config,
+        )
+        # 绿色 → &H00FF00
+        assert "&H00FF00" in content
 
     def test_size_24_preserved(self):
-        """size=24 应原样使用（小于36，不cap）。"""
+        """size=24 应原样使用。"""
         config = {"size": 24}
         content = build_ass_content(
             video_width=1080,
@@ -69,7 +97,7 @@ class TestFontSizeCap:
             title_text="测试标题",
             title_config=config,
         )
-        assert ",24," in content
+        assert ",32," in content  # 24*1.35=32.4→32
 
 
 class TestBooleanStrokeNormalization:
@@ -202,9 +230,9 @@ class TestFullStyleConsistency:
         fields = [f.strip() for f in style_line.split(",")]
 
         # Fontname
-        assert fields[1] == "思源黑体"
-        # Fontsize = 28 (小于36，不cap)
-        assert fields[2] == "28"
+        assert fields[1] == "Noto Sans SC"
+        # Fontsize = 28*1.35=37.8→38
+        assert fields[2] == "38"
         # Bold = -1 (True)
         assert fields[7] == "-1"
         # Outline width = 2 (前端默认 stroke width)
@@ -213,3 +241,39 @@ class TestFullStyleConsistency:
         assert int(fields[17]) == 2
         # Alignment = 8 (top)
         assert int(fields[18]) == 8
+
+
+class TestTitleSlashNewline:
+    """用户输入 / 或 ／ 应触发标题换行。"""
+
+    def test_halfwidth_slash_in_title(self):
+        content = build_ass_content(
+            video_width=1080,
+            video_height=1920,
+            video_duration=10.0,
+            title_text="第一行/第二行",
+            title_config={"size": 48},
+        )
+        for line in content.splitlines():
+            if "Dialogue" in line and "TitleStyle" in line:
+                assert "\\N" in line, f"斜杠应转为换行: {line}"
+                assert "第一行" in line
+                assert "第二行" in line
+                break
+        else:
+            pytest.fail("未找到 TitleStyle Dialogue 行")
+
+    def test_fullwidth_slash_in_title(self):
+        content = build_ass_content(
+            video_width=1080,
+            video_height=1920,
+            video_duration=10.0,
+            title_text="第一行／第二行",
+            title_config={"size": 48},
+        )
+        for line in content.splitlines():
+            if "Dialogue" in line and "TitleStyle" in line:
+                assert "\\N" in line, f"全角斜杠应转为换行: {line}"
+                break
+        else:
+            pytest.fail("未找到 TitleStyle Dialogue 行")

tests/unit/test_tts_preview.py

@@ -0,0 +1,441 @@
+"""Tests for POST /tts/preview endpoint."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+from unittest.mock import MagicMock, patch
+
+import pytest
+from pydantic import ValidationError
+
+
+@dataclass
+class FakeSynthesizeResult:
+    audio_url: str
+    duration: float = 0.0
+    file_size: int = 0
+    request_id: str = ""
+
+
+class TestTTSPreviewEndpoint:
+    """Integration-style tests for the /tts/preview route."""
+
+    def _make_client(self, app):
+        from fastapi.testclient import TestClient
+
+        return TestClient(app)
+
+    def test_schema_preview_request_validation(self):
+        """TTSPreviewRequest rejects text > 200 chars and empty voice_id."""
+        from app.schemas.tts import TTSPreviewRequest
+
+        # Valid
+        req = TTSPreviewRequest(text="hello", voice_id="v1")
+        assert req.text == "hello"
+        assert req.voice_id == "v1"
+        assert req.speed == 1.0
+
+        # Empty voice_id rejected
+        with pytest.raises(ValidationError):
+            TTSPreviewRequest(text="hello", voice_id="")
+
+        # Text > 200 chars rejected
+        with pytest.raises(ValidationError):
+            TTSPreviewRequest(text="a" * 201, voice_id="v1")
+
+    def test_schema_preview_response(self):
+        """TTSPreviewResponse serialization."""
+        from app.schemas.tts import TTSPreviewResponse
+
+        resp = TTSPreviewResponse(audio_url="https://example.com/audio.mp3")
+        assert resp.audio_url == "https://example.com/audio.mp3"
+        assert resp.duration is None
+
+        resp2 = TTSPreviewResponse(audio_url="https://x.com/a.mp3", duration=3.5)
+        assert resp2.duration == 3.5
+
+    def test_preview_success(self):
+        """Successful preview returns audio_url."""
+        from app.schemas.tts import TTSPreviewRequest
+        from fastapi import FastAPI
+
+        app = FastAPI()
+
+        # We need to register the route with proper dependencies
+        from app.api.routes.tts import router
+
+        app.include_router(router, prefix="/tts")
+
+        # Override dependencies
+        from app.auth import get_current_user
+        from app.dependencies import get_cosyvoice_service
+
+        fake_user = MagicMock()
+        fake_user.user.id = "user-1"
+        app.dependency_overrides[get_current_user] = lambda: fake_user
+
+        mock_service = MagicMock()
+        mock_service.synthesize_speech.return_value = FakeSynthesizeResult(
+            audio_url="https://cosyvoice.example.com/audio.mp3",
+            duration=2.5,
+        )
+        app.dependency_overrides[get_cosyvoice_service] = lambda: mock_service
+
+        from app.dependencies import get_voice_clone_profile_repository
+
+        mock_clone_repo = MagicMock()
+        mock_clone_repo.get.return_value = None  # no profile found = preset voice
+        app.dependency_overrides[get_voice_clone_profile_repository] = lambda: mock_clone_repo
+
+        client = self._make_client(app)
+        resp = client.post(
+            "/tts/preview",
+            json={"text": "你好世界", "voice_id": "longxiaochun"},
+        )
+        assert resp.status_code == 200
+        data = resp.json()
+        assert data["audio_url"] == "https://cosyvoice.example.com/audio.mp3"
+        assert data["duration"] == 2.5
+
+        mock_service.synthesize_speech.assert_called_once_with(
+            text="你好世界",
+            voice_id="longxiaochun",
+            speed=1.0,
+        )
+
+    def test_preview_with_speed(self):
+        """Custom speed is passed through to CosyVoice."""
+        from fastapi import FastAPI
+
+        app = FastAPI()
+        from app.api.routes.tts import router
+
+        app.include_router(router, prefix="/tts")
+
+        from app.auth import get_current_user
+        from app.dependencies import get_cosyvoice_service
+
+        fake_user = MagicMock()
+        fake_user.user.id = "user-1"
+        app.dependency_overrides[get_current_user] = lambda: fake_user
+
+        mock_service = MagicMock()
+        mock_service.synthesize_speech.return_value = FakeSynthesizeResult(
+            audio_url="https://x.com/a.mp3",
+            duration=0.0,
+        )
+        app.dependency_overrides[get_cosyvoice_service] = lambda: mock_service
+
+        from app.dependencies import get_voice_clone_profile_repository
+
+        mock_clone_repo = MagicMock()
+        mock_clone_repo.get.return_value = None  # no profile found = preset voice
+        app.dependency_overrides[get_voice_clone_profile_repository] = lambda: mock_clone_repo
+
+        client = self._make_client(app)
+        resp = client.post(
+            "/tts/preview",
+            json={"text": "测试", "voice_id": "v1", "speed": 1.5},
+        )
+        assert resp.status_code == 200
+        data = resp.json()
+        assert data["audio_url"] == "https://x.com/a.mp3"
+        assert data["duration"] is None  # 0.0 -> None
+
+        mock_service.synthesize_speech.assert_called_once_with(
+            text="测试",
+            voice_id="v1",
+            speed=1.5,
+        )
+
+    def test_preview_cosyvoice_error_returns_502(self):
+        """CosyVoice failure returns 502."""
+        from fastapi import FastAPI
+
+        from packages.application.cosyvoice_service import CosyVoiceError
+
+        app = FastAPI()
+        from app.api.routes.tts import router
+
+        app.include_router(router, prefix="/tts")
+
+        from app.auth import get_current_user
+        from app.dependencies import get_cosyvoice_service
+
+        fake_user = MagicMock()
+        fake_user.user.id = "user-1"
+        app.dependency_overrides[get_current_user] = lambda: fake_user
+
+        mock_service = MagicMock()
+        mock_service.synthesize_speech.side_effect = CosyVoiceError("API timeout")
+        app.dependency_overrides[get_cosyvoice_service] = lambda: mock_service
+
+        from app.dependencies import get_voice_clone_profile_repository
+
+        mock_clone_repo = MagicMock()
+        mock_clone_repo.get.return_value = None  # no profile found = preset voice
+        app.dependency_overrides[get_voice_clone_profile_repository] = lambda: mock_clone_repo
+
+        client = self._make_client(app)
+        resp = client.post(
+            "/tts/preview",
+            json={"text": "测试", "voice_id": "v1"},
+        )
+        assert resp.status_code == 502
+        assert "TTS 合成失败" in resp.json()["detail"]
+
+    def test_preview_value_error_returns_400(self):
+        """Invalid params return 400."""
+        from fastapi import FastAPI
+
+        app = FastAPI()
+        from app.api.routes.tts import router
+
+        app.include_router(router, prefix="/tts")
+
+        from app.auth import get_current_user
+        from app.dependencies import get_cosyvoice_service
+
+        fake_user = MagicMock()
+        fake_user.user.id = "user-1"
+        app.dependency_overrides[get_current_user] = lambda: fake_user
+
+        mock_service = MagicMock()
+        mock_service.synthesize_speech.side_effect = ValueError("text 不能为空")
+        app.dependency_overrides[get_cosyvoice_service] = lambda: mock_service
+
+        from app.dependencies import get_voice_clone_profile_repository
+
+        mock_clone_repo = MagicMock()
+        mock_clone_repo.get.return_value = None  # no profile found = preset voice
+        app.dependency_overrides[get_voice_clone_profile_repository] = lambda: mock_clone_repo
+
+        client = self._make_client(app)
+        resp = client.post(
+            "/tts/preview",
+            json={"text": "测试", "voice_id": "v1"},
+        )
+        assert resp.status_code == 400
+        assert "text 不能为空" in resp.json()["detail"]
+
+    def test_preview_text_too_long_returns_422(self):
+        """Text > 200 chars is rejected by Pydantic validation."""
+        from fastapi import FastAPI
+
+        app = FastAPI()
+        from app.api.routes.tts import router
+
+        app.include_router(router, prefix="/tts")
+
+        from app.auth import get_current_user
+        from app.dependencies import get_cosyvoice_service
+
+        fake_user = MagicMock()
+        fake_user.user.id = "user-1"
+        app.dependency_overrides[get_current_user] = lambda: fake_user
+
+        mock_service = MagicMock()
+        app.dependency_overrides[get_cosyvoice_service] = lambda: mock_service
+
+        from app.dependencies import get_voice_clone_profile_repository
+
+        mock_clone_repo = MagicMock()
+        mock_clone_repo.get.return_value = None  # no profile found = preset voice
+        app.dependency_overrides[get_voice_clone_profile_repository] = lambda: mock_clone_repo
+
+        client = self._make_client(app)
+        resp = client.post(
+            "/tts/preview",
+            json={"text": "a" * 201, "voice_id": "v1"},
+        )
+        assert resp.status_code == 422
+
+    def test_preview_empty_voice_id_returns_422(self):
+        """Empty voice_id is rejected by Pydantic validation."""
+        from fastapi import FastAPI
+
+        app = FastAPI()
+        from app.api.routes.tts import router
+
+        app.include_router(router, prefix="/tts")
+
+        from app.auth import get_current_user
+        from app.dependencies import get_cosyvoice_service
+
+        fake_user = MagicMock()
+        fake_user.user.id = "user-1"
+        app.dependency_overrides[get_current_user] = lambda: fake_user
+
+        mock_service = MagicMock()
+        app.dependency_overrides[get_cosyvoice_service] = lambda: mock_service
+
+        from app.dependencies import get_voice_clone_profile_repository
+
+        mock_clone_repo = MagicMock()
+        mock_clone_repo.get.return_value = None  # no profile found = preset voice
+        app.dependency_overrides[get_voice_clone_profile_repository] = lambda: mock_clone_repo
+
+        client = self._make_client(app)
+        resp = client.post(
+            "/tts/preview",
+            json={"text": "hello", "voice_id": ""},
+        )
+        assert resp.status_code == 422
+
+    def test_preview_clone_voice_resolves_to_cosyvoice_id(self):
+        """Clone voice UUID is resolved to CosyVoice voice_id."""
+        from dataclasses import dataclass
+
+        from fastapi import FastAPI
+
+        app = FastAPI()
+        from app.api.routes.tts import router
+
+        app.include_router(router, prefix="/tts")
+
+        from app.auth import get_current_user
+        from app.dependencies import get_cosyvoice_service, get_voice_clone_profile_repository
+
+        fake_user = MagicMock()
+        fake_user.user.id = "user-1"
+        app.dependency_overrides[get_current_user] = lambda: fake_user
+
+        mock_service = MagicMock()
+        mock_service.synthesize_speech.return_value = FakeSynthesizeResult(
+            audio_url="https://x.com/cloned.mp3",
+            duration=1.8,
+        )
+        app.dependency_overrides[get_cosyvoice_service] = lambda: mock_service
+
+        # Mock voice clone profile with voice_id
+        mock_profile = MagicMock()
+        mock_profile.user_id = "user-1"
+        mock_profile.voice_id = "cosyvoice_actual_voice_123"
+        mock_clone_repo = MagicMock()
+        mock_clone_repo.get.return_value = mock_profile
+        app.dependency_overrides[get_voice_clone_profile_repository] = lambda: mock_clone_repo
+
+        client = self._make_client(app)
+        # Frontend sends the profile UUID as voice_id
+        resp = client.post(
+            "/tts/preview",
+            json={"text": "克隆音色测试", "voice_id": "abc123-uuid-of-profile"},
+        )
+        assert resp.status_code == 200
+        data = resp.json()
+        assert data["audio_url"] == "https://x.com/cloned.mp3"
+
+        # Verify CosyVoice was called with the resolved voice_id, not the UUID
+        mock_service.synthesize_speech.assert_called_once_with(
+            text="克隆音色测试",
+            voice_id="cosyvoice_actual_voice_123",
+            speed=1.0,
+        )
+        # Verify repo was queried with the UUID
+        mock_clone_repo.get.assert_called_once_with("abc123-uuid-of-profile")
+
+    def test_preview_clone_voice_incomplete_returns_400(self):
+        """Clone profile with empty voice_id returns 400."""
+        from fastapi import FastAPI
+
+        app = FastAPI()
+        from app.api.routes.tts import router
+
+        app.include_router(router, prefix="/tts")
+
+        from app.auth import get_current_user
+        from app.dependencies import get_voice_clone_profile_repository
+
+        fake_user = MagicMock()
+        fake_user.user.id = "user-1"
+        app.dependency_overrides[get_current_user] = lambda: fake_user
+
+        # Mock voice clone profile with empty voice_id (clone not finished)
+        mock_profile = MagicMock()
+        mock_profile.user_id = "user-1"
+        mock_profile.voice_id = ""
+        mock_clone_repo = MagicMock()
+        mock_clone_repo.get.return_value = mock_profile
+        app.dependency_overrides[get_voice_clone_profile_repository] = lambda: mock_clone_repo
+
+        client = self._make_client(app)
+        resp = client.post(
+            "/tts/preview",
+            json={"text": "测试未完成克隆", "voice_id": "abc123-uuid"},
+        )
+        assert resp.status_code == 400
+        assert "音色克隆尚未完成" in resp.json()["detail"]
+
+    def test_preview_preset_voice_passthrough(self):
+        """Preset voice ID (not a profile UUID) passes through unchanged."""
+        from fastapi import FastAPI
+
+        app = FastAPI()
+        from app.api.routes.tts import router
+
+        app.include_router(router, prefix="/tts")
+
+        from app.auth import get_current_user
+        from app.dependencies import get_cosyvoice_service, get_voice_clone_profile_repository
+
+        fake_user = MagicMock()
+        fake_user.user.id = "user-1"
+        app.dependency_overrides[get_current_user] = lambda: fake_user
+
+        mock_service = MagicMock()
+        mock_service.synthesize_speech.return_value = FakeSynthesizeResult(
+            audio_url="https://x.com/preset.mp3",
+            duration=2.0,
+        )
+        app.dependency_overrides[get_cosyvoice_service] = lambda: mock_service
+
+        # Mock repo returns None (preset voice, not a clone profile)
+        mock_clone_repo = MagicMock()
+        mock_clone_repo.get.return_value = None
+        app.dependency_overrides[get_voice_clone_profile_repository] = lambda: mock_clone_repo
+
+        client = self._make_client(app)
+        resp = client.post(
+            "/tts/preview",
+            json={"text": "预设音色测试", "voice_id": "longxiaoxia_v3"},
+        )
+        assert resp.status_code == 200
+
+        # Verify CosyVoice was called with the original preset voice_id
+        mock_service.synthesize_speech.assert_called_once_with(
+            text="预设音色测试",
+            voice_id="longxiaoxia_v3",
+            speed=1.0,
+        )
+
+    def test_preview_clone_voice_wrong_user_returns_403(self):
+        """Accessing another user's clone profile returns 403."""
+        from fastapi import FastAPI
+
+        app = FastAPI()
+        from app.api.routes.tts import router
+
+        app.include_router(router, prefix="/tts")
+
+        from app.auth import get_current_user
+        from app.dependencies import get_voice_clone_profile_repository
+
+        fake_user = MagicMock()
+        fake_user.user.id = "user-1"
+        app.dependency_overrides[get_current_user] = lambda: fake_user
+
+        # Mock profile belonging to a different user
+        mock_profile = MagicMock()
+        mock_profile.user_id = "user-2"
+        mock_profile.voice_id = "cosyvoice_voice_xyz"
+        mock_clone_repo = MagicMock()
+        mock_clone_repo.get.return_value = mock_profile
+        app.dependency_overrides[get_voice_clone_profile_repository] = lambda: mock_clone_repo
+
+        client = self._make_client(app)
+        resp = client.post(
+            "/tts/preview",
+            json={"text": "越权测试", "voice_id": "other-user-profile-uuid"},
+        )
+        assert resp.status_code == 403
+        assert "无权访问该音色" in resp.json()["detail"]