feat(#1970): 新 API 字段 + 叙事模式 PR3 - assembly_mode/script_id/tts_*/video_ratio (#1976)

Co-authored-by: xiaoxia <dev@xiaoxiajianji.com>
Co-committed-by: xiaoxia <dev@xiaoxiajianji.com>

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

@@ -16,10 +16,12 @@ from app.core.task_enqueue import (
 from app.dependencies import (
     get_asset_library_repository,
     get_asset_repository,
+    get_cosyvoice_service,
     get_db_session,
     get_generated_video_repository,
     get_generation_task_repository,
     get_project_repository,
+    get_voice_clone_profile_repository,
 )
 from app.schemas.generated_video import (
     GeneratedVideoResponse,
@@ -132,6 +134,8 @@ def _select_assets_from_library(
     mode: str,
     count: int,
     rng=None,
+    script_tags: list | None = None,
+    tag_names_by_id: dict | None = None,
 ) -> list[str]:
     """根据选取模式从素材库中选取 ready 状态的视频素材 ID。
 
@@ -141,6 +145,8 @@ def _select_assets_from_library(
         count: 选取数量，0 表示全部（仅 smart 模式有效）
         rng: 可选随机源（smart 模式排序噪声用），生产环境不传则内部随机；
             测试可注入固定种子或零噪声随机源获得确定性结果。
+        script_tags: #1970 叙事模式文案标签；非空时标签命中素材优先，不足再用其余素材兜底。
+        tag_names_by_id: asset_id → 素材标签名列表（素材只存 tag_ids 时由调用方查名称注入）。
 
     Returns:
         选中的素材 ID 列表
@@ -150,6 +156,20 @@ def _select_assets_from_library(
     if not ready_video_assets:
         return []
 
+    # 叙事模式（#1970 PR3）：文案标签命中池优先；无任何命中时完全降级为现有随机逻辑。
+    if script_tags:
+        from packages.domain.narrative_match import pick_narrative_assets
+
+        limit = count if count > 0 else None
+        picked = pick_narrative_assets(
+            ready_video_assets,
+            script_tags=script_tags,
+            tag_names_by_id=tag_names_by_id,
+            limit=limit,
+            rng=rng,
+        )
+        return [a.id for a in picked]
+
     if mode == "smart":
         # 智能匹配：统一使用 packages/domain/smart_match.py 的多维评分+多样性选取
         # 评分维度：质量分(40%) + 时长适配(30%) + 新鲜度(20%) + 未使用加分(10%)
@@ -162,6 +182,53 @@ def _select_assets_from_library(
     return [a.id for a in ready_video_assets]
 
 
+# #1970 PR3：video_ratio → 默认输出分辨率（显式 output_width/output_height 优先）
+_VIDEO_RATIO_DIMENSIONS = {
+    "9:16": (1080, 1920),
+    "16:9": (1920, 1080),
+    "1:1": (1080, 1080),
+    "3:4": (1080, 1440),
+    "4:3": (1440, 1080),
+}
+
+
+def _resolve_output_dimensions(request: CreateGenerationTaskRequest) -> tuple[int, int]:
+    """解析输出分辨率：显式 output_width/output_height 非旧默认值时优先，否则按 video_ratio。
+
+    前端 #1973 总是同时传 video_ratio 与具体分辨率，两者一致；此函数主要服务
+    只传比例的调用方，并保证旧调用（不传比例）维持 1280x720 行为。
+    """
+    width, height = request.output_width, request.output_height
+    ratio = (request.video_ratio or "").strip()
+    if ratio in _VIDEO_RATIO_DIMENSIONS and (width, height) == (1280, 720):
+        return _VIDEO_RATIO_DIMENSIONS[ratio]
+    return width, height
+
+
+def _load_asset_tag_names(db: Session, assets: list, user_id: str) -> dict[str, list[str]]:
+    """叙事模式：查 TagModel 名称，构造 asset_id → 标签名列表（失败返回空 dict 降级随机）。"""
+    try:
+        from packages.adapters.sqlalchemy_impl.models import AssetTagModel, TagModel
+
+        tag_ids = {tid for a in assets for tid in (getattr(a, "tag_ids", None) or [])}
+        if not tag_ids:
+            return {}
+        name_rows = (
+            db.query(TagModel.id, TagModel.name).filter(TagModel.id.in_(tag_ids), TagModel.user_id == user_id).all()
+        )
+        name_by_id = {row.id: row.name for row in name_rows}
+        links = db.query(AssetTagModel.asset_id, AssetTagModel.tag_id).filter(AssetTagModel.tag_id.in_(tag_ids)).all()
+        index: dict[str, list[str]] = {}
+        for asset_id, tag_id in links:
+            name = name_by_id.get(tag_id)
+            if name:
+                index.setdefault(asset_id, []).append(name)
+        return index
+    except Exception:  # noqa: BLE001 - 标签匹配是加分项，查询失败不阻断生成
+        logger.warning("[叙事模式] 素材标签查询失败，降级随机选片", exc_info=True)
+        return {}
+
+
 def _writeback_edit_plan_config(
     plan_id: str,
     task_id: str,
@@ -169,12 +236,23 @@ def _writeback_edit_plan_config(
     db: Session,
     dedup_enabled: bool | None = None,
     video_index: int | None = None,
+    assembly_mode: str | None = None,
+    script_id: str | None = None,
+    video_ratio: str | None = None,
 ) -> None:
     """[已下沉] 路由层兼容别名 → app.services.generation_common.writeback_edit_plan_config。"""
     from app.services.generation_common import writeback_edit_plan_config
 
     return writeback_edit_plan_config(
-        plan_id, task_id, title_config, db, dedup_enabled=dedup_enabled, video_index=video_index
+        plan_id,
+        task_id,
+        title_config,
+        db,
+        dedup_enabled=dedup_enabled,
+        video_index=video_index,
+        assembly_mode=assembly_mode,
+        script_id=script_id,
+        video_ratio=video_ratio,
     )
 
 
@@ -225,16 +303,63 @@ def create_generation_task(
     asset_library_repository: Any = Depends(get_asset_library_repository),
     asset_repository: Any = Depends(get_asset_repository),
     db: Session = Depends(get_db_session),
+    cosyvoice_service: Any = Depends(get_cosyvoice_service),
+    voice_clone_repository: Any = Depends(get_voice_clone_profile_repository),
 ) -> BatchGenerationTaskResponse:
     logger.info(
-        "[生成任务] 接收请求: user_id=%s, template_id=%s, asset_count=%d, mode=%s, count=%d",
+        "[生成任务] 接收请求: user_id=%s, template_id=%s, asset_count=%d, mode=%s, assembly=%s, count=%d",
         authenticated_user.user.id,
         request.template_id,
         len(request.asset_ids),
         request.asset_select_mode,
+        request.assembly_mode,
         request.count,
     )
 
+    # video_ratio → 默认分辨率（显式分辨率优先）
+    request.output_width, request.output_height = _resolve_output_dimensions(request)
+
+    # ── #1970 PR3 叙事模式：入队前同步合成配音并落为 audio asset ──
+    # 合成结果覆盖 voice_library_id（下游按 audio asset id 消费），失败直接 4xx 不入队。
+    narrative_script_tags: list = []
+    if request.assembly_mode == "narrative":
+        from app.config import settings as _settings
+        from app.services.narrative_service import NarrativeError, prepare_narrative_voice
+
+        from packages.adapters.sqlalchemy_impl.tts_job_repository import SQLAlchemyTTSJobRepository
+
+        try:
+            narrative_ctx = prepare_narrative_voice(
+                db=db,
+                user_id=authenticated_user.user.id,
+                script_id=request.script_id,
+                tts_voice_id=request.tts_voice_id,
+                tts_voice_source=request.tts_voice_source,
+                tts_repository=SQLAlchemyTTSJobRepository(db),
+                cosyvoice_service=cosyvoice_service,
+                voice_clone_repository=voice_clone_repository,
+                asset_repository=asset_repository,
+                asset_library_repository=asset_library_repository,
+                project_repository=project_repository,
+                storage_service=get_storage_service(),
+                points_enabled=bool(getattr(_settings, "points_enabled", False)),
+                is_member=bool(getattr(authenticated_user.user, "is_member", False)),
+                member_type=getattr(authenticated_user.user, "member_type", None),
+            )
+        except NarrativeError as e:
+            logger.warning("[叙事模式] 配音前置处理失败: %s", e.message)
+            raise HTTPException(status_code=e.status_code, detail=e.message) from e
+
+        request.voice_library_id = narrative_ctx.voice_asset_id
+        narrative_script_tags = list(getattr(narrative_ctx.script, "tags", None) or [])
+        logger.info(
+            "[叙事模式] 配音已就绪: script_id=%s, tts_job=%s, voice_asset=%s, duration=%.2f",
+            request.script_id,
+            narrative_ctx.tts_job_id,
+            narrative_ctx.voice_asset_id,
+            narrative_ctx.audio_duration,
+        )
+
     try:
         project_id, asset_library_id = _resolve_project_and_library(
             request, project_repository, asset_library_repository, asset_repository, authenticated_user
@@ -260,19 +385,29 @@ def create_generation_task(
 
         # 素材库自动匹配：当未显式指定 asset_ids 时，按模式自动选取
         if not resolved_asset_ids:
+            _tag_index = (
+                _load_asset_tag_names(db, assets, authenticated_user.user.id) if narrative_script_tags else None
+            )
             resolved_asset_ids = _select_assets_from_library(
                 assets,
                 mode=request.asset_select_mode,
                 count=request.asset_select_count,
+                script_tags=narrative_script_tags or None,
+                tag_names_by_id=_tag_index,
             )
-    elif project_id and not resolved_asset_ids and request.asset_select_mode in ("smart",):
-        # 项目级模式：未指定 asset_ids 且选择了 smart 模式时，也自动选取
+    elif project_id and not resolved_asset_ids and (request.asset_select_mode in ("smart",) or narrative_script_tags):
+        # 项目级模式：未指定 asset_ids 且选择了 smart 模式（或叙事模式按标签匹配）时自动选取
         assets = asset_repository.find_by_project(project_id)
         if assets:
+            _tag_index = (
+                _load_asset_tag_names(db, assets, authenticated_user.user.id) if narrative_script_tags else None
+            )
             resolved_asset_ids = _select_assets_from_library(
                 assets,
                 mode=request.asset_select_mode,
                 count=request.asset_select_count,
+                script_tags=narrative_script_tags or None,
+                tag_names_by_id=_tag_index,
             )
             if not resolved_asset_ids:
                 raise HTTPException(
@@ -337,6 +472,9 @@ def create_generation_task(
                             title_config=fallback_title_config,
                             db=db,
                             dedup_enabled=request.dedup_enabled,
+                            assembly_mode=request.assembly_mode,
+                            script_id=request.script_id or None,
+                            video_ratio=request.video_ratio or None,
                         )
 
                     logger.info(
@@ -685,6 +823,9 @@ def create_generation_task(
                         db=db,
                         dedup_enabled=request.dedup_enabled,
                         video_index=task_index,
+                        assembly_mode=request.assembly_mode,
+                        script_id=request.script_id or None,
+                        video_ratio=request.video_ratio or None,
                     )
 
                 if safe_enqueue_generation_task(