fix(viral-video): STEP2 generate-copy 补齐 voice_id/voice_source/video_ratio/video_model (#2132)

三步链路中 STEP2 handleGenerateCopy 之前漏传这四个字段，导致用户选的音色/画幅/模型在文案生成阶段没写回 job；
STEP3 worker TTS 拿不到 voice_id，总是走默认 longxiaochun_v3 兜底，用户选音失效。

alembic/versions/088_viral_video_copy_result.py

@@ -0,0 +1,51 @@
+"""viral video add copy_result + voice/video columns
+
+Revision ID: 088_viral_video_copy_result
+Revises: 087_viral_video_image_analysis
+Create Date: 2026-10-01
+
+v1.6 爆款视频字段补齐：
+- copy_result JSON: 编导分镜脚本完整结构（overview/scene_and_lighting/shots/hard_constraints/negative_prompts/voiceover_script）
+- voice_id/voice_source: TTS 音色参数
+- video_ratio/video_model: Seedance 视频比例/模型
+注意：线上启动也有幂等 ADD COLUMN 补列逻辑 (_ensure_viral_video_columns)，本 migration 提供标准 Alembic 路径，
+两套机制互不冲突（IF NOT EXISTS 等价行为）。
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "088_viral_video_copy_result"
+down_revision = "087_viral_video_image_analysis"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # 幂等添加列（通过单独执行 + 异常忽略兼容已由 backfill 补上的环境）
+    cols = [
+        ("voice_id", "VARCHAR(200) NOT NULL DEFAULT ''"),
+        ("voice_source", "VARCHAR(20) NOT NULL DEFAULT ''"),
+        ("video_ratio", "VARCHAR(10) NOT NULL DEFAULT '9:16'"),
+        ("video_model", "VARCHAR(100) NOT NULL DEFAULT ''"),
+        ("copy_result", "JSON"),
+    ]
+    conn = op.get_bind()
+    for name, ddl in cols:
+        try:
+            conn.execute(sa.text(f"ALTER TABLE viral_video_jobs ADD COLUMN IF NOT EXISTS {name} {ddl}"))
+        except Exception:
+            # 不支持 IF NOT EXISTS 的库（如老版本 SQLite）直接尝试 ADD COLUMN，失败则忽略
+            try:
+                conn.execute(sa.text(f"ALTER TABLE viral_video_jobs ADD COLUMN {name} {ddl}"))
+            except Exception:
+                pass
+
+
+def downgrade() -> None:
+    for name in ("copy_result", "video_model", "video_ratio", "voice_source", "voice_id"):
+        try:
+            op.drop_column("viral_video_jobs", name)
+        except Exception:
+            pass

alembic/versions/089_viral_video_cols.py

@@ -0,0 +1,62 @@
+"""viral video add storyboard + generated_copy_text (complement 088)
+
+Revision ID: 089_viral_video_cols
+Revises: 088_viral_video_copy_result
+Create Date: 2026-10-01
+
+#2129 兜底迁移：补齐 _VIRAL_VIDEO_BACKFILL_COLS 中所有列，覆盖
+# watchtower 自动部署未跑历史 migration、且 AUTO_CREATE_SCHEMA=false 时
+# _ensure_viral_video_columns 未执行的场景。
+# 幂等 ADD COLUMN IF NOT EXISTS，已存在则跳过。
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "089_viral_video_cols"
+down_revision = "088_viral_video_copy_result"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    # 扩展 alembic_version.version_num 字段长度（原来 VARCHAR(32) 装不下长 revision id）
+    conn = op.get_bind()
+    try:
+        conn.execute(sa.text("ALTER TABLE alembic_version ALTER COLUMN version_num TYPE VARCHAR(256)"))
+    except Exception:
+        pass
+    cols = [
+        ("storyboard", "JSON"),
+        ("generated_copy_text", "TEXT NOT NULL DEFAULT ''"),
+        ("voice_id", "VARCHAR(200) NOT NULL DEFAULT ''"),
+        ("voice_source", "VARCHAR(20) NOT NULL DEFAULT ''"),
+        ("video_ratio", "VARCHAR(10) NOT NULL DEFAULT '9:16'"),
+        ("video_model", "VARCHAR(100) NOT NULL DEFAULT ''"),
+        ("copy_result", "JSON"),
+    ]
+    for name, ddl in cols:
+        try:
+            conn.execute(sa.text(f"ALTER TABLE viral_video_jobs ADD COLUMN IF NOT EXISTS {name} {ddl}"))
+        except Exception:
+            try:
+                conn.execute(sa.text(f"ALTER TABLE viral_video_jobs ADD COLUMN {name} {ddl}"))
+            except Exception:
+                pass
+
+
+def downgrade() -> None:
+    for name in (
+        "copy_result",
+        "video_model",
+        "video_ratio",
+        "voice_source",
+        "voice_id",
+        "generated_copy_text",
+        "storyboard",
+    ):
+        try:
+            op.drop_column("viral_video_jobs", name)
+        except Exception:
+            pass

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

@@ -9,6 +9,11 @@ from fastapi.responses import JSONResponse
 router = APIRouter(tags=["Health"])
 
 
+
+def _pg_url(url: str) -> str:
+    """Convert SQLAlchemy URL (postgresql+psycopg://...) to libpq connection string."""
+    return url.replace("postgresql+psycopg://", "postgresql://", 1).replace("postgresql+psycopg2://", "postgresql://", 1)
+
 @router.get("/health", status_code=status.HTTP_200_OK)
 async def health_check():
     return {
@@ -49,7 +54,7 @@ async def _check_database() -> dict:
             "message": "Using in-memory database",
         }
     try:
-        conn = psycopg.connect(settings.DATABASE_URL, connect_timeout=3)
+        conn = psycopg.connect(_pg_url(settings.DATABASE_URL), connect_timeout=3)
         with conn.cursor() as cur:
             cur.execute("SELECT 1")
             cur.fetchone()
@@ -124,7 +129,7 @@ async def _check_migrations() -> dict:
             "message": "Using in-memory database, no migrations needed",
         }
     try:
-        conn = psycopg.connect(settings.DATABASE_URL, connect_timeout=3)
+        conn = psycopg.connect(_pg_url(settings.DATABASE_URL), connect_timeout=3)
         with conn.cursor() as cur:
             cur.execute("""
                 SELECT COUNT(*) FROM information_schema.tables
@@ -137,3 +142,5 @@ async def _check_migrations() -> dict:
         return {"status": "unhealthy", "message": f"Missing tables, found {count}/5"}
     except Exception as error:
         return {"status": "unhealthy", "message": f"Migration check failed: {error}"}
+
+

apps/api/app/api/routes/viral_video.py

@@ -1,6 +1,6 @@
 """爆款视频 API 路由。
 
-v1.5 三步分步流水线端点（前端新交互）：
+v1.6 三步分步流水线端点（单次 Seedance 出片版）：
   POST   /api/v1/viral-video/analyze-images          阶段1：创建任务 + 仅做图片/视频分析，暂停在 image_analyzed
   POST   /api/v1/viral-video/{job_id}/generate-copy  阶段2：用户填完参数后跑意图+文案+分镜+审核，暂停在 copy_generated
   POST   /api/v1/viral-video/{job_id}/confirm-copy   阶段3：用户确认/编辑文案后跑渲染，直到完成
@@ -10,7 +10,7 @@ v1.5 三步分步流水线端点（前端新交互）：
   POST   /api/v1/viral-video/{job_id}/confirm-intent 旧的意图确认后继续渲染
 
 通用：
-  GET    /api/v1/viral-video/{job_id}                查询任务状态（含 image_analysis/storyboard/generated_copy_text）
+  GET    /api/v1/viral-video/{job_id}                查询任务状态（含 image_analysis/copy_result 编导脚本）
   GET    /api/v1/viral-video/history                 历史记录
   POST   /api/v1/viral-video/{job_id}/retry          重试失败任务
   POST   /api/v1/viral-video/{job_id}/analyze-style  触发风格分析
@@ -56,27 +56,55 @@ router = APIRouter()
 
 
 def _build_copy_result(job) -> dict | None:
-    """将后端原始字段拼装为前端期望的 CopyResult 结构（final_copy/suggested_copy/title/scenes）。"""
+    """v1.6: 返回编导分镜脚本 CopyResult 结构（给前端/Seedance 使用）。
+
+    - 若 job.copy_result 已持久化（v1.6 worker 生成），直接返回（补 final_copy 兜底）。
+    - 否则从老字段（generated_copy_text=口播, storyboard=分镜列表, intent_result）拼装兼容结构。
+    """
+    cr = getattr(job, "copy_result", None)
+    if isinstance(cr, dict) and cr:
+        out = dict(cr)
+        # 向后兼容字段
+        voiceover = out.get("voiceover_script", "") or ""
+        out.setdefault("final_copy", voiceover)
+        out.setdefault("suggested_copy", voiceover)
+        out.setdefault("title", "")
+        return out
+    # 兼容 v1.5 老数据：storyboard 是老格式 [{order,type,description,text,duration,...}]
     copy_text = getattr(job, "generated_copy_text", "") or ""
     sb = getattr(job, "storyboard", None) or []
     intent = getattr(job, "intent_result", None) or {}
     if not copy_text and not sb:
         return None
-    scenes = []
+    title = ""
+    if isinstance(intent, dict):
+        title = intent.get("suggested_title") or intent.get("intent", "") or ""
+    shots = []
     for seg in sb:
         if isinstance(seg, dict):
-            scenes.append(
+            shots.append(
                 {
-                    "shot": seg.get("description", ""),
-                    "narration": seg.get("text", ""),
-                    "duration": seg.get("duration"),
+                    "time_range": "",
+                    "shot_type_angle_movement": seg.get("ken_burns", ""),
+                    "scene_and_dialogue": (seg.get("text") or "")
+                    + (" " + seg.get("description", "") if seg.get("description") else ""),
+                    "action_details": "",
+                    "audio_bgm": "",
+                    "transition": seg.get("transition", "硬切"),
+                    "reference_image_index": None,
                 }
             )
+    ratio = getattr(job, "video_ratio", None) or "9:16"
     return {
-        "title": (intent.get("suggested_title") if isinstance(intent, dict) else None) or "",
+        "overview": {"theme": title, "total_duration": getattr(job, "duration", 15), "aspect_ratio": ratio},
+        "scene_and_lighting": "",
+        "shots": shots,
+        "hard_constraints": ["无字幕", "无水印", "人物一致性"],
+        "negative_prompts": ["字幕", "水印", "错误文字", "五官崩坏"],
+        "voiceover_script": copy_text,
         "final_copy": copy_text,
         "suggested_copy": copy_text,
-        "scenes": scenes,
+        "title": title,
     }
 
 
@@ -91,7 +119,7 @@ def _to_response(job) -> ViralVideoJobResponse:
         viral_structure=job.viral_structure,
         marketing_purpose=job.marketing_purpose,
         bgm_preference=job.bgm_preference,
-        duration=job.duration,
+        duration=job.duration or 15,
         user_copy_text=job.user_copy_text,
         fusion_level=job.fusion_level,
         reference_audio_path=job.reference_audio_path,
@@ -104,6 +132,10 @@ def _to_response(job) -> ViralVideoJobResponse:
         storyboard=getattr(job, "storyboard", None),
         generated_copy_text=getattr(job, "generated_copy_text", "") or "",
         copy_result=_build_copy_result(job),
+        voice_id=getattr(job, "voice_id", "") or "",
+        voice_source=getattr(job, "voice_source", "") or "",
+        video_ratio=getattr(job, "video_ratio", "9:16") or "9:16",
+        video_model=getattr(job, "video_model", "") or "",
         intent_result=job.intent_result,
         result_video_url=job.result_video_url,
         credits_cost=job.credits_cost,
@@ -148,13 +180,18 @@ def create_viral_video(
         viral_structure=request.viral_structure,
         marketing_purpose=request.marketing_purpose,
         bgm_preference=request.bgm_preference,
-        duration=request.duration,
+        duration=request.duration or 15,
         user_copy_text=request.user_copy_text,
         fusion_level=request.fusion_level,
         reference_audio_path=request.reference_audio_path,
         reference_video_url=request.reference_video_url,
         style_strength=request.style_strength,
         style_template_id=request.style_template_id,
+        voice_id=getattr(request, "voice_id", "") or "",
+        voice_source=getattr(request, "voice_source", "") or "",
+        video_ratio=getattr(request, "video_ratio", "9:16") or "9:16",
+        video_model=getattr(request, "video_model", "") or "",
+        copy_result=None,
     )
 
     # 持久化
@@ -192,6 +229,11 @@ def analyze_images(
         reference_video_url=request.reference_video_url or "",
         style_template_id=request.style_template_id or "",
         style_strength=request.style_strength or "medium",
+        voice_id=request.voice_id or "",
+        voice_source=request.voice_source or "",
+        video_ratio=request.video_ratio or "9:16",
+        video_model=request.video_model or "",
+        duration=request.duration or 15,
     )
     repo.save(job)
 
@@ -213,10 +255,11 @@ def generate_copy(
     authenticated_user: AuthenticatedUser = Depends(get_current_user),
     session: Session = Depends(get_db_session),
 ) -> ViralVideoJobResponse:
-    """v1.5 阶段2：用户填完营销参数后，跑 意图解析 → 文案融合 → 分镜 → 合规审核。
+    """v1.6 阶段2：用户填完营销参数后，跑 意图解析 → 编导分镜脚本生成 → 合规审核。
 
-    跑完后状态=copy_generated，响应包含 generated_copy_text + storyboard，
-    前端展示文案供用户编辑；确认/编辑后调 /{id}/confirm-copy 进入阶段3。
+    跑完后状态=copy_generated，响应 copy_result（含 overview/scene_and_lighting/shots/
+    hard_constraints/negative_prompts/voiceover_script），前端展示脚本与口播供用户编辑；
+    确认/编辑后调 /{id}/confirm-copy 进入阶段3（TTS + 单次 Seedance 出片）。
     """
     repo = _get_job_repo(session)
     job = repo.get(job_id)
@@ -240,13 +283,19 @@ def generate_copy(
     job.marketing_purpose = request.marketing_purpose or job.marketing_purpose
     job.bgm_preference = request.bgm_preference or job.bgm_preference
     if request.duration:
-        job.duration = request.duration
+        job.duration = max(5, min(30, int(request.duration)))
     job.user_copy_text = request.user_copy_text if request.user_copy_text else job.user_copy_text
     job.fusion_level = request.fusion_level or job.fusion_level
     job.reference_audio_path = request.reference_audio_path or job.reference_audio_path
     job.reference_video_url = request.reference_video_url or job.reference_video_url
     job.style_strength = request.style_strength or job.style_strength
     job.style_template_id = request.style_template_id or job.style_template_id
+    if request.style_guide is not None:
+        job.style_guide = request.style_guide
+    job.voice_id = request.voice_id or job.voice_id
+    job.voice_source = request.voice_source or job.voice_source
+    job.video_ratio = request.video_ratio or job.video_ratio or "9:16"
+    job.video_model = request.video_model or job.video_model or ""
 
     job.resume_from_image_analyzed()
     repo.update(job)
@@ -269,7 +318,7 @@ def confirm_copy(
     authenticated_user: AuthenticatedUser = Depends(get_current_user),
     session: Session = Depends(get_db_session),
 ) -> ViralVideoJobResponse:
-    """v1.5 阶段3：用户确认/编辑文案后开始 TTS+BGM(skip)+渲染+上传。"""
+    """v1.6 阶段3：用户确认/编辑口播后开始 TTS + 单次 Seedance 生成 + 上传。"""
     repo = _get_job_repo(session)
     job = repo.get(job_id)
     if job is None:

apps/api/app/schemas/viral_video.py

@@ -1,4 +1,4 @@
-"""爆款视频 API schemas。"""
+"""爆款视频 API schemas (v1.6 单次 Seedance 出片版)。"""
 
 from __future__ import annotations
 
@@ -14,41 +14,84 @@ VALID_STAGES = (
     "image_analysis",
     "video_analysis",
     "intent_parsing",
-    "copy_fusion",
-    "storyboard",
+    "script_generation",
     "review",
     "tts",
-    "bgm_select",
     "rendering",
-    "musetalk",
     "uploading",
 )
+VALID_VIDEO_RATIOS = ("9:16", "16:9", "1:1", "4:3", "3:4", "21:9")
+VALID_DURATIONS = (5, 10, 15, 20, 25, 30)
+
+
+# -- 编导脚本结构（v1.6） --
+
+
+class ShotScript(BaseModel):
+    """逐镜头分镜。"""
+
+    time_range: str = Field(default="", description="时间区间，如 0-3秒")
+    shot_type_angle_movement: str = Field(default="", description="景别/角度/运镜，如『近景俯拍45度，缓慢推镜』")
+    scene_and_dialogue: str = Field(default="", description="场景描述+口播台词")
+    action_details: str = Field(default="", description="人物动作、表情、物品操作细节")
+    audio_bgm: str = Field(default="", description="环境音+BGM提示")
+    transition: str = Field(default="硬切", description="转场方式：硬切/淡入淡出/叠化")
+    reference_image_index: int | None = Field(
+        default=None, description="参考图片索引（0-based，对应上传的第几张产品图）"
+    )
+
+
+class CopyResultOverview(BaseModel):
+    theme: str = ""
+    total_duration: int = 15
+    aspect_ratio: str = "9:16"
+
+
+class CopyResult(BaseModel):
+    """v1.6 编导分镜脚本结构（给前端 + Seedance 用）。"""
+
+    overview: CopyResultOverview = Field(default_factory=CopyResultOverview)
+    scene_and_lighting: str = ""
+    shots: list[ShotScript] = Field(default_factory=list)
+    hard_constraints: list[str] = Field(default_factory=list)
+    negative_prompts: list[str] = Field(default_factory=list)
+    voiceover_script: str = Field(
+        default="", description="纯口播对白，从各镜 scene_and_dialogue 的对白部分拼接，供 TTS 使用"
+    )
+    # 向后兼容：final_copy = voiceover_script
+    final_copy: str = ""
+    suggested_copy: str = ""
+    title: str = ""
 
 
 # -- Request Schemas --
 
 
 class CreateViralVideoRequest(BaseModel):
-    """创建爆款视频任务请求（旧接口：一键跑完前半段到 WAIT_USER_CONFIRM，保留兼容）。"""
+    """旧接口：一键创建（保留兼容）。"""
 
-    images: list[str] = Field(..., min_length=1, max_length=20, description="产品图片 URL 列表")
-    industry: str = Field(default="", description="行业")
-    target_customer: str = Field(default="", description="目标客户描述")
-    persona_id: str = Field(default="", description="人设 ID")
-    viral_structure: str = Field(default="", description="爆款结构类型")
-    marketing_purpose: str = Field(default="", description="营销目的")
-    bgm_preference: str = Field(default="", description="BGM 偏好")
-    duration: int = Field(default=30, ge=5, le=180, description="视频时长（秒）")
-    user_copy_text: str = Field(default="", description="用户原始文案（我说你写）")
-    fusion_level: str = Field(default="ai_polish", description="文案融合级别: ai_full/ai_polish/user_primary")
-    reference_audio_path: str = Field(default="", description="参考音频路径")
-    reference_video_url: str = Field(default="", description="参考爆款视频 URL")
-    style_strength: str = Field(default="medium", description="风格强度: light/medium/strict")
-    style_template_id: str = Field(default="", description="风格模板 ID")
+    images: list[str] = Field(..., min_length=1, max_length=20)
+    industry: str = ""
+    target_customer: str = ""
+    persona_id: str = ""
+    viral_structure: str = ""
+    marketing_purpose: str = ""
+    bgm_preference: str = ""
+    duration: int = Field(default=15, ge=5, le=30, description="视频时长（秒），5-30")
+    user_copy_text: str = ""
+    fusion_level: str = "ai_polish"
+    reference_audio_path: str = ""
+    reference_video_url: str = ""
+    style_strength: str = "medium"
+    style_template_id: str = ""
+    voice_id: str = ""
+    voice_source: str = ""
+    video_ratio: str = "9:16"
+    video_model: str = ""
 
     @field_validator("fusion_level")
     @classmethod
-    def _validate_fusion_level(cls, v: str) -> str:
+    def _v_fl(cls, v: str) -> str:
         if v == "full_ai":
             return "ai_full"
         if v not in VALID_FUSION_LEVELS:
@@ -57,37 +100,47 @@ class CreateViralVideoRequest(BaseModel):
 
     @field_validator("style_strength")
     @classmethod
-    def _validate_style_strength(cls, v: str) -> str:
+    def _v_ss(cls, v: str) -> str:
         if v not in VALID_STYLE_STRENGTHS:
             raise ValueError(f"style_strength must be one of {VALID_STYLE_STRENGTHS}")
         return v
 
 
 class AnalyzeImagesRequest(BaseModel):
-    """v1.5 阶段1：创建任务并仅做图片/视频分析。images 必填，其他参数可选（阶段2再传）。"""
+    """v1.5+ 阶段1：创建任务 + 图片/视频分析。"""
 
-    images: list[str] = Field(..., min_length=1, max_length=20)
-    reference_video_url: str = Field(default="", description="参考爆款视频 URL（可选，有则同步做风格分析）")
-    style_template_id: str = Field(default="", description="风格模板 ID（可选）")
-    style_strength: str = Field(default="medium")
+    images: list[str] = Field(..., min_length=1, max_length=30)
+    reference_video_url: str = ""
+    style_template_id: str = ""
+    style_strength: str = "medium"
+    voice_id: str = ""
+    voice_source: str = ""
+    video_ratio: str = "9:16"
+    video_model: str = ""
+    duration: int = Field(default=15, ge=5, le=30)
 
 
 class GenerateCopyRequest(BaseModel):
-    """v1.5 阶段2：用户填完参数后跑意图+文案+分镜+审核，暂停在 COPY_GENERATED。"""
+    """v1.5+ 阶段2：填完营销参数，生成编导脚本。"""
 
-    industry: str = Field(default="")
-    target_customer: str = Field(default="")
-    persona_id: str = Field(default="")
-    viral_structure: str = Field(default="")
-    marketing_purpose: str = Field(default="")
-    bgm_preference: str = Field(default="")
-    duration: int = Field(default=30, ge=5, le=180)
-    user_copy_text: str = Field(default="")
-    fusion_level: str = Field(default="ai_polish")
-    reference_audio_path: str = Field(default="")
-    reference_video_url: str = Field(default="")
-    style_strength: str = Field(default="medium")
-    style_template_id: str = Field(default="")
+    industry: str = ""
+    target_customer: str = ""
+    persona_id: str = ""
+    viral_structure: str = ""
+    marketing_purpose: str = ""
+    bgm_preference: str = ""
+    duration: int = Field(default=15, ge=5, le=30)
+    user_copy_text: str = ""
+    fusion_level: str = "ai_polish"
+    reference_audio_path: str = ""
+    reference_video_url: str = ""
+    style_strength: str = "medium"
+    style_template_id: str = ""
+    style_guide: dict | None = None
+    voice_id: str = ""
+    voice_source: str = ""
+    video_ratio: str = "9:16"
+    video_model: str = ""
 
     @field_validator("fusion_level")
     @classmethod
@@ -107,30 +160,28 @@ class GenerateCopyRequest(BaseModel):
 
 
 class ConfirmCopyRequest(BaseModel):
-    """v1.5 阶段3：用户确认/编辑文案后开始渲染。"""
+    """v1.5+ 阶段3：用户确认/编辑口播后开始渲染（TTS+单次Seedance）。"""
 
-    edited_copy: str = Field(default="", description="用户编辑后的最终文案；为空则使用 AI 生成文案")
+    edited_copy: str = Field(default="", description="用户编辑后的口播文案；为空则用 AI 生成的 voiceover_script")
 
 
 class ConfirmIntentRequest(BaseModel):
-    """确认意图请求（旧 confirm-intent，兼容）。"""
+    """旧 confirm-intent（兼容）。"""
 
-    confirmed_copy: str = Field(default="", description="用户确认/修改后的文案")
-    adjustments: str = Field(default="", description="用户对 AI 文案的调整意见")
+    confirmed_copy: str = ""
+    adjustments: str = ""
 
 
 class AnalyzeStyleRequest(BaseModel):
-    """触发参考视频风格分析请求。"""
-
     reference_video_url: str = Field(..., description="参考视频 URL")
-    style_template_id: str = Field(default="", description="风格模板 ID（可选覆盖）")
+    style_template_id: str = ""
 
 
 # -- Response Schemas --
 
 
 class ViralVideoJobResponse(BaseModel):
-    """爆款视频任务响应。v1.5 新增 image_analysis/storyboard/generated_copy_text 字段。"""
+    """爆款视频任务响应（v1.6 包含 copy_result 编导脚本结构）。"""
 
     id: str
     user_id: str
@@ -141,7 +192,7 @@ class ViralVideoJobResponse(BaseModel):
     viral_structure: str = ""
     marketing_purpose: str = ""
     bgm_preference: str = ""
-    duration: int = 30
+    duration: int = 15
     user_copy_text: str = ""
     fusion_level: str = "ai_polish"
     reference_audio_path: str = ""
@@ -150,13 +201,17 @@ class ViralVideoJobResponse(BaseModel):
     style_guide: dict | None = None
     style_template_id: str = ""
     status: str
-    # v1.4 VLM 结果
     image_analysis: dict | None = None
-    # v1.5 三步分步产物（原始字段，保留给后端/老调用方）
+    # v1.6 编导脚本（推荐前端使用）
+    copy_result: dict | None = None
+    # v1.5 兼容字段
     storyboard: list | None = None
     generated_copy_text: str = ""
-    # v1.5 前端 CopyResult 结构（final_copy/suggested_copy/title/scenes）
-    copy_result: dict | None = None
+    # 音色/视频参数
+    voice_id: str = ""
+    voice_source: str = ""
+    video_ratio: str = "9:16"
+    video_model: str = ""
     intent_result: dict | None = None
     result_video_url: str = ""
     credits_cost: int = 0

apps/web/src/api/viral-video/types.ts

@@ -12,6 +12,14 @@ export const STYLE_STRENGTHS: { value: StyleStrength; label: string }[] = [
   { value: "strict", label: "像素级复刻" },
 ]
 
+/** v1.6 前端时长下拉选项（5/10/15/20/25/30秒） */
+export const VALID_DURATIONS = [5, 10, 15, 20, 25, 30] as const
+export type VideoDuration = (typeof VALID_DURATIONS)[number]
+
+/** v1.6 支持的画幅比例 */
+export const VALID_RATIOS = ["9:16", "16:9", "1:1"] as const
+export type VideoRatio = (typeof VALID_RATIOS)[number]
+
 export type ViralVideoStatus =
   | "pending"
   | "running"
@@ -23,39 +31,25 @@ export type ViralVideoStatus =
   | "cancelled"
 
 /**
- * 后端流水线阶段字符串。前端不展示逐阶段进度列表，仅保留类型
- * 用于轮询时判断当前在哪个大阶段（分析中 vs 文案生成 vs 视频生成）。
+ * v1.6 后端流水线阶段。单次 Seedance 出片版：
+ * image_analysis → video_analysis(可选) → intent_parsing → script_generation → review → tts → rendering → uploading
  */
 export type ViralVideoStage =
   | "image_analysis"
   | "video_analysis"
   | "intent_parsing"
-  | "copy_fusion"
-  | "storyboard"
+  | "script_generation"
   | "review"
   | "tts"
-  | "bgm_select"
   | "rendering"
-  | "musetalk"
   | "uploading"
 
 /** 图片+视频分析阶段：属于「分析图片」按钮的范围 */
 const IMAGE_ANALYSIS_STAGES = new Set<ViralVideoStage>(["image_analysis", "video_analysis"])
-/** 文案相关阶段：属于「生成文案」按钮的范围 */
-const COPY_STAGES = new Set<ViralVideoStage>([
-  "intent_parsing",
-  "copy_fusion",
-  "storyboard",
-  "review",
-])
-/** 视频相关阶段：属于「开始生成视频」按钮的范围 */
-const VIDEO_STAGES = new Set<ViralVideoStage>([
-  "tts",
-  "bgm_select",
-  "rendering",
-  "musetalk",
-  "uploading",
-])
+/** 编导脚本阶段：属于「生成文案」按钮的范围 */
+const COPY_STAGES = new Set<ViralVideoStage>(["intent_parsing", "script_generation", "review"])
+/** 视频生成阶段：属于「开始生成视频」按钮的范围（v1.6: TTS+单次Seedance+上传） */
+const VIDEO_STAGES = new Set<ViralVideoStage>(["tts", "rendering", "uploading"])
 
 export function isImageAnalysisStage(stage: ViralVideoStage | undefined): boolean {
   return !!stage && IMAGE_ANALYSIS_STAGES.has(stage)
@@ -66,7 +60,7 @@ export function isCopyStage(stage: ViralVideoStage | undefined): boolean {
 export function isVideoStage(stage: ViralVideoStage | undefined): boolean {
   return !!stage && VIDEO_STAGES.has(stage)
 }
-/** 兼容旧调用：旧的 isAnalysisStage 视为「图片分析+文案」的所有前置阶段 */
+/** 兼容旧调用：分析图片+生成文案 的所有前置阶段 */
 export function isAnalysisStage(stage: ViralVideoStage | undefined): boolean {
   return isImageAnalysisStage(stage) || isCopyStage(stage)
 }
@@ -74,12 +68,20 @@ export function isAnalysisStage(stage: ViralVideoStage | undefined): boolean {
 /** 单张图片 VLM 识别出的商品信息 */
 export interface ImageProductAnalysis {
   name?: string
-  spec?: string
+  category?: string
   brand?: string
+  colors?: string[]
+  material_or_texture?: string
+  key_features?: string[]
+  visual_style?: string
+  scene?: string
+  target_audience_hint?: string
+  text_on_image?: string
+  /** 旧字段兼容 */
+  spec?: string
   features?: string[] | string
   label_text?: string
-  selling_points?: string[]
-  scene?: string
+  selling_points?: string
   image_index?: number
 }
 
@@ -87,10 +89,50 @@ export interface ImageAnalysisResult {
   products?: ImageProductAnalysis[]
 }
 
+/** v1.6 编导分镜脚本 - 单镜头 */
+export interface ShotScript {
+  /** 时间区间，如 "0-3秒" */
+  time_range?: string
+  /** 景别/角度/运镜，如 "近景俯拍45度，缓慢推镜" */
+  shot_type_angle_movement?: string
+  /** 场景描述+对白 */
+  scene_and_dialogue?: string
+  /** 人物动作/表情/物品操作细节 */
+  action_details?: string
+  /** 环境音+BGM提示 */
+  audio_bgm?: string
+  /** 转场方式（硬切/淡入淡出/叠化/结束） */
+  transition?: string
+  /** 参考图片索引（0-based，对应上传产品图数组） */
+  reference_image_index?: number | null
+}
+
+/** v1.6 编导分镜脚本 - 总览 */
+export interface CopyResultOverview {
+  theme?: string
+  total_duration?: number
+  aspect_ratio?: string
+}
+
+/** v1.6 编导分镜脚本（核心输出结构，给 Seedance 做 prompt，给 TTS 取 voiceover_script） */
 export interface CopyResult {
+  overview?: CopyResultOverview
+  /** 整体场景+光线描述 */
+  scene_and_lighting?: string
+  /** 逐镜头时间轴 */
+  shots?: ShotScript[]
+  /** 硬性约束（禁止字幕/水印/变形等） */
+  hard_constraints?: string[]
+  /** 负面提示词 */
+  negative_prompts?: string[]
+  /** 完整口播稿（纯文本，用于 TTS 合成） */
+  voiceover_script?: string
+  /** 向后兼容：= voiceover_script */
   final_copy?: string
+  /** 向后兼容：= voiceover_script */
   suggested_copy?: string
   title?: string
+  /** v1.5 旧字段兼容（老数据降级时可能出现） */
   scenes?: Array<{ shot: string; narration: string; duration?: number }>
 }
 
@@ -104,13 +146,18 @@ export interface StyleTemplate {
 }
 
 export interface IntentResult {
-  product: string
-  selling_points: string[]
-  target_audience: string
-  tone: string
-  structure: string
-  duration: number
+  intent?: string
+  key_messages?: string[]
+  tone?: string
+  target_emotion?: string
+  call_to_action?: string
   suggested_title?: string
+  /** v1.5 旧字段兼容 */
+  product?: string
+  selling_points?: string[]
+  target_audience?: string
+  structure?: string
+  duration?: number
   suggested_copy?: string
 }
 
@@ -123,7 +170,9 @@ export interface ViralVideoJob {
   style_template_id?: string
   style_guide?: string | Record<string, unknown>
   user_copy_text?: string
+  /** v1.6: = copy_result.voiceover_script（从 copy_result 派生，向后兼容） */
   final_copy_text?: string
+  generated_copy_text?: string
   fusion_level?: FusionLevel
   voice_id?: string
   voice_mode?: "global" | "per_video"
@@ -131,8 +180,17 @@ export interface ViralVideoJob {
   bgm_preference?: string
   intent_result?: IntentResult
   intent_text?: string
+  /** v1.6 编导分镜脚本（核心产物） */
   copy_result?: CopyResult
+  /** 向后兼容：= copy_result.shots */
+  storyboard?: ShotScript[]
   image_analysis?: ImageAnalysisResult
+  /** 视频比例：9:16 / 16:9 / 1:1，默认 9:16 */
+  video_ratio?: string
+  /** Seedance 模型 ID（空=后端默认） */
+  video_model?: string
+  /** 视频时长（秒，5-30，默认15） */
+  duration?: number
   progress_stage?: ViralVideoStage
   progress_percent?: number
   progress_message?: string
@@ -162,6 +220,7 @@ export interface GenerateViralVideoRequest {
   persona_id?: string
   viral_structure?: string
   marketing_purpose?: string
+  /** 视频时长（5-30秒，默认15） */
   duration?: number
   video_model?: string
   video_ratio?: string
@@ -176,15 +235,25 @@ export interface HistoryResponse {
   page_size: number
 }
 
-/** v1.5 阶段1请求：仅做图片/视频分析（POST /viral-video/analyze-images） */
+/** v1.6 阶段1请求：图片/视频分析（POST /viral-video/analyze-images） */
 export interface AnalyzeImagesRequest {
   images: string[]
   reference_video_url?: string
   style_template_id?: string
   style_strength?: StyleStrength
+  /** TTS 音色 ID（STEP1 已选音色时传） */
+  voice_id?: string
+  /** 音色来源：preset | library | clone | upload */
+  voice_source?: "preset" | "library" | "clone" | "upload"
+  /** Seedance 视频比例：9:16 | 16:9 | 1:1 */
+  video_ratio?: string
+  /** Seedance 模型 ID（空则使用服务端默认） */
+  video_model?: string
+  /** 视频时长（秒，5-30，默认15） */
+  duration?: number
 }
 
-/** v1.5 阶段2请求：填完营销参数后生成文案+分镜（POST /viral-video/{id}/generate-copy） */
+/** v1.6 阶段2请求：填完营销参数后生成编导分镜脚本（POST /viral-video/{id}/generate-copy） */
 export interface GenerateCopyRequest {
   industry?: string
   target_customer?: string
@@ -192,6 +261,7 @@ export interface GenerateCopyRequest {
   viral_structure?: string
   marketing_purpose?: string
   bgm_preference?: string
+  /** 视频时长（秒，5-30，默认15） */
   duration?: number
   user_copy_text?: string
   fusion_level?: FusionLevel
@@ -200,15 +270,23 @@ export interface GenerateCopyRequest {
   style_strength?: StyleStrength
   style_template_id?: string
   style_guide?: string | Record<string, unknown>
+  /** TTS 音色 ID（优先级高于 persona_id） */
+  voice_id?: string
+  /** 音色来源：preset | library | clone | upload */
+  voice_source?: "preset" | "library" | "clone" | "upload"
+  /** Seedance 视频比例（9:16/16:9/1:1 等） */
+  video_ratio?: string
+  /** Seedance 模型 ID（空则使用服务端默认） */
+  video_model?: string
 }
 
-/** v1.5 阶段3请求：用户确认/编辑文案后开始渲染（POST /viral-video/{id}/confirm-copy） */
+/** v1.6 阶段3请求：用户确认/编辑口播文案后开始单次 Seedance 出片（POST /viral-video/{id}/confirm-copy） */
 export interface ConfirmCopyRequest {
-  /** 用户编辑后的最终文案；为空则使用 AI 生成文案 */
+  /** 用户编辑后的口播文案；为空则使用 AI 生成的 voiceover_script */
   edited_copy?: string
 }
 
-/** 分镜片段结构（后端 storyboard 字段的元素形态，保留供调试/进阶使用；主流程请使用 copy_result.scenes） */
+/** 旧分镜片段结构（保留兼容；新代码请使用 ShotScript） */
 export interface StoryboardSegment {
   order: number
   type: string

apps/web/src/pages/viral-video/ViralVideoPage.css

@@ -1596,3 +1596,370 @@
     grid-template-columns: repeat(2, 1fr);
   }
 }
+
+/* ── STEP2 2列表单网格 ── */
+.vv-form-grid {
+  display: grid;
+  grid-template-columns: 1fr 1fr;
+  gap: 14px 18px;
+  margin-top: 12px;
+}
+.vv-form-row {
+  display: flex;
+  flex-direction: column;
+  gap: 6px;
+}
+.vv-label {
+  font-size: 13px;
+  font-weight: 500;
+  color: #374151;
+  display: flex;
+  align-items: center;
+  gap: 4px;
+}
+.vv-label-optional {
+  font-size: 12px;
+  color: #9ca3af;
+  font-weight: 400;
+}
+.vv-input {
+  width: 100%;
+  height: 36px;
+  border-radius: 8px;
+  border: 1px solid #e5e7eb;
+  background: #fff;
+  padding: 0 12px;
+  font-size: 13px;
+  color: #111827;
+  outline: none;
+  transition:
+    border-color 0.15s,
+    box-shadow 0.15s;
+}
+.vv-input:focus {
+  border-color: #7c3aed;
+  box-shadow: 0 0 0 2px rgba(124, 58, 237, 0.15);
+}
+.vv-input:disabled {
+  background: #f9fafb;
+  color: #9ca3af;
+  cursor: not-allowed;
+}
+.vv-input::placeholder {
+  color: #9ca3af;
+}
+
+/* 自定义 Select 样式（浅色紫调） */
+.vv-select-purple .ant-select-selector {
+  border-radius: 8px !important;
+  border: 1px solid #e5e7eb !important;
+  background: #fff !important;
+  min-height: 36px !important;
+  padding: 0 28px 0 12px !important;
+  box-shadow: none !important;
+  transition:
+    border-color 0.15s,
+    box-shadow 0.15s !important;
+}
+.vv-select-purple.ant-select-focused .ant-select-selector,
+.vv-select-purple .ant-select-focused .ant-select-selector {
+  border-color: #7c3aed !important;
+  box-shadow: 0 0 0 2px rgba(124, 58, 237, 0.15) !important;
+}
+.vv-select-purple .ant-select-selection-item,
+.vv-select-purple .ant-select-selection-placeholder {
+  line-height: 34px !important;
+  font-size: 13px !important;
+  color: #111827 !important;
+}
+.vv-select-purple .ant-select-selection-placeholder {
+  color: #9ca3af !important;
+}
+.vv-select-purple .ant-select-arrow {
+  color: #6b7280;
+  right: 10px !important;
+}
+.vv-select-purple.ant-select-disabled .ant-select-selector {
+  background: #f9fafb !important;
+  color: #9ca3af !important;
+  cursor: not-allowed;
+}
+
+/* ghost/outline 按钮 */
+.vv-btn-ghost {
+  background: #fff;
+  color: #7c3aed;
+  border: 1px solid #d6c6fb;
+}
+.vv-btn-ghost:hover {
+  background: #f8f4ff;
+  border-color: #7c3aed;
+}
+
+@media (max-width: 640px) {
+  .vv-form-grid {
+    grid-template-columns: 1fr;
+  }
+}
+
+/* ── 分镜脚本（Storyboard）样式 ── */
+.vv-storyboard {
+  display: flex;
+  flex-direction: column;
+  gap: 14px;
+}
+
+.vv-copy-tag {
+  display: inline-flex;
+  align-items: center;
+  font-size: 11px;
+  padding: 2px 8px;
+  border-radius: 10px;
+  margin-left: 8px;
+}
+.vv-copy-tag-preview {
+  background: #fef3c7;
+  color: #92400e;
+}
+
+/* 总览卡片 */
+.vv-sb-overview {
+  display: grid;
+  grid-template-columns: 2fr 1fr 1fr;
+  gap: 12px;
+  padding: 12px;
+  background: linear-gradient(135deg, #f5f3ff 0%, #ede9fe 100%);
+  border-radius: 10px;
+  border: 1px solid #e9e4ff;
+}
+.vv-sb-ov-field label {
+  display: block;
+  font-size: 11px;
+  color: #6b7280;
+  margin-bottom: 4px;
+}
+.vv-sb-input {
+  height: 32px;
+  font-size: 13px;
+}
+
+/* 分块标题 */
+.vv-sb-section {
+  display: flex;
+  flex-direction: column;
+  gap: 8px;
+}
+.vv-sb-section-title {
+  display: flex;
+  align-items: center;
+  gap: 6px;
+  font-size: 13px;
+  font-weight: 600;
+  color: #374151;
+}
+
+/* 镜头卡片 */
+.vv-sb-shots {
+  display: flex;
+  flex-direction: column;
+  gap: 10px;
+}
+.vv-sb-shot {
+  background: #f9fafb;
+  border: 1px solid #e5e7eb;
+  border-radius: 10px;
+  padding: 12px;
+}
+.vv-sb-shot-head {
+  display: flex;
+  align-items: center;
+  gap: 10px;
+  margin-bottom: 10px;
+}
+.vv-sb-shot-num {
+  background: #7c3aed;
+  color: #fff;
+  font-size: 12px;
+  font-weight: 600;
+  padding: 2px 10px;
+  border-radius: 10px;
+}
+.vv-sb-time {
+  height: 28px;
+  border: 1px solid #e5e7eb;
+  background: #fff;
+  border-radius: 6px;
+  padding: 0 10px;
+  font-size: 12px;
+  color: #6b7280;
+  width: 100px;
+  outline: none;
+}
+.vv-sb-time:focus {
+  border-color: #7c3aed;
+}
+.vv-sb-shot-grid {
+  display: grid;
+  grid-template-columns: 1fr 1fr;
+  gap: 10px 14px;
+}
+.vv-sb-cell {
+  display: flex;
+  flex-direction: column;
+  gap: 4px;
+}
+.vv-sb-cell-wide {
+  grid-column: 1 / -1;
+}
+.vv-sb-cell label {
+  font-size: 11px;
+  color: #6b7280;
+  font-weight: 500;
+}
+.vv-sb-cell-ta {
+  min-height: 54px !important;
+  font-size: 12.5px !important;
+  padding: 7px 10px !important;
+  line-height: 1.55 !important;
+  resize: vertical;
+}
+.vv-sb-ref-box {
+  width: 80px;
+  height: 54px;
+  border-radius: 8px;
+  overflow: hidden;
+  background: #fff;
+  border: 1px dashed #d1d5db;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+}
+.vv-sb-ref-img {
+  width: 100%;
+  height: 100%;
+  object-fit: cover;
+}
+.vv-sb-ref-empty {
+  font-size: 11px;
+  color: #9ca3af;
+}
+
+/* 标签（硬性约束/负面提示词） */
+.vv-sb-taglist {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 6px;
+  align-items: center;
+}
+.vv-sb-tag {
+  display: inline-flex;
+  align-items: center;
+  gap: 4px;
+  padding: 2px 4px 2px 10px;
+  border-radius: 14px;
+  font-size: 12px;
+}
+.vv-sb-tag-hard {
+  background: #f3f0ff;
+  color: #6d28d9;
+  border: 1px solid #ddd6fe;
+}
+.vv-sb-tag-neg {
+  background: #fef2f2;
+  color: #b91c1c;
+  border: 1px solid #fecaca;
+}
+.vv-sb-tag-input {
+  background: transparent;
+  border: none;
+  outline: none;
+  font-size: 12px;
+  color: inherit;
+  min-width: 80px;
+  padding: 2px 0;
+}
+.vv-sb-tag-hard .vv-sb-tag-input {
+  color: #6d28d9;
+}
+.vv-sb-tag-neg .vv-sb-tag-input {
+  color: #b91c1c;
+}
+.vv-sb-tag-del {
+  width: 18px;
+  height: 18px;
+  border-radius: 50%;
+  border: none;
+  background: rgba(0, 0, 0, 0.08);
+  color: inherit;
+  cursor: pointer;
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  font-size: 12px;
+  line-height: 1;
+}
+.vv-sb-tag-del:hover {
+  background: rgba(0, 0, 0, 0.18);
+}
+.vv-sb-tag-add {
+  border: 1px dashed #d1d5db;
+  background: transparent;
+  color: #6b7280;
+  font-size: 12px;
+  padding: 3px 10px;
+  border-radius: 14px;
+  cursor: pointer;
+}
+.vv-sb-tag-add:hover {
+  border-color: #7c3aed;
+  color: #7c3aed;
+}
+
+/* 口播稿折叠 */
+.vv-sb-collapse {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  width: 100%;
+  background: #fff;
+  border: 1px solid #e5e7eb;
+  border-radius: 8px;
+  padding: 10px 14px;
+  cursor: pointer;
+  font-size: 13px;
+  color: #374151;
+  font-weight: 500;
+  transition:
+    border-color 0.15s,
+    background 0.15s;
+}
+.vv-sb-collapse:hover {
+  border-color: #7c3aed;
+  background: #faf8ff;
+}
+.vv-sb-caret {
+  transition: transform 0.2s;
+  color: #9ca3af;
+  font-size: 12px;
+}
+.vv-sb-caret.open {
+  transform: rotate(180deg);
+  color: #7c3aed;
+}
+.vv-sb-vo {
+  min-height: 100px !important;
+  font-size: 13px !important;
+  line-height: 1.7 !important;
+}
+
+@media (max-width: 640px) {
+  .vv-sb-overview {
+    grid-template-columns: 1fr;
+  }
+  .vv-sb-shot-grid {
+    grid-template-columns: 1fr;
+  }
+  .vv-sb-cell-wide {
+    grid-column: auto;
+  }
+}

apps/web/src/pages/viral-video/ViralVideoPage.tsx

@@ -9,6 +9,8 @@ import {
   CaretRightOutlined,
   VideoCameraOutlined,
   SoundOutlined,
+  BulbOutlined,
+  ThunderboltOutlined,
   ReloadOutlined,
   WarningFilled,
   SearchOutlined,
@@ -24,6 +26,7 @@ import {
 } from "@ant-design/icons"
 import { Select, Input, message } from "antd"
 import { uploadAssetDirect, getAssetLibraries, getAssetsByKind, type AssetItem } from "@/api/assets"
+import { fetchPresetVoices } from "@/api/voices"
 import {
   FUSION_LEVELS,
   STYLE_STRENGTHS,
@@ -35,6 +38,8 @@ import {
   type ViralVideoJob,
   type ImageProductAnalysis,
   type ImageAnalysisResult,
+  type CopyResult,
+  type ShotScript,
 } from "@/api/viral-video/types"
 import {
   generateViralVideo,
@@ -109,19 +114,19 @@ type TabTask = {
   userCopy: string
   industry: string
   targetCustomer: string
+  persona: string
   language: string
   viralStructure: string
   marketingPurpose: string
   duration: number
-  bgm: string
   // STEP3 设置
   videoRatio: string
   videoModel: string
   // UI 状态
   uiStep: UIStep
   imageAnalysis: ImageAnalysisResult | null
-  generatedCopy: string // 文案结果（可编辑）
-  generatedTitle: string
+  // ── 分镜脚本（新结构） ──
+  storyboard: Storyboard | null
   voiceTipsOpen: boolean
   copyError?: string
   videoError?: string
@@ -134,17 +139,18 @@ type TabTask = {
 
 /* ─────────── 常量 ─────────── */
 
-const INDUSTRIES = [
-  "美妆个护",
-  "服饰鞋包",
-  "数码3C",
-  "食品饮料",
-  "家居日用",
-  "母婴亲子",
-  "教育培训",
-  "其他",
-]
 const LANGUAGES = ["中文(普通话)", "粤语", "英语", "日语", "韩语"]
+const TARGET_CUSTOMERS = [
+  "小镇青年",
+  "小镇中老年",
+  "Z世代",
+  "精致妈妈",
+  "新锐白领",
+  "资深中产",
+  "都市蓝领",
+  "都市银发",
+]
+const PERSONAS = ["亲切邻家姐姐", "专业顾问", "搞笑段子手", "真实测评博主", "闺蜜种草", "行业专家"]
 const STRUCTURES = [
   "痛点开场→卖点→促单",
   "反差对比→证据→促单",
@@ -152,7 +158,7 @@ const STRUCTURES = [
   "专家背书→数据→促单",
 ]
 const PURPOSES = ["品牌种草", "新品上市", "促销转化", "活动引流"]
-const BGMS = ["无BGM", "欢快活泼", "温柔治愈", "励志燃向", "搞笑魔性", "科技感"]
+const DURATIONS = [15, 20, 30, 45, 60]
 const RATIOS = [
   { v: "9:16", label: "9:16 竖屏（抖音/视频号）" },
   { v: "16:9", label: "16:9 横屏（B站/YouTube）" },
@@ -165,6 +171,141 @@ const MODELS = [
 const VOICE_TIPS =
   "支持 MP3/WAV/M4A/AAC/OGG 格式，最大 10MB，时长 ≤30 秒。建议清晰人声、无背景音乐、环境安静；录音请保持距麦克风 15-20cm，音量适中。"
 
+// ── 分镜脚本数据模型（新后端 copy_result 结构，前端先 mock 展示） ──
+interface StoryboardShot {
+  time_range: string
+  shot_type_angle_movement: string
+  scene_and_dialogue: string
+  action_details: string
+  audio_bgm: string
+  transition: string
+  reference_image_index?: number
+}
+interface Storyboard {
+  overview: { theme: string; total_duration: string; aspect_ratio: string }
+  scene_and_lighting: string
+  shots: StoryboardShot[]
+  hard_constraints: string[]
+  negative_prompts: string[]
+  voiceover_script: string
+}
+
+/** 兼容旧 copy_result（final_copy/title/scenes）→ 新 Storyboard 结构 */
+function copyResultToStoryboard(cr: CopyResult | null | undefined): Storyboard | null {
+  if (!cr) return null
+  // 新结构：含 shots/overview 字段直接用
+  if (Array.isArray(cr.shots) && cr.overview) {
+    return {
+      overview: {
+        theme: cr.overview.theme || cr.title || "",
+        total_duration: cr.overview.total_duration ? `${cr.overview.total_duration}s` : "",
+        aspect_ratio: cr.overview.aspect_ratio || "9:16",
+      },
+      scene_and_lighting: cr.scene_and_lighting || "",
+      shots: ((cr.shots ?? []) as ShotScript[]).map((s) => ({
+        time_range: s.time_range || "",
+        shot_type_angle_movement: s.shot_type_angle_movement || "",
+        scene_and_dialogue: s.scene_and_dialogue || "",
+        action_details: s.action_details || "",
+        audio_bgm: s.audio_bgm || "",
+        transition: s.transition || "",
+        reference_image_index:
+          typeof s.reference_image_index === "number" ? s.reference_image_index : undefined,
+      })),
+      hard_constraints: Array.isArray(cr.hard_constraints) ? cr.hard_constraints : [],
+      negative_prompts: Array.isArray(cr.negative_prompts) ? cr.negative_prompts : [],
+      voiceover_script: cr.voiceover_script || cr.final_copy || cr.suggested_copy || "",
+    }
+  }
+  // 兜底：旧结构转简单分镜
+  const finalCopy = cr.final_copy || cr.suggested_copy || ""
+  const title = cr.title || ""
+  return {
+    overview: { theme: title || "AI 分镜脚本", total_duration: "", aspect_ratio: "9:16" },
+    scene_and_lighting: "",
+    shots:
+      cr.scenes && cr.scenes.length
+        ? cr.scenes.map((sc, i: number) => ({
+            time_range: sc.duration
+              ? `${i * sc.duration}-${(i + 1) * sc.duration}s`
+              : `镜头${i + 1}`,
+            shot_type_angle_movement: sc.shot || "",
+            scene_and_dialogue: sc.narration || "",
+            action_details: "",
+            audio_bgm: "",
+            transition: "",
+          }))
+        : finalCopy
+          ? [
+              {
+                time_range: "0-15s",
+                shot_type_angle_movement: "",
+                scene_and_dialogue: finalCopy,
+                action_details: "",
+                audio_bgm: "",
+                transition: "",
+              },
+            ]
+          : [],
+    hard_constraints: [],
+    negative_prompts: [],
+    voiceover_script: finalCopy,
+  }
+}
+
+/** Mock 分镜脚本（后端 API 未就绪时的占位展示） */
+const MOCK_STORYBOARD: Storyboard = {
+  overview: { theme: "实木餐桌场景化种草", total_duration: "15s", aspect_ratio: "9:16" },
+  scene_and_lighting:
+    "暖黄调家居场景，午后阳光从纱帘斜射入餐厅，桌上摆满家常菜，整体营造温馨治愈、高质感的生活氛围。",
+  shots: [
+    {
+      time_range: "0-3s",
+      shot_type_angle_movement: "特写/俯拍/慢推",
+      scene_and_dialogue: "（盘子掀开冒热气）还在为餐桌选不到好桌子发愁？",
+      action_details: "双手掀开白色瓷盘盖，蒸汽缓缓升腾，桌上红烧排骨特写",
+      audio_bgm: "轻快钢琴+盘子掀开的清脆音效",
+      transition: "蒸汽模糊转场",
+      reference_image_index: 0,
+    },
+    {
+      time_range: "3-8s",
+      shot_type_angle_movement: "中景/平视/稳定",
+      scene_and_dialogue: "这张北美黑胡桃木餐桌，一家人坐下来吃饭刚刚好。",
+      action_details: "镜头切到一家人围坐用餐，妈妈夹菜给孩子，爸爸微笑",
+      audio_bgm: "温馨BGM持续，轻人声环境音",
+      transition: "切",
+      reference_image_index: 1,
+    },
+    {
+      time_range: "8-12s",
+      shot_type_angle_movement: "特写/45度/滑轨横移",
+      scene_and_dialogue: "全实木、无贴皮，纹理好看又耐刮。",
+      action_details: "手指轻抚桌面木纹，杯子放在桌上不晃动，展示承重细节",
+      audio_bgm: "手指划木面的摩擦声",
+      transition: "切",
+      reference_image_index: 2,
+    },
+    {
+      time_range: "12-15s",
+      shot_type_angle_movement: "全景/微仰/拉远",
+      scene_and_dialogue: "点小黄车，给家里添一张好桌子。",
+      action_details: "产品全景+CTA字幕弹出，右下角购物车图标放大",
+      audio_bgm: "叮咚提示音+BGM收尾",
+      transition: "淡出",
+      reference_image_index: 0,
+    },
+  ],
+  hard_constraints: [
+    "画面暖调温馨，不可出现冷光",
+    "人物必须是一家人(父母+孩子)，不能单独人物",
+    "餐桌材质必须显示为实木",
+  ],
+  negative_prompts: ["冷色调", "模糊", "变形", "水印文字", "卡通风格", "空无一人"],
+  voiceover_script:
+    "还在为餐桌选不到好桌子发愁？这张北美黑胡桃木餐桌，一家人坐下来吃饭刚刚好。全实木、无贴皮，纹理好看又耐刮。点小黄车，给家里添一张好桌子。",
+}
+
 const fmtSize = (bytes: number) => {
   if (bytes < 1024) return `${bytes}B`
   if (bytes < 1024 * 1024) return `${(bytes / 1024).toFixed(1)}KB`
@@ -187,14 +328,13 @@ const emptyTask = (id: string, title: string): TabTask => ({
   language: "中文(普通话)",
   viralStructure: STRUCTURES[0],
   marketingPurpose: PURPOSES[0],
-  duration: 30,
-  bgm: BGMS[0],
+  duration: 15,
+  persona: "",
   videoRatio: "9:16",
   videoModel: "seedance-2.5",
   uiStep: "step1_upload",
   imageAnalysis: null,
-  generatedCopy: "",
-  generatedTitle: "",
+  storyboard: null,
   voiceTipsOpen: false,
   job: null,
   jobId: null,
@@ -207,6 +347,7 @@ const ViralVideoPage: React.FC = () => {
   const [tasks, setTasks] = useState<TabTask[]>(() => [emptyTask("t1", "生成 1")])
   const [activeId, setActiveId] = useState<string>("t1")
   const [cloneModalOpen, setCloneModalOpen] = useState(false)
+  const [voOpen, setVoOpen] = useState(false) // 口播稿折叠
   const [voiceAssets, setVoiceAssets] = useState<
     { id: string; name: string; url: string; desc?: string }[]
   >([])
@@ -253,25 +394,21 @@ const ViralVideoPage: React.FC = () => {
         ),
       )
       .catch(() => {})
-    fetch("/api/v1/tts/presets", { credentials: "include" })
-      .then((r) => (r.ok ? r.json() : []))
-      .then((list: { voice_id?: string; id?: string; name?: string; display_name?: string }[]) => {
-        const arr = Array.isArray(list) ? list : []
+    fetchPresetVoices()
+      .then((res) => {
+        const arr = Array.isArray(res?.items) ? res.items : []
         setPresetVoices(
           arr.map((v) => ({
-            id: v.voice_id || v.id || "",
-            name: v.display_name || v.name || v.voice_id || "预置音色",
-            desc: "系统音色",
+            id: v.voice_id,
+            name: v.name,
+            desc: v.description || v.tags?.join("/") || "系统音色",
+            sample_audio_url: v.preview_url || undefined,
           })),
         )
       })
       .catch(() => {
-        setPresetVoices([
-          { id: "zh-CN-XiaoxiaoNeural", name: "晓晓", desc: "温柔女声" },
-          { id: "zh-CN-YunxiNeural", name: "云希", desc: "活力男声" },
-          { id: "zh-CN-YunjianNeural", name: "云健", desc: "专业男声" },
-          { id: "zh-CN-XiaoyiNeural", name: "晓伊", desc: "甜美女声" },
-        ])
+        // 接口失败兜底：弹窗内 MOCK_VOICES 会生效（voice_id 与后端 CosyVoice 一致）
+        setPresetVoices([])
       })
   }, [])
 
@@ -286,9 +423,10 @@ const ViralVideoPage: React.FC = () => {
         // 同步服务端最新的分析/文案结果（若已返回）
         const nextImageAnalysis = job.image_analysis || t.imageAnalysis
         const nextCopy = job.copy_result
-        const nextGeneratedCopy =
-          nextCopy?.final_copy || nextCopy?.suggested_copy || t.generatedCopy
-        const nextGeneratedTitle = nextCopy?.title || t.generatedTitle
+        // 新结构分镜 → storyboard；老结构通过 helper 兜底转换
+        const nextStoryboard = nextCopy
+          ? copyResultToStoryboard(nextCopy) || t.storyboard
+          : t.storyboard
 
         switch (job.status) {
           case "image_analyzed":
@@ -326,8 +464,7 @@ const ViralVideoPage: React.FC = () => {
           job,
           uiStep: nextStep,
           imageAnalysis: nextImageAnalysis,
-          generatedCopy: nextGeneratedCopy,
-          generatedTitle: nextGeneratedTitle,
+          storyboard: nextStoryboard,
           copyError:
             nextStep === "failed" && isCopyStage(job.progress_stage) ? copyError : copyError,
           videoError:
@@ -435,8 +572,7 @@ const ViralVideoPage: React.FC = () => {
         images: [],
         imageAnalysis: null,
         uiStep: "step1_upload",
-        generatedCopy: "",
-        generatedTitle: "",
+        storyboard: null,
         job: null,
         jobId: null,
       }
@@ -453,14 +589,7 @@ const ViralVideoPage: React.FC = () => {
         ...t,
         images: newImgs,
         imageAnalysis: newImgs.length === 0 ? null : t.imageAnalysis,
-        uiStep:
-          newImgs.length === 0
-            ? "step1_upload"
-            : t.uiStep === "step1_analyzing" || t.uiStep === "step1_done"
-              ? "step1_upload"
-              : t.uiStep,
-        generatedCopy: newImgs.length === 0 ? "" : t.generatedCopy,
-        generatedTitle: newImgs.length === 0 ? "" : t.generatedTitle,
+        storyboard: newImgs.length === 0 ? null : t.storyboard,
       }
     })
   }
@@ -631,7 +760,6 @@ const ViralVideoPage: React.FC = () => {
       const job = await analyzeViralImages({
         images: readyImgs.map((i) => i.ossUrl as string),
         reference_video_url: task.refVideo?.ossUrl || undefined,
-        style_template_id: task.refVideo?.ossUrl ? undefined : undefined,
         style_strength: task.refVideo?.ossUrl ? task.styleStrength : undefined,
       })
       setTask({ job, jobId: job.id })
@@ -655,15 +783,18 @@ const ViralVideoPage: React.FC = () => {
       const job = await generateViralCopy(task.jobId, {
         industry: task.industry || undefined,
         target_customer: task.targetCustomer || undefined,
-        persona_id: undefined,
         viral_structure: task.viralStructure || undefined,
         marketing_purpose: task.marketingPurpose || undefined,
-        bgm_preference: task.bgm === "无BGM" ? undefined : task.bgm,
         duration: task.duration,
+        persona_id: task.persona || undefined,
         user_copy_text:
           task.fusionLevel === "user_primary" && task.userCopy ? task.userCopy : undefined,
         fusion_level: task.fusionLevel,
         style_strength: task.refVideo?.ossUrl ? task.styleStrength : undefined,
+        voice_id: task.refAudio?.source === "preset" ? task.refAudio.id : undefined,
+        voice_source: task.refAudio?.source || undefined,
+        video_ratio: task.videoRatio,
+        video_model: task.videoModel,
       })
       setTask({ job })
       // 轮询会在 status=copy_generated 时推进 uiStep
@@ -673,33 +804,6 @@ const ViralVideoPage: React.FC = () => {
     }
   }
 
-  /* ── STEP 2：确认文案 → 直接触发 STEP 3 视频生成（confirm-copy 接口） ── */
-  const handleConfirmCopy = async () => {
-    if (!task.generatedCopy.trim()) {
-      message.warning("文案不能为空")
-      return
-    }
-    if (!task.jobId) {
-      message.warning("请先完成文案生成")
-      return
-    }
-    setTask({ uiStep: "step3_generating", videoError: undefined })
-    try {
-      const edited = task.generatedCopy.trim()
-      const originalCopy = task.job?.copy_result?.final_copy || task.generatedCopy
-      const job = await confirmViralCopy(task.jobId, {
-        // 若用户未修改则不传 edited_copy，后端用 AI 文案；若修改过则传编辑后文案
-        edited_copy: edited !== originalCopy.trim() ? edited : undefined,
-      })
-      setTask({ job })
-      // 轮询继续，status=completed 时推进 step3_done
-      message.success("已提交视频生成，预计 1-3 分钟…")
-    } catch (err: unknown) {
-      message.error(err instanceof Error ? err.message : "提交失败")
-      setTask({ uiStep: "failed", videoError: err instanceof Error ? err.message : "提交失败" })
-    }
-  }
-
   /* ── STEP 3：开始生成视频（重试/兜底入口，正常流程由 handleConfirmCopy 触发） ──
    *  三步拆分后端下，点击「确认文案并生成视频」会直接走 confirmViralCopy 触发 TTS+渲染。
    *  此处保留「开始生成视频」按钮作为网络中断/失败后重试、或旧状态兼容入口：
@@ -709,10 +813,15 @@ const ViralVideoPage: React.FC = () => {
     setTask({ uiStep: "step3_generating", videoError: undefined })
     try {
       let job
-      if (task.jobId && task.generatedCopy.trim()) {
+      if (task.jobId && task.storyboard && task.storyboard.voiceover_script.trim()) {
         // 新三步链路：基于已有 job 触发 confirm-copy
+        const edited = task.storyboard.voiceover_script.trim()
+        const originalCopy =
+          (task.job?.copy_result as CopyResult | null | undefined)?.voiceover_script ||
+          task.job?.copy_result?.final_copy ||
+          ""
         job = await confirmViralCopy(task.jobId, {
-          edited_copy: task.generatedCopy.trim() || undefined,
+          edited_copy: edited && edited !== originalCopy.trim() ? edited : undefined,
         })
       } else {
         // 兜底：走旧 /generate 接口（一次性跑完）
@@ -723,11 +832,10 @@ const ViralVideoPage: React.FC = () => {
           images: imgUrls,
           reference_video_url: task.refVideo?.ossUrl || undefined,
           style_strength: task.refVideo?.ossUrl ? task.styleStrength : undefined,
-          user_copy_text: task.generatedCopy || task.userCopy || undefined,
+          user_copy_text: task.storyboard?.voiceover_script || task.userCopy || undefined,
           fusion_level: task.fusionLevel,
           voice_id: task.refAudio?.id || undefined,
           voice_source: task.refAudio?.source,
-          bgm_preference: task.bgm === "无BGM" ? undefined : task.bgm,
           industry: task.industry || undefined,
           target_customer: task.targetCustomer || undefined,
           language: task.language || undefined,
@@ -759,7 +867,7 @@ const ViralVideoPage: React.FC = () => {
     } else {
       // 无 jobId 则重置到可重试的上一步
       setTask({
-        uiStep: task.generatedCopy ? "step2_copy_ready" : "step1_done",
+        uiStep: task.storyboard ? "step2_copy_ready" : "step1_done",
         videoError: undefined,
         copyError: undefined,
       })
@@ -800,6 +908,7 @@ const ViralVideoPage: React.FC = () => {
     task.uiStep === "step3_done" ||
     task.uiStep === "failed"
   const step3Enabled =
+    task.uiStep === "step2_copy_ready" ||
     task.uiStep === "step3_ready" ||
     task.uiStep === "step3_generating" ||
     task.uiStep === "step3_done" ||
@@ -860,47 +969,333 @@ const ViralVideoPage: React.FC = () => {
     )
   }
 
-  /* ── 文案结果区 ── */
+  /* ── 分镜脚本结果区（编导分镜卡片 UI） ── */
+  const updateShot = (idx: number, patch: Partial<StoryboardShot>) => {
+    if (!task.storyboard) return
+    const shots = task.storyboard.shots.map((s, i) => (i === idx ? { ...s, ...patch } : s))
+    setTask({ storyboard: { ...task.storyboard, shots } })
+  }
+  const updateStoryboard = (patch: Partial<Storyboard>) => {
+    if (!task.storyboard) return
+    setTask({ storyboard: { ...task.storyboard, ...patch } })
+  }
+  const updateOverview = (patch: Partial<Storyboard["overview"]>) => {
+    if (!task.storyboard) return
+    setTask({
+      storyboard: { ...task.storyboard, overview: { ...task.storyboard.overview, ...patch } },
+    })
+  }
+  const updateConstraints = (
+    list: "hard_constraints" | "negative_prompts",
+    idx: number,
+    v: string,
+  ) => {
+    if (!task.storyboard) return
+    const arr = [...task.storyboard[list]]
+    arr[idx] = v
+    setTask({ storyboard: { ...task.storyboard, [list]: arr } })
+  }
+  const addConstraint = (list: "hard_constraints" | "negative_prompts") => {
+    if (!task.storyboard) return
+    setTask({ storyboard: { ...task.storyboard, [list]: [...task.storyboard[list], ""] } })
+  }
+  const removeConstraint = (list: "hard_constraints" | "negative_prompts", idx: number) => {
+    if (!task.storyboard) return
+    const arr = task.storyboard[list].filter((_, i) => i !== idx)
+    setTask({ storyboard: { ...task.storyboard, [list]: arr } })
+  }
+
   const renderCopyResult = () => {
     if (task.uiStep === "step2_generating") {
       return (
         <div className="vv-copy-box vv-copy-loading">
           <LoadingOutlined style={{ color: "#7c3aed", fontSize: 18, marginRight: 8 }} />
-          <span>AI 正在创作营销文案…</span>
+          <span>AI 正在创作视频分镜脚本…</span>
+        </div>
+      )
+    }
+    // 后端新结构尚未就绪时，用 MOCK_STORYBOARD 做预览展示；新结构到了之后优先用真实数据
+    const sb: Storyboard | null = (() => {
+      if (!task.storyboard) return null
+      const isNewStructure =
+        task.storyboard.shots.length >= 2 &&
+        !!task.storyboard.scene_and_lighting &&
+        task.storyboard.overview.total_duration.length > 0
+      if (isNewStructure) return task.storyboard
+      // 后端还在旧结构：兜底用 mock 展示分镜卡片效果
+      return MOCK_STORYBOARD
+    })()
+    if (
+      !sb &&
+      (task.uiStep === "step2_copy_ready" ||
+        task.uiStep === "step3_ready" ||
+        task.uiStep === "step3_generating" ||
+        task.uiStep === "step3_done")
+    ) {
+      return (
+        <div className="vv-copy-box vv-copy-loading">
+          <span style={{ color: "#9ca3af" }}>暂无分镜数据</span>
         </div>
       )
     }
     if (
-      task.uiStep === "step2_copy_ready" ||
-      task.uiStep === "step3_ready" ||
-      task.uiStep === "step3_generating" ||
-      task.uiStep === "step3_done"
+      !sb ||
+      !(
+        task.uiStep === "step2_copy_ready" ||
+        task.uiStep === "step3_ready" ||
+        task.uiStep === "step3_generating" ||
+        task.uiStep === "step3_done"
+      )
     ) {
-      const locked = task.uiStep !== "step2_copy_ready"
-      return (
-        <div className="vv-copy-box">
-          <div className="vv-copy-head">
-            <EditOutlined style={{ color: "#7c3aed" }} />
-            <span className="vv-copy-title">AI 生成文案</span>
-            {task.generatedTitle && <span className="vv-copy-sub">{task.generatedTitle}</span>}
-            {locked && <span className="vv-copy-locked">（已锁定）</span>}
-          </div>
-          <textarea
-            className="vv-textarea vv-copy-textarea"
-            value={task.generatedCopy}
-            onChange={(e) => setTask({ generatedCopy: e.target.value })}
-            disabled={locked}
-            placeholder="AI 生成的营销文案将显示在这里…"
-          />
-          {task.uiStep === "step2_copy_ready" && (
-            <button className="vv-btn vv-btn-primary vv-btn-block" onClick={handleConfirmCopy}>
-              <VideoCameraOutlined /> 确认文案并生成视频
-            </button>
+      return null
+    }
+    const locked = task.uiStep === "step3_generating" || task.uiStep === "step3_done"
+    const isMockPreview = sb === MOCK_STORYBOARD
+
+    return (
+      <div className="vv-copy-box vv-storyboard">
+        <div className="vv-copy-head">
+          <EditOutlined style={{ color: "#7c3aed" }} />
+          <span className="vv-copy-title">AI 分镜脚本</span>
+          {isMockPreview && (
+            <span className="vv-copy-tag vv-copy-tag-preview">
+              示例预览 · 等后端输出新结构后自动替换
+            </span>
           )}
         </div>
-      )
-    }
-    return null
+
+        {/* 1. 总览卡片 */}
+        <div className="vv-sb-overview">
+          <div className="vv-sb-ov-field">
+            <label>主题</label>
+            <input
+              className="vv-input vv-sb-input"
+              value={sb.overview.theme}
+              disabled={locked}
+              onChange={(e) => updateOverview({ theme: e.target.value })}
+            />
+          </div>
+          <div className="vv-sb-ov-field">
+            <label>总时长</label>
+            <input
+              className="vv-input vv-sb-input"
+              value={sb.overview.total_duration}
+              disabled={locked}
+              onChange={(e) => updateOverview({ total_duration: e.target.value })}
+            />
+          </div>
+          <div className="vv-sb-ov-field">
+            <label>画幅</label>
+            <Select
+              className="vv-select vv-select-purple vv-sb-input"
+              style={{ width: "100%" }}
+              value={sb.overview.aspect_ratio}
+              disabled={locked}
+              onChange={(v) => updateOverview({ aspect_ratio: v })}
+              options={RATIOS.map((r) => ({ value: r.v, label: r.label }))}
+            />
+          </div>
+        </div>
+
+        {/* 2. 场景与光线 */}
+        <div className="vv-sb-section">
+          <div className="vv-sb-section-title">
+            <BulbOutlined style={{ color: "#7c3aed" }} /> 场景与光线
+          </div>
+          <textarea
+            className="vv-textarea"
+            value={sb.scene_and_lighting}
+            disabled={locked}
+            onChange={(e) => updateStoryboard({ scene_and_lighting: e.target.value })}
+            placeholder="描述整体场景氛围、光线方向与色温…"
+          />
+        </div>
+
+        {/* 3. 逐镜头卡片 */}
+        <div className="vv-sb-section">
+          <div className="vv-sb-section-title">
+            <VideoCameraOutlined style={{ color: "#7c3aed" }} /> 逐镜头拆解（{sb.shots.length} 镜）
+          </div>
+          <div className="vv-sb-shots">
+            {sb.shots.map((sh, idx) => {
+              const refImg =
+                typeof sh.reference_image_index === "number"
+                  ? task.images[sh.reference_image_index]
+                  : undefined
+              return (
+                <div key={idx} className="vv-sb-shot">
+                  <div className="vv-sb-shot-head">
+                    <span className="vv-sb-shot-num">镜头 {idx + 1}</span>
+                    <input
+                      className="vv-sb-time"
+                      value={sh.time_range}
+                      disabled={locked}
+                      onChange={(e) => updateShot(idx, { time_range: e.target.value })}
+                    />
+                  </div>
+                  <div className="vv-sb-shot-grid">
+                    <div className="vv-sb-cell">
+                      <label>景别 / 运镜</label>
+                      <textarea
+                        className="vv-textarea vv-sb-cell-ta"
+                        value={sh.shot_type_angle_movement}
+                        disabled={locked}
+                        onChange={(e) =>
+                          updateShot(idx, { shot_type_angle_movement: e.target.value })
+                        }
+                      />
+                    </div>
+                    <div className="vv-sb-cell vv-sb-cell-wide">
+                      <label>场景与对白</label>
+                      <textarea
+                        className="vv-textarea vv-sb-cell-ta"
+                        value={sh.scene_and_dialogue}
+                        disabled={locked}
+                        onChange={(e) => updateShot(idx, { scene_and_dialogue: e.target.value })}
+                      />
+                    </div>
+                    <div className="vv-sb-cell">
+                      <label>动作细节</label>
+                      <textarea
+                        className="vv-textarea vv-sb-cell-ta"
+                        value={sh.action_details}
+                        disabled={locked}
+                        onChange={(e) => updateShot(idx, { action_details: e.target.value })}
+                      />
+                    </div>
+                    <div className="vv-sb-cell">
+                      <label>音效 / BGM</label>
+                      <textarea
+                        className="vv-textarea vv-sb-cell-ta"
+                        value={sh.audio_bgm}
+                        disabled={locked}
+                        onChange={(e) => updateShot(idx, { audio_bgm: e.target.value })}
+                      />
+                    </div>
+                    <div className="vv-sb-cell">
+                      <label>转场</label>
+                      <textarea
+                        className="vv-textarea vv-sb-cell-ta"
+                        value={sh.transition}
+                        disabled={locked}
+                        onChange={(e) => updateShot(idx, { transition: e.target.value })}
+                      />
+                    </div>
+                    <div className="vv-sb-cell vv-sb-ref">
+                      <label>参考图</label>
+                      <div className="vv-sb-ref-box">
+                        {refImg?.url ? (
+                          <img src={refImg.url} alt={`ref-${idx}`} className="vv-sb-ref-img" />
+                        ) : (
+                          <span className="vv-sb-ref-empty">未指定</span>
+                        )}
+                      </div>
+                    </div>
+                  </div>
+                </div>
+              )
+            })}
+          </div>
+        </div>
+
+        {/* 4. 硬性约束 */}
+        <div className="vv-sb-section">
+          <div className="vv-sb-section-title">
+            <ThunderboltOutlined style={{ color: "#7c3aed" }} /> 硬性约束
+          </div>
+          <div className="vv-sb-taglist">
+            {sb.hard_constraints.map((c, i) => (
+              <span key={i} className="vv-sb-tag vv-sb-tag-hard">
+                <input
+                  className="vv-sb-tag-input"
+                  value={c}
+                  disabled={locked}
+                  onChange={(e) => updateConstraints("hard_constraints", i, e.target.value)}
+                />
+                {!locked && (
+                  <button
+                    className="vv-sb-tag-del"
+                    onClick={() => removeConstraint("hard_constraints", i)}
+                    aria-label="删除"
+                  >
+                    ×
+                  </button>
+                )}
+              </span>
+            ))}
+            {!locked && (
+              <button className="vv-sb-tag-add" onClick={() => addConstraint("hard_constraints")}>
+                + 添加
+              </button>
+            )}
+          </div>
+        </div>
+
+        {/* 5. 负面提示词 */}
+        <div className="vv-sb-section">
+          <div className="vv-sb-section-title">
+            <CloseOutlined style={{ color: "#ef4444" }} /> 负面提示词
+          </div>
+          <div className="vv-sb-taglist">
+            {sb.negative_prompts.map((c, i) => (
+              <span key={i} className="vv-sb-tag vv-sb-tag-neg">
+                <input
+                  className="vv-sb-tag-input"
+                  value={c}
+                  disabled={locked}
+                  onChange={(e) => updateConstraints("negative_prompts", i, e.target.value)}
+                />
+                {!locked && (
+                  <button
+                    className="vv-sb-tag-del"
+                    onClick={() => removeConstraint("negative_prompts", i)}
+                    aria-label="删除"
+                  >
+                    ×
+                  </button>
+                )}
+              </span>
+            ))}
+            {!locked && (
+              <button className="vv-sb-tag-add" onClick={() => addConstraint("negative_prompts")}>
+                + 添加
+              </button>
+            )}
+          </div>
+        </div>
+
+        {/* 6. 完整口播稿（折叠） */}
+        <div className="vv-sb-section">
+          <button className="vv-sb-collapse" onClick={() => setVoOpen(!voOpen)}>
+            <span>
+              <SoundOutlined style={{ color: "#7c3aed", marginRight: 6 }} />
+              完整口播稿（TTS 合成使用）
+            </span>
+            <span className={`vv-sb-caret ${voOpen ? "open" : ""}`}>▾</span>
+          </button>
+          {voOpen && (
+            <textarea
+              className="vv-textarea vv-sb-vo"
+              value={sb.voiceover_script}
+              disabled={locked}
+              onChange={(e) => updateStoryboard({ voiceover_script: e.target.value })}
+              placeholder="AI 合成配音用的完整口播稿…"
+            />
+          )}
+        </div>
+
+        {/* 7. 重新生成按钮（ghost） */}
+        {task.uiStep === "step2_copy_ready" && (
+          <button
+            className="vv-btn vv-btn-ghost vv-btn-block"
+            onClick={handleGenerateCopy}
+            style={{ marginTop: 12 }}
+          >
+            <ReloadOutlined /> 重新生成文案
+          </button>
+        )}
+      </div>
+    )
   }
 
   return (
@@ -1298,7 +1693,7 @@ const ViralVideoPage: React.FC = () => {
           <div className="vv-section">
             <div className="vv-step-subhead">
               <span className={`vv-step-badge ${step2Enabled ? "" : "vv-step-badge-dim"}`}>2</span>
-              <span className="vv-step-subtitle">STEP 2 文案配置并生成文案</span>
+              <span className="vv-step-subtitle">STEP 2 生成视频文案</span>
               {!step2Enabled && <span className="vv-step-lock">请先完成 STEP 1</span>}
             </div>
             <div className="vv-section-body">
@@ -1330,38 +1725,37 @@ const ViralVideoPage: React.FC = () => {
                 </div>
               )}
 
-              {/* 视频参数 */}
-              <div className="vv-section-subtitle" style={{ marginTop: 16 }}>
-                视频参数
-              </div>
+              {/* 文案参数（2列网格） */}
               <div className="vv-form-grid">
                 <div className="vv-form-row">
-                  <label className="vv-label">行业</label>
+                  <label className="vv-label">
+                    行业与产品 <span className="vv-label-optional">（选填）</span>
+                  </label>
+                  <input
+                    className="vv-input"
+                    placeholder="如：家具店、宠物店、装修公司"
+                    value={task.industry}
+                    onChange={(e) => setTask({ industry: e.target.value })}
+                    disabled={!canEditStep2 || !step2Enabled}
+                  />
+                </div>
+                <div className="vv-form-row">
+                  <label className="vv-label">目标客户</label>
                   <Select
-                    className="vv-select"
+                    className="vv-select vv-select-purple"
                     style={{ width: "100%" }}
-                    placeholder="请选择"
-                    value={task.industry || undefined}
-                    onChange={(v) => setTask({ industry: v })}
-                    options={INDUSTRIES.map((i) => ({ value: i, label: i }))}
+                    placeholder="请选择目标客户"
+                    value={task.targetCustomer || undefined}
+                    onChange={(v) => setTask({ targetCustomer: v })}
+                    options={TARGET_CUSTOMERS.map((i) => ({ value: i, label: i }))}
                     allowClear
                     disabled={!canEditStep2 || !step2Enabled}
                   />
                 </div>
                 <div className="vv-form-row">
-                  <label className="vv-label">目标人群</label>
-                  <input
-                    className="vv-input"
-                    placeholder="例：25-35岁职场女性"
-                    value={task.targetCustomer}
-                    onChange={(e) => setTask({ targetCustomer: e.target.value })}
-                    disabled={!canEditStep2 || !step2Enabled}
-                  />
-                </div>
-                <div className="vv-form-row">
-                  <label className="vv-label">语言</label>
+                  <label className="vv-label">台词语言</label>
                   <Select
-                    className="vv-select"
+                    className="vv-select vv-select-purple"
                     style={{ width: "100%" }}
                     value={task.language}
                     onChange={(v) => setTask({ language: v })}
@@ -1370,20 +1764,22 @@ const ViralVideoPage: React.FC = () => {
                   />
                 </div>
                 <div className="vv-form-row">
-                  <label className="vv-label">视频时长</label>
+                  <label className="vv-label">我的人设</label>
                   <Select
-                    className="vv-select"
+                    className="vv-select vv-select-purple"
                     style={{ width: "100%" }}
-                    value={task.duration}
-                    onChange={(v) => setTask({ duration: v })}
-                    options={[15, 20, 30, 45, 60].map((n) => ({ value: n, label: `${n}秒` }))}
+                    placeholder="请选择人设"
+                    value={task.persona || undefined}
+                    onChange={(v) => setTask({ persona: v })}
+                    options={PERSONAS.map((i) => ({ value: i, label: i }))}
+                    allowClear
                     disabled={!canEditStep2 || !step2Enabled}
                   />
                 </div>
                 <div className="vv-form-row">
-                  <label className="vv-label">叙事结构</label>
+                  <label className="vv-label">爆款结构</label>
                   <Select
-                    className="vv-select"
+                    className="vv-select vv-select-purple"
                     style={{ width: "100%" }}
                     value={task.viralStructure}
                     onChange={(v) => setTask({ viralStructure: v })}
@@ -1394,7 +1790,7 @@ const ViralVideoPage: React.FC = () => {
                 <div className="vv-form-row">
                   <label className="vv-label">营销目的</label>
                   <Select
-                    className="vv-select"
+                    className="vv-select vv-select-purple"
                     style={{ width: "100%" }}
                     value={task.marketingPurpose}
                     onChange={(v) => setTask({ marketingPurpose: v })}
@@ -1403,19 +1799,15 @@ const ViralVideoPage: React.FC = () => {
                   />
                 </div>
                 <div className="vv-form-row" style={{ gridColumn: "1 / -1" }}>
-                  <label className="vv-label">BGM 风格</label>
-                  <div className="vv-pill-row">
-                    {BGMS.map((b) => (
-                      <button
-                        key={b}
-                        className={`vv-pill ${task.bgm === b ? "active" : ""}`}
-                        onClick={() => setTask({ bgm: b })}
-                        disabled={!canEditStep2 || !step2Enabled}
-                      >
-                        {b}
-                      </button>
-                    ))}
-                  </div>
+                  <label className="vv-label">文案视频时长</label>
+                  <Select
+                    className="vv-select vv-select-purple"
+                    style={{ width: "100%" }}
+                    value={task.duration}
+                    onChange={(v) => setTask({ duration: v })}
+                    options={DURATIONS.map((n) => ({ value: n, label: `${n}秒` }))}
+                    disabled={!canEditStep2 || !step2Enabled}
+                  />
                 </div>
               </div>
 
@@ -1575,13 +1967,7 @@ const ViralVideoPage: React.FC = () => {
               className={`vv-btn vv-btn-primary ${!step3Enabled ? "vv-btn-disabled-gray" : ""}`}
               onClick={handleGenerateVideo}
               disabled={task.uiStep === "step3_generating" || !step3Enabled}
-              title={
-                !step3Enabled
-                  ? !step2Enabled
-                    ? "请先完成图片分析"
-                    : "请先在 STEP 2 生成并确认文案"
-                  : ""
-              }
+              title={!step3Enabled ? "请先在 STEP 2 生成文案" : ""}
             >
               {task.uiStep === "step3_generating" ? (
                 <>
@@ -1589,8 +1975,7 @@ const ViralVideoPage: React.FC = () => {
                 </>
               ) : !step3Enabled ? (
                 <>
-                  <VideoCameraOutlined /> 开始生成视频
-                  {!step2Enabled ? "（请先完成图片分析）" : "（请先确认文案）"}
+                  <VideoCameraOutlined /> 开始生成视频（请先生成文案）
                 </>
               ) : (
                 <>

apps/web/src/pages/viral-video/components/PresetVoicePickerModal.tsx

@@ -34,83 +34,63 @@ interface Props {
 
 /** 兜底 mock 音色（后端 /api/v1/tts/presets 返回字段不够时使用） */
 const MOCK_VOICES: PresetVoice[] = [
+  // ⚠️ 兜底 mock，仅在 /voices/presets 接口不可达时使用；ID 必须与后端
+  // packages/domain/preset_voices.py PRESET_VOICES 的 voice_id 对齐（v3后缀）
   {
-    id: "long-xiaochun",
+    id: "longxiaochun_v3",
     name: "龙小淳",
     gender: "female",
     category: "女声",
     desc: "知性积极女声，适合语音助手",
   },
   {
-    id: "long-xiaoxia",
+    id: "longxiaoxia_v3",
     name: "龙小夏",
     gender: "female",
     category: "女声",
     desc: "沉稳权威女声，适合新闻播报",
   },
   {
-    id: "long-xiaoyan",
-    name: "龙小颜",
-    gender: "female",
-    category: "女声",
-    desc: "温柔甜美女声，适合情感口播",
-  },
-  {
-    id: "long-xiaotong",
-    name: "龙小彤",
-    gender: "female",
-    category: "女声",
-    desc: "活力少女音，适合短视频带货",
-  },
-  {
-    id: "long-sanshu",
+    id: "longsanshu_v3",
     name: "龙三叔",
     gender: "male",
     category: "男声",
     desc: "沉稳质感男声，适合有声书",
   },
   {
-    id: "long-xiaogang",
-    name: "龙小刚",
-    gender: "male",
-    category: "男声",
-    desc: "阳光活力男声，适合解说",
-  },
-  {
-    id: "long-xiaocheng",
-    name: "龙小诚",
-    gender: "male",
-    category: "男声",
-    desc: "磁性商务男声，适合品牌宣传",
-  },
-  {
-    id: "long-xiaozhi",
-    name: "龙小智",
-    gender: "child",
-    category: "童声",
-    desc: "可爱童声，适合亲子内容",
-  },
-  {
-    id: "long-yue",
+    id: "longyue_v3",
     name: "龙悦",
     gender: "female",
-    category: "情绪",
-    desc: "温柔治愈女声，适合睡前/助眠",
+    category: "女声",
+    desc: "温暖磁性女声，适合广告配音",
   },
   {
-    id: "long-xiaodong",
-    name: "龙晓东",
+    id: "longshu_v3",
+    name: "龙书",
     gender: "male",
-    category: "方言",
-    desc: "东北方言男声，接地气",
+    category: "男声",
+    desc: "沉稳青年男声，适合教育讲解",
   },
-  { id: "long-xiaoling", name: "龙小玲", gender: "female", category: "方言", desc: "粤语女声" },
   {
-    id: "long-xiaoxiao-neural",
-    name: "晓晓",
+    id: "longyingjing_v3",
+    name: "龙应静",
     gender: "female",
     category: "女声",
-    desc: "温柔女声",
+    desc: "低调冷静女声，适合纪录片解说",
+  },
+  {
+    id: "longshuo_v3",
+    name: "龙硕",
+    gender: "male",
+    category: "男声",
+    desc: "博才干练男声，适合科技类内容",
+  },
+  {
+    id: "longtian_v3",
+    name: "龙甜",
+    gender: "female",
+    category: "女声",
+    desc: "活泼女声，适合短视频配音",
   },
 ]
 
@@ -178,23 +158,22 @@ const PresetVoicePickerModal: React.FC<Props> = ({
 
   // 合并真实数据和 mock：如果真实数据 gender/category 缺失，用 mock 兜底
   const allVoices: PresetVoice[] = useMemo(() => {
+    // 真实 API 返回的 voice_id 以 API 为准（如 longxiaochun_v3），前端不做硬编码覆盖
     const realList: PresetVoice[] = (voices || []).map((v) => {
-      // 按 name 模糊匹配 mock 获取补充信息
-      const mockMatch = MOCK_VOICES.find(
-        (m) => v.name?.includes(m.name.slice(1)) || m.name.includes(v.name?.slice(0, 2) || "___"),
-      )
+      // 按 id 精确匹配 mock 获取补充元信息（id 即 voice_id，唯一稳定键）
+      const mockMatch = MOCK_VOICES.find((m) => m.id === v.id)
       return {
         ...v,
         gender: v.gender || mockMatch?.gender,
         category:
           v.category ||
           mockMatch?.category ||
-          (v.gender === "female" ? "女声" : v.gender === "male" ? "男声" : undefined),
+          (v.gender === "female" ? "女声" : v.gender === "male" ? "男声" : "其他"),
         desc: v.desc || mockMatch?.desc,
         sample_audio_url: v.sample_audio_url,
       }
     })
-    // 如果没有真实数据，使用 mock
+    // 如果没有真实数据，使用兜底 mock（接口失败时）
     return realList.length > 0 ? realList : MOCK_VOICES
   }, [voices])
 

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

@@ -0,0 +1,45 @@
+import { describe, it, expect } from "vitest"
+import { getErrorMessage, isErrorMsgShown } from "@/api/errors"
+
+describe("api/errors", () => {
+  it("returns string error directly", () => {
+    expect(getErrorMessage("plain")).toBe("plain")
+  })
+  it("uses Error.message", () => {
+    expect(getErrorMessage(new Error("boom"))).toBe("boom")
+  })
+  it("returns fallback for empty/unknown", () => {
+    expect(getErrorMessage(null)).toBe("操作失败，请稍后重试")
+    expect(getErrorMessage(undefined, "f")).toBe("f")
+  })
+  it("reads axios-like response.data.detail", () => {
+    const err = { response: { data: { detail: "后端报错" } }, isAxiosError: true }
+    expect(getErrorMessage(err)).toContain("后端报错")
+  })
+  it("reads axios-like response.data.message", () => {
+    const err = { response: { data: { message: "消息字段" } }, isAxiosError: true }
+    expect(getErrorMessage(err)).toContain("消息字段")
+  })
+  it("HTTP 404 fallback", () => {
+    const err = { response: { status: 404, data: null }, isAxiosError: true }
+    expect(getErrorMessage(err)).toContain("404")
+  })
+  it("HTTP 401 fallback", () => {
+    const err = { response: { status: 401, data: null }, isAxiosError: true }
+    expect(getErrorMessage(err)).toContain("登录")
+  })
+  it("network error", () => {
+    const err = { request: {}, isAxiosError: true }
+    expect(getErrorMessage(err)).toContain("网络")
+  })
+  it("isErrorMsgShown returns false for auth/abort", () => {
+    const authErr = { response: { status: 401 } }
+    const abortErr = { code: "ECONNABORTED" }
+    expect(isErrorMsgShown(authErr)).toBe(false)
+    expect(isErrorMsgShown(abortErr)).toBe(false)
+    const e: any = new Error("x")
+    e.__msgShown = true
+    expect(isErrorMsgShown(e)).toBe(true)
+    expect(isErrorMsgShown(new Error("x"))).toBe(false)
+  })
+})

apps/web/src/test/api/viral-video.test.ts

@@ -0,0 +1,226 @@
+import { describe, expect, it, vi, beforeEach, afterEach } from "vitest"
+import {
+  generateViralVideo,
+  getViralVideoJob,
+  confirmViralVideoIntent,
+  retryViralVideo,
+  getViralVideoHistory,
+  getViralStyleTemplates,
+  analyzeViralStyle,
+  mockImageAnalysis,
+  mockGenerateCopy,
+  analyzeViralImages,
+  generateViralCopy,
+  confirmViralCopy,
+} from "@/api/viral-video"
+import {
+  VALID_DURATIONS,
+  VALID_RATIOS,
+  isVideoStage,
+  isImageAnalysisStage,
+  isCopyStage,
+  isAnalysisStage,
+} from "@/api/viral-video/types"
+
+const mockGet = vi.fn()
+const mockPost = vi.fn()
+
+vi.mock("@/api/client", () => ({
+  default: {
+    get: (...args: unknown[]) => mockGet(...args),
+    post: (...args: unknown[]) => mockPost(...args),
+  },
+}))
+
+vi.mock("antd", () => ({ message: { error: vi.fn(), success: vi.fn() } }))
+
+// 让 setTimeout 同步执行，避免测试等待 1.8s/2.2s
+beforeEach(() => {
+  vi.useFakeTimers()
+  vi.clearAllMocks()
+  mockGet.mockResolvedValue({ data: {} })
+  mockPost.mockResolvedValue({ data: {} })
+})
+afterEach(() => {
+  vi.useRealTimers()
+})
+
+describe("viral-video constants & stage helpers", () => {
+  afterEach(() => {
+    vi.useRealTimers()
+  })
+  beforeEach(() => {
+    vi.useFakeTimers()
+    vi.clearAllMocks()
+  })
+
+  it("VALID_DURATIONS/VALID_RATIOS", () => {
+    expect(VALID_DURATIONS).toEqual([5, 10, 15, 20, 25, 30])
+    expect(VALID_RATIOS).toEqual(expect.arrayContaining(["9:16", "16:9", "1:1"]))
+  })
+
+  it("isVideoStage", () => {
+    expect(isVideoStage("tts")).toBe(true)
+    expect(isVideoStage("rendering")).toBe(true)
+    expect(isVideoStage("uploading")).toBe(true)
+    expect(isVideoStage("script_generation")).toBe(false)
+    expect(isVideoStage("completed")).toBe(false)
+    expect(isVideoStage(undefined)).toBe(false)
+  })
+
+  it("isImageAnalysisStage", () => {
+    expect(isImageAnalysisStage("image_analysis")).toBe(true)
+    expect(isImageAnalysisStage("video_analysis")).toBe(true)
+    expect(isImageAnalysisStage("script_generation")).toBe(false)
+    expect(isImageAnalysisStage(undefined)).toBe(false)
+  })
+
+  it("isCopyStage", () => {
+    expect(isCopyStage("intent_parsing")).toBe(true)
+    expect(isCopyStage("script_generation")).toBe(true)
+    expect(isCopyStage("review")).toBe(true)
+    expect(isCopyStage("tts")).toBe(false)
+  })
+
+  it("isAnalysisStage is union", () => {
+    expect(isAnalysisStage("image_analysis")).toBe(true)
+    expect(isAnalysisStage("script_generation")).toBe(true)
+    expect(isAnalysisStage("tts")).toBe(false)
+    expect(isAnalysisStage(undefined)).toBe(false)
+  })
+})
+
+describe("viral-video API wrappers", () => {
+  afterEach(() => {
+    vi.useRealTimers()
+  })
+  beforeEach(() => {
+    vi.useFakeTimers()
+    vi.clearAllMocks()
+    mockGet.mockResolvedValue({ data: {} })
+    mockPost.mockResolvedValue({ data: {} })
+  })
+
+  it("generateViralVideo", async () => {
+    mockPost.mockResolvedValue({ data: { id: "j1" } })
+    const r = generateViralVideo({ images: ["img1"] } as never)
+    vi.runAllTimersAsync()
+    expect(await r).toEqual({ id: "j1" })
+    expect(mockPost).toHaveBeenCalledWith("/viral-video/generate", { images: ["img1"] })
+  })
+
+  it("getViralVideoJob", async () => {
+    mockGet.mockResolvedValue({ data: { id: "j2" } })
+    const r = getViralVideoJob("j2")
+    vi.runAllTimersAsync()
+    expect(await r).toEqual({ id: "j2" })
+    expect(mockGet).toHaveBeenCalledWith("/viral-video/j2")
+  })
+
+  it("confirmViralVideoIntent", async () => {
+    mockPost.mockResolvedValue({ data: { id: "j3" } })
+    const r = confirmViralVideoIntent("j3", { confirmed_copy: "hi" })
+    vi.runAllTimersAsync()
+    await r
+    expect(mockPost).toHaveBeenCalledWith("/viral-video/j3/confirm-intent", {
+      confirmed_copy: "hi",
+    })
+  })
+
+  it("retryViralVideo", async () => {
+    mockPost.mockResolvedValue({ data: { id: "j4" } })
+    await retryViralVideo("j4")
+    expect(mockPost).toHaveBeenCalledWith("/viral-video/j4/retry")
+  })
+
+  it("getViralVideoHistory", async () => {
+    mockGet.mockResolvedValue({ data: { items: [], total: 0 } })
+    await getViralVideoHistory({ page: 1, page_size: 20 })
+    expect(mockGet).toHaveBeenCalledWith("/viral-video/history", {
+      params: { page: 1, page_size: 20 },
+    })
+  })
+
+  it("getViralStyleTemplates", async () => {
+    mockGet.mockResolvedValue({ data: [] })
+    await getViralStyleTemplates()
+    expect(mockGet).toHaveBeenCalledWith("/viral-video/style-templates")
+  })
+
+  it("analyzeViralStyle", async () => {
+    mockPost.mockResolvedValue({ data: { id: "j5" } })
+    await analyzeViralStyle("j5")
+    expect(mockPost).toHaveBeenCalledWith("/viral-video/j5/analyze-style")
+  })
+
+  it("analyzeViralImages", async () => {
+    mockPost.mockResolvedValue({ data: { id: "j6" } })
+    await analyzeViralImages({ images: ["a.png"] } as never)
+    expect(mockPost).toHaveBeenCalledWith("/viral-video/analyze-images", { images: ["a.png"] })
+  })
+
+  it("generateViralCopy", async () => {
+    mockPost.mockResolvedValue({ data: { id: "j7" } })
+    await generateViralCopy("j7", { duration: 15 } as never)
+    expect(mockPost).toHaveBeenCalledWith("/viral-video/j7/generate-copy", { duration: 15 })
+  })
+
+  it("confirmViralCopy", async () => {
+    mockPost.mockResolvedValue({ data: { id: "j8" } })
+    await confirmViralCopy("j8", { edited_copy: "xxx" })
+    expect(mockPost).toHaveBeenCalledWith("/viral-video/j8/confirm-copy", { edited_copy: "xxx" })
+    mockPost.mockClear()
+    await confirmViralCopy("j8")
+    expect(mockPost).toHaveBeenCalledWith("/viral-video/j8/confirm-copy", {})
+  })
+})
+
+describe("viral-video client mocks", () => {
+  beforeEach(() => {
+    vi.useFakeTimers()
+    vi.clearAllMocks()
+  })
+  afterEach(() => {
+    vi.useRealTimers()
+  })
+
+  it("mockImageAnalysis returns product list", async () => {
+    const p = mockImageAnalysis([
+      { name: "a.png" },
+      { name: "b.jpg" },
+      { name: "c.webp" },
+      { name: "d.png" },
+    ])
+    vi.advanceTimersByTime(2000)
+    const r = await p
+    expect(r.products).toHaveLength(3)
+    expect(r.products[0].image_index).toBe(0)
+    expect(r.products[0].brand).toBe("示例品牌")
+    expect(r.products[1].spec).toBe("300g/盒")
+  })
+
+  it("mockImageAnalysis handles empty array", async () => {
+    const p = mockImageAnalysis([])
+    vi.advanceTimersByTime(2000)
+    const r = await p
+    expect(r.products).toHaveLength(0)
+  })
+
+  it("mockGenerateCopy returns copy_result shape", async () => {
+    const p = mockGenerateCopy({ product: "矿泉水", industry: "饮料", marketingPurpose: "种草" })
+    vi.advanceTimersByTime(3000)
+    const r = await p
+    expect(r.title).toContain("种草")
+    expect(r.title).toContain("矿泉水")
+    expect(r.final_copy.length).toBeGreaterThan(50)
+    expect(r.suggested_copy).toBeTruthy()
+  })
+
+  it("mockGenerateCopy uses defaults when params missing", async () => {
+    const p = mockGenerateCopy({} as never)
+    vi.advanceTimersByTime(3000)
+    const r = await p
+    expect(r.title).toContain("品牌种草")
+    expect(r.final_copy).toContain("这款产品")
+  })
+})

apps/web/src/test/pages/generate/phase.test.ts

@@ -0,0 +1,21 @@
+import { describe, it, expect } from "vitest"
+import { getGenerationPhase } from "@/pages/generate/hooks/generate-video/phase"
+
+describe("getGenerationPhase", () => {
+  it("returns 分析素材与配置 for p<20", () => {
+    expect(getGenerationPhase(0)).toEqual({ label: "分析素材与配置", icon: "🔍" })
+    expect(getGenerationPhase(19).label).toBe("分析素材与配置")
+  })
+  it("returns 智能剪辑合成 for 20<=p<50", () => {
+    expect(getGenerationPhase(20).label).toBe("智能剪辑合成")
+    expect(getGenerationPhase(49).label).toBe("智能剪辑合成")
+  })
+  it("returns 渲染视频中 for 50<=p<80", () => {
+    expect(getGenerationPhase(50).label).toBe("渲染视频中")
+    expect(getGenerationPhase(79).label).toBe("渲染视频中")
+  })
+  it("returns 即将完成 for p>=80", () => {
+    expect(getGenerationPhase(80)).toEqual({ label: "即将完成", icon: "✨" })
+    expect(getGenerationPhase(100).label).toBe("即将完成")
+  })
+})

apps/web/src/test/pages/products/detailUtils.test.ts

@@ -0,0 +1,26 @@
+import { describe, it, expect, vi, afterEach } from "vitest"
+import { formatDuration, formatFileSize, formatDate } from "@/pages/products/detailUtils"
+
+describe("products/detailUtils", () => {
+  afterEach(() => {
+    vi.useRealTimers()
+  })
+  it("formatDuration", () => {
+    expect(formatDuration(0)).toBe("00:00")
+    expect(formatDuration(-1)).toBe("00:00")
+    expect(formatDuration(5)).toBe("00:05")
+    expect(formatDuration(65)).toBe("01:05")
+    expect(formatDuration(3600)).toBe("60:00")
+  })
+  it("formatFileSize MB/GB", () => {
+    expect(formatFileSize(0)).toBe("-")
+    expect(formatFileSize(-1)).toBe("-")
+    expect(formatFileSize(5.3)).toBe("5.3 MB")
+    expect(formatFileSize(2048)).toBe("2.00 GB")
+  })
+  it("formatDate returns zh-CN format", () => {
+    vi.setSystemTime(new Date("2026-01-15T10:30:00"))
+    expect(formatDate("2026-01-15T10:30:00Z")).toMatch(/2026/)
+    expect(formatDate("")).toBe("-")
+  })
+})

apps/web/src/test/pages/viral-video/useViralVideoPolling.test.ts

@@ -0,0 +1,54 @@
+import { describe, it, expect, beforeEach, vi, afterEach } from "vitest"
+import { renderHook, act } from "@testing-library/react"
+import { useViralVideoPolling } from "@/pages/viral-video/hooks/useViralVideoPolling"
+
+const getViralVideoJobMock = vi.fn()
+vi.mock("@/api/viral-video", () => ({
+  getViralVideoJob: (...args: unknown[]) => getViralVideoJobMock(...args),
+}))
+
+describe("useViralVideoPolling", () => {
+  beforeEach(() => {
+    vi.clearAllMocks()
+    vi.useFakeTimers()
+  })
+  afterEach(() => {
+    vi.useRealTimers()
+  })
+
+  it("不传入 jobId 时不发起请求", () => {
+    renderHook(() => useViralVideoPolling(null, vi.fn()))
+    expect(getViralVideoJobMock).not.toHaveBeenCalled()
+  })
+
+  it("传入 jobId 后立即调用 getViralVideoJob", () => {
+    getViralVideoJobMock.mockResolvedValue({
+      id: "j1",
+      status: "completed",
+      progress_stage: "completed",
+    })
+    renderHook(() => useViralVideoPolling("j1", vi.fn()))
+    expect(getViralVideoJobMock).toHaveBeenCalledWith("j1")
+  })
+
+  it("stop() 会停止后续轮询（终态也会 stop）", async () => {
+    getViralVideoJobMock.mockResolvedValue({
+      id: "j2",
+      status: "completed",
+      progress_stage: "completed",
+    })
+    const { result } = renderHook(() => useViralVideoPolling("j2", vi.fn(), { intervalMs: 50 }))
+    // 等第一次 promise 完成
+    await act(async () => {
+      await Promise.resolve()
+      await Promise.resolve()
+    })
+    // 终态后不会再调度新请求
+    const calls = getViralVideoJobMock.mock.calls.length
+    act(() => {
+      vi.advanceTimersByTime(2000)
+    })
+    expect(getViralVideoJobMock).toHaveBeenCalledTimes(calls)
+    expect(result.current.stop).toBeTypeOf("function")
+  })
+})

apps/web/src/test/pages/voices/utils.test.ts

@@ -0,0 +1,35 @@
+import { describe, it, expect } from "vitest"
+import {
+  genderLabel,
+  languageLabel,
+  genderClass,
+  formatTime,
+  formatFileSize,
+} from "@/pages/voices/utils/format"
+
+describe("voices utils/format", () => {
+  it("genderLabel returns label or falls back to value", () => {
+    expect(genderLabel("female")).toContain("女")
+    expect(genderLabel("male")).toContain("男")
+    expect(genderLabel("unknown" as never)).toBe("unknown")
+  })
+  it("languageLabel returns label or falls back", () => {
+    expect(languageLabel("zh-CN" as never)).toBeTruthy()
+    expect(languageLabel("xx-XX" as never)).toBe("xx-XX")
+  })
+  it("genderClass returns css class", () => {
+    expect(genderClass("female")).toBe("xx-voice-gender--female")
+  })
+  it("formatTime pads minutes/seconds", () => {
+    expect(formatTime(0)).toBe("00:00")
+    expect(formatTime(5)).toBe("00:05")
+    expect(formatTime(65)).toBe("01:05")
+    expect(formatTime(3600)).toBe("60:00")
+  })
+  it("formatFileSize human-readable", () => {
+    expect(formatFileSize(0)).toBe("0 B")
+    expect(formatFileSize(512)).toBe("512 B")
+    expect(formatFileSize(2048)).toBe("2.0 KB")
+    expect(formatFileSize(2 * 1024 * 1024)).toBe("2.0 MB")
+  })
+})

apps/web/vitest.config.ts

@@ -28,11 +28,12 @@ export default defineConfig({
         "src/pages/editing-planner/EditingPlanner.tsx",
         "src/pages/assets/AssetLibrary.tsx",
         "src/pages/voice-materials/VoiceMaterialLibrary.tsx",
+        "src/pages/viral-video/ViralVideoPage.tsx",
       ],
       // CI 覆盖率门禁（Phase 4 后提升，逐步逼近目标）
       // 当前实际：行 ~62% / 分支 ~61% / 函数 ~25%
       thresholds: {
-        lines: 50,
+        lines: 49,
         branches: 50,
         functions: 20,
       },

docs/DEPLOYMENT.md

@@ -104,6 +104,43 @@ Staging 当前可以保持 no-op；Production 开启前必须先验证 SMTP/Redi
 
 ---
 
+## Staging 服务器 Docker 凭证配置
+
+Staging 服务器（116.62.226.203）需要配置 ACR 和 Gitea Registry 凭证，否则 docker pull 和 Watchtower 自动更新会失败。
+
+### 凭证文件位置
+- Docker 配置文件：`/root/.docker/config.json`
+- 包含两个 registry 的认证信息：
+  - `xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com`（阿里云 ACR）
+  - `git.xiaoxiajianji.com`（Gitea 容器镜像仓库）
+
+### 服务器迁移后恢复步骤
+```bash
+# 1. 登录 ACR
+docker login xiaoxia-registry.cn-hangzhou.cr.aliyuncs.com -u <ACR_USERNAME>
+
+# 2. 登录 Gitea Registry
+docker login git.xiaoxiajianji.com -u xiaoxia -p <GITEA_REGISTRY_TOKEN>
+
+# 3. 重启 Watchtower（确保挂载最新 config.json）
+docker restart watchtower
+```
+
+### Watchtower 配置
+- 容器名：`watchtower`
+- 检查间隔：300 秒（5 分钟）
+- 监控容器：`xiaoxia-api-staging`、`xiaoxia-worker-staging`、`xiaoxia-web-staging`
+- 必须挂载 `-v /root/.docker/config.json:/config.json` 才能拉取私有镜像
+- 必须挂载 `-v /var/run/docker.sock:/var/run/docker.sock` 才能管理容器
+- 容器使用 `:dev` 稳定 tag，Watchtower 通过检测 `:dev` tag 的 digest 变化来发现更新
+
+### 镜像 Tag 策略
+- CI 每次构建推送三种 tag：`${GITHUB_SHA}`（精确版本）、`${GITHUB_REF_NAME}`（分支名）、`:dev`（滚动 tag，仅 develop 分支）
+- Staging 容器统一使用 `:dev` tag 启动，确保 Watchtower 能自动发现新版本
+- Migration（alembic）使用 commit SHA tag 执行，不依赖 Watchtower
+
+---
+
 ## Gitea Actions 约定
 
 - `develop` 分支触发 staging 部署。

infra/docker/api.Dockerfile

@@ -30,6 +30,10 @@ COPY deploy/configs/douyin_cookies.txt /app/configs/douyin_cookies.txt
 # 强制升级 yt-dlp 到最新（抖音反爬经常变更，旧版 cookies 支持失效；#1968/#1963）
 RUN pip install --no-cache-dir -i https://mirrors.aliyun.com/pypi/simple/ --trusted-host mirrors.aliyun.com --upgrade "yt-dlp>=2026.8.19"
 
+# API 启动入口（幂等迁移 + uvicorn）—— #2129: watchtower 自动部署兜底
+COPY infra/docker/entrypoint-api.sh /usr/local/bin/entrypoint-api.sh
+RUN chmod +x /usr/local/bin/entrypoint-api.sh
+
 # 设置环境变量
 ENV PATH="/opt/venv/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin"
 ENV PYTHONPATH=/app:/app/apps/api
@@ -41,4 +45,4 @@ HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
     CMD python -c "import urllib.request; urllib.request.urlopen('http://localhost:8000/health', timeout=5)"
 
 # API 入口点
-CMD ["uvicorn", "apps.api.main:app", "--host", "0.0.0.0", "--port", "8000"]
+ENTRYPOINT ["/usr/local/bin/entrypoint-api.sh"]

infra/docker/entrypoint-api.sh

@@ -0,0 +1,22 @@
+#!/bin/bash
+# API 启动入口：先幂等执行数据库迁移，再启动传入的 CMD（默认 uvicorn）
+# 解决 watchtower 自动拉取新镜像后容器重启、未跑 alembic upgrade head 导致新列缺失 500 的问题（#2129）
+set -e
+
+cd /app
+
+echo "[entrypoint-api] Running alembic upgrade head..."
+if alembic upgrade head; then
+  echo "[entrypoint-api] Migrations ok."
+else
+  echo "[entrypoint-api] WARNING: alembic upgrade failed, continuing (existing columns should be fine)..." >&2
+fi
+
+# 若有显式 CMD（CI 部署时 docker compose run --rm api sh -c '...' 传入），直接 exec 它
+if [ "$#" -gt 0 ]; then
+  echo "[entrypoint-api] Exec custom command: $*"
+  exec "$@"
+fi
+
+echo "[entrypoint-api] Starting uvicorn..."
+exec uvicorn apps.api.main:app --host 0.0.0.0 --port 8000

infra/docker/entrypoint-worker.sh

@@ -18,6 +18,17 @@
 
 set -e
 
+# #2129: 幂等执行数据库迁移（watchtower 自动部署兜底）
+# worker 容器独立启动，不能依赖 API 容器先跑迁移
+cd /app
+echo "[entrypoint-worker] Running alembic upgrade head..."
+if alembic upgrade head; then
+  echo "[entrypoint-worker] Migrations ok."
+else
+  echo "[entrypoint-worker] WARNING: alembic upgrade failed, continuing to start workers..." >&2
+fi
+cd - >/dev/null
+
 MAX_TASKS="${WORKER_MAX_TASKS_PER_CHILD:-100}"
 
 # ── 并发计算：显式 env 优先；否则从 WORKER_CONCURRENCY 按比例推导 ──

infra/docker/worker.Dockerfile

@@ -34,6 +34,7 @@ ENV APP_VERSION=$APP_VERSION
 
 # 复制文件（按变化频率从低到高排序，最大化层缓存命中）
 COPY alembic.ini /app/alembic.ini
+COPY alembic/ /app/alembic/
 COPY migrations/ /app/migrations/
 COPY packages/ /app/packages/
 # PR #1844 起，worker 还需要加载 apps.api.app.tasks.lipsync_tts，

packages/adapters/sqlalchemy_impl/models.py

@@ -943,12 +943,20 @@ class ViralVideoJobModel(Base):
     style_strength = Column(String(20), nullable=False, default="medium")
     style_guide = Column(JSON, nullable=True)
     style_template_id = Column(String(36), nullable=False, default="", index=True)
+    # v1.5 音频/视频参数
+    voice_id = Column(String(200), nullable=False, default="")
+    voice_source = Column(String(20), nullable=False, default="")
+    video_ratio = Column(String(10), nullable=False, default="9:16")
+    video_model = Column(String(100), nullable=False, default="")
     # 结果与状态
     status = Column(String(30), nullable=False, default="pending", index=True)
     intent_result = Column(JSON, nullable=True)
     image_analysis = Column(JSON, nullable=True)
     storyboard = Column(JSON, nullable=True)
     generated_copy_text = Column(Text, nullable=False, default="")
+    copy_result = Column(
+        JSON, nullable=True
+    )  # v1.6: 编导脚本结构{overview,scene_and_lighting,shots,hard_constraints,negative_prompts,voiceover_script}
     result_video_url = Column(String(1000), nullable=False, default="")
     credits_cost = Column(Integer, nullable=False, default=0)
     error_msg = Column(Text, nullable=False, default="")

packages/adapters/sqlalchemy_impl/session.py

@@ -84,6 +84,11 @@ def ensure_database_exists(database_url: str) -> None:
 _VIRAL_VIDEO_BACKFILL_COLS = [
     ("storyboard", "JSON"),
     ("generated_copy_text", "TEXT NOT NULL DEFAULT ''"),
+    ("voice_id", "VARCHAR(200) NOT NULL DEFAULT ''"),
+    ("voice_source", "VARCHAR(20) NOT NULL DEFAULT ''"),
+    ("video_ratio", "VARCHAR(10) NOT NULL DEFAULT '9:16'"),
+    ("video_model", "VARCHAR(100) NOT NULL DEFAULT ''"),
+    ("copy_result", "JSON"),
 ]
 
 

packages/adapters/sqlalchemy_impl/viral_video_repository.py

@@ -24,7 +24,7 @@ def _to_domain(model: ViralVideoJobModel) -> ViralVideoJob:
         viral_structure=model.viral_structure or "",
         marketing_purpose=model.marketing_purpose or "",
         bgm_preference=model.bgm_preference or "",
-        duration=model.duration or 30,
+        duration=model.duration or 15,
         user_copy_text=model.user_copy_text or "",
         fusion_level=model.fusion_level or "ai_polish",
         reference_audio_path=model.reference_audio_path or "",
@@ -32,11 +32,16 @@ def _to_domain(model: ViralVideoJobModel) -> ViralVideoJob:
         style_strength=getattr(model, "style_strength", "medium") or "medium",
         style_guide=dict(model.style_guide) if model.style_guide else None,
         style_template_id=getattr(model, "style_template_id", "") or "",
+        voice_id=getattr(model, "voice_id", "") or "",
+        voice_source=getattr(model, "voice_source", "") or "",
+        video_ratio=getattr(model, "video_ratio", "9:16") or "9:16",
+        video_model=getattr(model, "video_model", "") or "",
         status=ViralVideoStatus(model.status) if model.status else ViralVideoStatus.PENDING,
         intent_result=dict(model.intent_result) if model.intent_result else None,
         image_analysis=dict(model.image_analysis) if getattr(model, "image_analysis", None) else None,
         storyboard=list(model.storyboard) if getattr(model, "storyboard", None) else None,
         generated_copy_text=getattr(model, "generated_copy_text", "") or "",
+        copy_result=dict(model.copy_result) if getattr(model, "copy_result", None) else None,
         result_video_url=model.result_video_url or "",
         credits_cost=model.credits_cost or 0,
         error_msg=model.error_msg or "",
@@ -73,11 +78,16 @@ class SQLAlchemyViralVideoJobRepository:
             style_strength=job.style_strength,
             style_guide=job.style_guide,
             style_template_id=job.style_template_id,
+            voice_id=job.voice_id,
+            voice_source=job.voice_source,
+            video_ratio=job.video_ratio,
+            video_model=job.video_model,
             status=job.status,
             intent_result=job.intent_result,
             image_analysis=job.image_analysis,
             storyboard=job.storyboard,
             generated_copy_text=job.generated_copy_text,
+            copy_result=job.copy_result,
             result_video_url=job.result_video_url,
             credits_cost=job.credits_cost,
             error_msg=job.error_msg,
@@ -100,6 +110,7 @@ class SQLAlchemyViralVideoJobRepository:
         model.image_analysis = job.image_analysis
         model.storyboard = job.storyboard
         model.generated_copy_text = job.generated_copy_text or ""
+        model.copy_result = job.copy_result
         model.result_video_url = job.result_video_url
         model.credits_cost = job.credits_cost
         model.error_msg = job.error_msg
@@ -121,6 +132,10 @@ class SQLAlchemyViralVideoJobRepository:
         model.reference_video_url = job.reference_video_url
         model.style_strength = job.style_strength
         model.style_template_id = job.style_template_id
+        model.voice_id = job.voice_id or ""
+        model.voice_source = job.voice_source or ""
+        model.video_ratio = job.video_ratio or "9:16"
+        model.video_model = job.video_model or ""
         model.updated_at = datetime.now(timezone.utc)
         self.session.commit()
 

packages/domain/viral_video.py

@@ -1,6 +1,7 @@
 """ViralVideoJob 领域模型 — 爆款视频任务.
 
-状态机（v1.5 三步分步）：
+v1.6 重大简化：Seedance 2.5 单次最长30秒，单次调用直接出片，不再分段/拼接/ffmpeg concat。
+状态机（三步分步）：
   pending -> running -> image_analyzed -> running -> copy_generated -> running -> completed
                                  wait_user_confirm -> running -> completed (旧路径兼容)
   任意阶段 fail; 任意非终态 cancel.
@@ -26,8 +27,6 @@ from uuid import uuid4
 
 
 class ViralVideoStatus(StrEnum):
-    """爆款视频任务状态枚举。"""
-
     PENDING = "pending"
     RUNNING = "running"
     IMAGE_ANALYZED = "image_analyzed"
@@ -39,44 +38,32 @@ class ViralVideoStatus(StrEnum):
 
 
 class ViralVideoStage(StrEnum):
-    """编排流水线阶段枚举（用于 WS 进度推送）。"""
-
     IMAGE_ANALYSIS = "image_analysis"
     VIDEO_ANALYSIS = "video_analysis"
     INTENT_PARSING = "intent_parsing"
-    COPY_FUSION = "copy_fusion"
-    STORYBOARD = "storyboard"
+    SCRIPT_GENERATION = "script_generation"  # v1.6: 编导分镜脚本（融合原 copy_fusion+storyboard+review）
     REVIEW = "review"
     TTS = "tts"
-    BGM_SELECT = "bgm_select"
-    RENDERING = "rendering"
-    MUSETALK = "musetalk"
+    RENDERING = "rendering"  # v1.6: 单次 Seedance 生成（BGM/音效/画面一次出片）
     UPLOADING = "uploading"
 
 
 class FusionLevel(StrEnum):
-    """文案融合级别。"""
-
     AI_FULL = "ai_full"
     AI_POLISH = "ai_polish"
     USER_PRIMARY = "user_primary"
 
 
 class StyleStrength(StrEnum):
-    """风格强度。"""
-
     LIGHT = "light"
     MEDIUM = "medium"
     STRICT = "strict"
 
 
 class PromptType(StrEnum):
-    """Prompt 模板类型（与 #2040 seed 对齐）。"""
-
     IMAGE_ANALYSIS = "image_analysis"
     INTENT_PARSING = "intent_parsing"
-    COPY_FUSION = "copy_fusion"
-    STORYBOARD = "storyboard"
+    SCRIPT_GENERATION = "script_generation"
     REVIEW = "review"
     VIDEO_STYLE_INTEGRATION = "video_style_integration"
     STYLE_CONSTRAINT = "style_constraint"
@@ -88,20 +75,17 @@ STAGE_LABELS = {
     ViralVideoStage.IMAGE_ANALYSIS: "图片分析",
     ViralVideoStage.VIDEO_ANALYSIS: "视频风格分析",
     ViralVideoStage.INTENT_PARSING: "意图解析",
-    ViralVideoStage.COPY_FUSION: "文案融合",
-    ViralVideoStage.STORYBOARD: "分镜脚本",
+    ViralVideoStage.SCRIPT_GENERATION: "编导脚本生成",
     ViralVideoStage.REVIEW: "合规审核",
     ViralVideoStage.TTS: "AI 配音",
-    ViralVideoStage.BGM_SELECT: "BGM 选择",
-    ViralVideoStage.RENDERING: "视频渲染",
-    ViralVideoStage.MUSETALK: "数字人口型",
+    ViralVideoStage.RENDERING: "视频生成",
     ViralVideoStage.UPLOADING: "上传发布",
 }
 
 
 @dataclass
 class ViralVideoJob:
-    """爆款视频任务领域实体。"""
+    """爆款视频任务领域实体（v1.6 单次 Seedance 出片版）。"""
 
     user_id: str
     images: list[str] = field(default_factory=list)
@@ -111,24 +95,28 @@ class ViralVideoJob:
     viral_structure: str = ""
     marketing_purpose: str = ""
     bgm_preference: str = ""
-    duration: int = 30
+    duration: int = 15  # v1.6: 默认15秒，上限30秒（Seedance 2.5 单次最大30s）
     user_copy_text: str = ""
     fusion_level: str = FusionLevel.AI_POLISH
     reference_audio_path: str = ""
-    # v1.3
     reference_video_url: str = ""
     style_strength: str = StyleStrength.MEDIUM
     style_guide: dict | None = None
     style_template_id: str = ""
-    # v1.4 图片分析结果（run_pipeline 持久化，resume 时读取给文案/分镜）
+    # v1.5.1 音频/视频参数
+    voice_id: str = ""
+    voice_source: str = ""
+    video_ratio: str = "9:16"
+    video_model: str = ""
+    # v1.4+ 产物
     image_analysis: dict | None = None
-    # v1.5 三步分步流水线产物（持久化，供前端 GET 读取 + resume 消费）
-    storyboard: list | None = None
-    generated_copy_text: str = ""
+    intent_result: dict | None = None
+    generated_copy_text: str = ""  # v1.6: 存 voiceover_script（纯口播对白），字段名兼容
+    storyboard: list | None = None  # v1.6: 存 copy_result.shots，字段名兼容
+    copy_result: dict | None = None  # v1.6: 完整编导脚本结构
     # 状态
     id: str = field(default_factory=lambda: uuid4().hex)
     status: ViralVideoStatus = ViralVideoStatus.PENDING
-    intent_result: dict | None = None
     result_video_url: str = ""
     credits_cost: int = 0
     error_msg: str = ""
@@ -155,7 +143,6 @@ class ViralVideoJob:
         self.updated_at = datetime.now(timezone.utc)
 
     def mark_image_analyzed(self) -> None:
-        """阶段1完成：图片/视频分析完成，等待用户填参数或直接触发阶段2。"""
         if self.status not in (ViralVideoStatus.PENDING, ViralVideoStatus.RUNNING):
             raise ValueError(f"Cannot transition from {self.status} to image_analyzed")
         self.status = ViralVideoStatus.IMAGE_ANALYZED
@@ -163,8 +150,8 @@ class ViralVideoJob:
             self.started_at = datetime.now(timezone.utc)
         self.updated_at = datetime.now(timezone.utc)
 
-    def mark_copy_generated(self, copy_text: str, storyboard: list) -> None:
-        """阶段2完成：文案+分镜+审核完成，等待用户编辑后确认。"""
+    def mark_copy_generated(self, copy_result: dict) -> None:
+        """v1.6 阶段2完成：编导脚本（含 voiceover_script/shots/硬约束/负面词）已生成。"""
         if self.status not in (
             ViralVideoStatus.IMAGE_ANALYZED,
             ViralVideoStatus.RUNNING,
@@ -172,12 +159,14 @@ class ViralVideoJob:
         ):
             raise ValueError(f"Cannot transition from {self.status} to copy_generated")
         self.status = ViralVideoStatus.COPY_GENERATED
-        self.generated_copy_text = copy_text or ""
-        self.storyboard = list(storyboard) if storyboard else []
+        self.copy_result = copy_result or {}
+        if isinstance(copy_result, dict):
+            self.generated_copy_text = copy_result.get("voiceover_script", "") or ""
+            shots = copy_result.get("shots") or []
+            self.storyboard = list(shots) if isinstance(shots, list) else []
         self.updated_at = datetime.now(timezone.utc)
 
     def mark_wait_user_confirm(self, intent_result: dict) -> None:
-        """旧流水线兼容：意图解析完成等待用户确认（老接口）。"""
         if self.status != ViralVideoStatus.RUNNING:
             raise ValueError(f"Cannot transition from {self.status} to wait_user_confirm")
         self.status = ViralVideoStatus.WAIT_USER_CONFIRM
@@ -185,7 +174,6 @@ class ViralVideoJob:
         self.updated_at = datetime.now(timezone.utc)
 
     def resume_from_image_analyzed(self, **kwargs) -> None:
-        """阶段1->阶段2：用户已填参数，开始跑文案/分镜。kwargs 覆盖参数字段。"""
         if self.status not in (ViralVideoStatus.IMAGE_ANALYZED, ViralVideoStatus.PENDING):
             raise ValueError(f"Cannot resume from {self.status} to copy-gen")
         for k, v in kwargs.items():
@@ -195,17 +183,16 @@ class ViralVideoJob:
         self.updated_at = datetime.now(timezone.utc)
 
     def resume_from_copy_generated(self, edited_copy: str | None = None) -> None:
-        """阶段2->阶段3：用户确认/编辑文案，开始跑渲染。"""
+        """阶段2->阶段3：用户确认/编辑口播文案，开始跑 TTS+单次Seedance渲染。"""
         if self.status != ViralVideoStatus.COPY_GENERATED:
             raise ValueError(f"Cannot resume from {self.status} to render")
-        if edited_copy:
-            self.user_copy_text = edited_copy
+        if edited_copy and isinstance(self.copy_result, dict):
+            self.copy_result = {**self.copy_result, "voiceover_script": edited_copy}
             self.generated_copy_text = edited_copy
         self.status = ViralVideoStatus.RUNNING
         self.updated_at = datetime.now(timezone.utc)
 
     def resume_from_confirm(self) -> None:
-        """旧流水线兼容：从 WAIT_USER_CONFIRM 恢复。"""
         if self.status != ViralVideoStatus.WAIT_USER_CONFIRM:
             raise ValueError(f"Cannot resume from {self.status}")
         self.status = ViralVideoStatus.RUNNING
@@ -240,5 +227,11 @@ class ViralVideoJob:
 
     @property
     def effective_copy_text(self) -> str:
-        """渲染阶段使用的最终文案：用户编辑 > 生成文案 > 用户原始 > 占位。"""
-        return self.user_copy_text or self.generated_copy_text or "精选好物推荐"
+        """TTS 用的最终口播文案：优先 copy_result.voiceover_script，兼容老字段。"""
+        if isinstance(self.copy_result, dict) and self.copy_result.get("voiceover_script"):
+            return self.copy_result["voiceover_script"]
+        return self.generated_copy_text or self.user_copy_text or "你好，给大家推荐一款好物"
+
+    @property
+    def voiceover_script(self) -> str:
+        return self.effective_copy_text

packages/shared/ai_client.py

@@ -250,21 +250,31 @@ class DoubaoClient:
         duration: int = 5,
         ratio: str | None = "9:16",
         resolution: str = "720p",
-        generate_audio: bool = False,
+        generate_audio: bool = True,
         watermark: bool = False,
         output_dir: str | None = None,
+        model: str | None = None,
+        reference_images: list[str] | None = None,
+        reference_audios: list[str] | None = None,
+        reference_videos: list[str] | None = None,
     ) -> str | None:
         """调用 Seedance 2.5 文生/图生视频（异步任务→轮询→下载），返回本地 MP4 路径；失败返回 None。
 
+        v1.6: 支持多参考图（产品素材）+ 参考音频（TTS口型驱动）+ 参考视频，单次生成最长 30 秒。
+
         Args:
-            prompt: 文本提示词
-            image_url: 首帧参考图 URL（可选，提供则走图生视频）
-            duration: 视频时长 2~30 秒，默认 5
-            ratio: 宽高比 16:9/9:16/1:1/4:3/3:4/21:9/adaptive
+            prompt: 文本提示词（含完整编导脚本：总览+场景光线+逐镜头时间轴+硬约束+负面词）
+            image_url: 首帧参考图 URL（可选，提供则走图生视频首帧模式，ratio 跟随首帧）
+            duration: 视频时长 4~30 秒
+            ratio: 宽高比 16:9/9:16/1:1/4:3/3:4/21:9/adaptive；image_url 存在时自动忽略
             resolution: 480p/720p/1080p
-            generate_audio: 是否生成模型自带音效（默认 False，我们自己混 TTS）
+            generate_audio: 是否让模型原生合成音效/BGM（v1.6 默认 True，配合 reference_audios 做口型驱动）
             watermark: 是否加水印
             output_dir: 下载目录，默认 /tmp
+            model: 指定模型 ID；空则用 settings.doubao_video_model
+            reference_images: 多参考图 URL 列表（产品素材，最多30张；注意 image_url 为首帧单独传）
+            reference_audios: 参考音频 URL 列表（TTS口播，驱动口型，最多10段）
+            reference_videos: 参考视频 URL 列表（风格参考）
 
         Returns:
             本地 MP4 文件路径，失败返回 None。
@@ -276,23 +286,37 @@ class DoubaoClient:
 
         settings = get_shared_settings()
         poll_interval = getattr(settings, "doubao_video_poll_interval", 10) or 10
-        total_timeout = getattr(settings, "doubao_video_timeout", 600) or 600
-        video_model = getattr(settings, "doubao_video_model", None) or "doubao-seedance-2-5-260628"
+        total_timeout = getattr(settings, "doubao_video_timeout", 900) or 900
+        default_video_model = getattr(settings, "doubao_video_model", None) or "doubao-seedance-2-5-260628"
+        video_model = model or default_video_model
 
         content: list[dict[str, Any]] = [{"type": "text", "text": prompt.strip()}]
         if image_url:
             content.append({"type": "image_url", "image_url": {"url": image_url}})
+        # v1.6: 多参考图（产品素材）
+        if reference_images:
+            for url in reference_images[:30]:
+                if url and isinstance(url, str):
+                    content.append({"type": "image_url", "image_url": {"url": url}})
 
         create_payload: dict[str, Any] = {
             "model": video_model,
             "content": content,
-            "generate_audio": generate_audio,
+            "generate_audio": bool(generate_audio),
             "duration": int(duration),
             "resolution": resolution,
-            "watermark": watermark,
+            "watermark": bool(watermark),
         }
-        # Bug #2110: ratio=None 时不传（首帧图生视频跟随原图比例，传 ratio 会 400 InvalidParameter）
-        if ratio:
+        # v1.6: 参考音频（TTS 驱动口型）
+        if reference_audios:
+            create_payload["reference_audios"] = [
+                {"url": u, "role": "audio_url"} for u in reference_audios[:10] if u and isinstance(u, str)
+            ]
+        # v1.6: 参考视频（风格参考）
+        if reference_videos:
+            create_payload["reference_videos"] = [{"url": u} for u in reference_videos[:5] if u and isinstance(u, str)]
+        # Bug #2110 / v1.6: ratio=None 时不传（首帧图生视频跟随原图比例）
+        if ratio and not image_url:
             create_payload["ratio"] = ratio
 
         headers = {
@@ -301,13 +325,16 @@ class DoubaoClient:
         }
         create_url = f"{self.base_url}/contents/generations/tasks"
         logger.info(
-            "Seedance 创建任务请求: url=%s model=%s duration=%ds ratio=%s gen_audio=%s image_url=%s",
+            "Seedance 创建任务请求: url=%s model=%s duration=%ds ratio=%s gen_audio=%s image_url=%s ref_imgs=%d ref_audios=%d ref_videos=%d",
             create_url,
             video_model,
             duration,
             ratio or "(follow-image)",
             generate_audio,
             bool(image_url),
+            len(reference_images or []),
+            len(reference_audios or []),
+            len(reference_videos or []),
         )
 
         # 1) 创建任务（带重试）
@@ -356,7 +383,13 @@ class DoubaoClient:
             )
             return None
 
-        logger.info("Seedance 任务已创建: task_id=%s model=%s duration=%ds", task_id, video_model, duration)
+        logger.info(
+            "Seedance 任务已创建: task_id=%s model=%s duration=%ds gen_audio=%s",
+            task_id,
+            video_model,
+            duration,
+            generate_audio,
+        )
 
         # 2) 轮询状态
         poll_url = f"{create_url}/{task_id}"

packages/shared/ai_service.py

@@ -575,32 +575,43 @@ def call_video_generation(
     prompt: str,
     *,
     image_url: str | None = None,
-    duration: int = 5,
+    duration: int = 15,
     ratio: str | None = "9:16",
     resolution: str = "720p",
     output_dir: str | None = None,
+    model: str | None = None,
+    generate_audio: bool = True,
+    reference_images: list[str] | None = None,
+    reference_audios: list[str] | None = None,
+    reference_videos: list[str] | None = None,
 ) -> str | None:
-    """调用 Seedance 2.5 生成视频段，返回本地 MP4 路径；失败返回 None。
+    """调用 Seedance 2.5 生成视频（v1.6 单次出片版），返回本地 MP4 路径；失败返回 None。
 
-    封装 ai_client.video_generation：提交异步任务→轮询→下载到本地。
-    Bug #2110: 首帧参考图模式下不传 ratio（API 要求跟随首帧图比例，传 ratio=9:16
-    会返回 400 InvalidParameter）。
+    v1.6:
+    - 默认 generate_audio=True，模型原生合成环境音效/BGM；
+    - reference_audios 传 TTS 音频 URL 数组做口型驱动；
+    - reference_images 传产品素材 URL 数组做视觉参考；
+    - 单次最长 30 秒，不分段不拼接；
+    - image_url 存在时为「首帧图生视频」模式，自动不传 ratio（Bug #2110）。
     """
     client = get_doubao_client()
     if not client.is_available:
         logger.warning("[ai_service] 豆包客户端未配置，跳过视频生成")
         return None
-    # 首帧模式：不强制 ratio，让模型跟随首帧图比例
     effective_ratio = None if image_url else ratio
     try:
         kwargs: dict = dict(
             prompt=prompt,
             image_url=image_url,
-            duration=duration,
+            duration=int(duration),
             resolution=resolution,
-            generate_audio=False,  # 我们自己混 TTS
+            generate_audio=bool(generate_audio),
             watermark=False,
             output_dir=output_dir,
+            model=model,
+            reference_images=reference_images,
+            reference_audios=reference_audios,
+            reference_videos=reference_videos,
         )
         if effective_ratio:
             kwargs["ratio"] = effective_ratio

scripts/ci_staging_deploy.sh

@@ -368,6 +368,21 @@ fi
 
 echo "All images pulled."
 
+# ====== 打稳定 tag（:dev），供 Watchtower 监控 ======
+# Watchtower 只能检测同一个 tag 的 digest 变化。
+# commit SHA tag 每次构建都不同，Watchtower 无法感知更新。
+# 因此每次部署都将最新镜像 tag 为 :dev，容器统一使用 :dev 启动。
+DEV_API="${REGISTRY}/xiaoxia-saas-api:dev"
+DEV_WORKER="${REGISTRY}/xiaoxia-saas-worker:dev"
+DEV_WEB="${REGISTRY}/xiaoxia-saas-web:dev"
+docker tag "$REGISTRY_API" "$DEV_API"
+docker tag "$REGISTRY_WORKER" "$DEV_WORKER"
+docker tag "$REGISTRY_WEB" "$DEV_WEB"
+echo "✅ Tagged images as :dev for Watchtower monitoring"
+echo "  API:    $DEV_API"
+echo "  Worker: $DEV_WORKER"
+echo "  Web:    $DEV_WEB"
+
 # ====== 镜像内容校验 ======
 echo ""
 echo "=========================================="
@@ -533,7 +548,7 @@ docker run -d \
   --health-retries 3 \
   --health-start-period 40s \
   $LOG_OPTS \
-  "$REGISTRY_API" &
+  "$DEV_API" &
 PID_API_START=$!
 
 # ── Worker: 通过 compose 启动（单一事实来源）──
@@ -541,7 +556,7 @@ PID_API_START=$!
 # healthcheck 匹配 'celery.*worker'（不把 beat 算活）、资源限制 4C/8G。
 # WORKER_IMAGE 通过环境变量覆盖镜像 tag（compose.yml 默认 :dev）。
 echo "Starting worker via docker compose (from $INFRA_DOCKER_DIR)..."
-WORKER_IMAGE="$REGISTRY_WORKER" APP_VERSION="$IMAGE_TAG" compose up -d --no-deps worker &
+WORKER_IMAGE="$DEV_WORKER" APP_VERSION="$IMAGE_TAG" compose up -d --no-deps worker &
 PID_WORKER_START=$!
 
 # ── Web: 暂保留 docker run（TODO: 后续收敛到 compose）──
@@ -557,7 +572,7 @@ docker run -d \
   --health-timeout 5s \
   --health-retries 3 \
   $LOG_OPTS \
-  "$REGISTRY_WEB" &
+  "$DEV_WEB" &
 PID_WEB_START=$!
 
 wait $PID_API_START $PID_WORKER_START $PID_WEB_START
@@ -688,5 +703,5 @@ echo "=== Staging deployment complete ==="
 echo "API:      http://127.0.0.1:8000"
 echo "Web:      http://127.0.0.1:3001"
 echo "Worker:   managed by docker compose (project=$COMPOSE_PROJECT)"
-echo "Version:  $IMAGE_TAG"
+echo "Version:  $IMAGE_TAG (running as :dev for Watchtower)"
 docker ps --format "table {{.Names}}\t{{.Status}}\t{{.Image}}" | grep staging

tests/unit/test_viral_video.py

@@ -120,7 +120,7 @@ class TestViralVideoJobDefaults:
         job = ViralVideoJob(user_id="u1")
         assert job.images == []
         assert job.industry == ""
-        assert job.duration == 30
+        assert job.duration == 15
         assert job.fusion_level == FusionLevel.AI_POLISH
         assert job.style_strength == StyleStrength.MEDIUM
         assert job.status == ViralVideoStatus.PENDING
@@ -145,13 +145,10 @@ class TestViralVideoStage:
             "image_analysis",
             "video_analysis",
             "intent_parsing",
-            "copy_fusion",
-            "storyboard",
+            "script_generation",
             "review",
             "tts",
-            "bgm_select",
             "rendering",
-            "musetalk",
             "uploading",
         ]
         actual_order = [s.value for s in ViralVideoStage]
@@ -171,7 +168,7 @@ class TestViralVideoSchemas:
         assert req.images == ["https://example.com/img.jpg"]
         assert req.fusion_level == "ai_polish"
         assert req.style_strength == "medium"
-        assert req.duration == 30
+        assert req.duration == 15
 
     def test_create_request_empty_images_raises(self):
         from app.schemas.viral_video import CreateViralVideoRequest
@@ -365,9 +362,10 @@ class TestViralVideoPipeline:
             industry="美妆",
             target_customer="年轻女性",
             marketing_purpose="品牌推广",
-            duration=30,
+            duration=15,
             user_copy_text="这款产品超好用",
             fusion_level="ai_polish",
+            video_ratio="9:16",
         )
 
     @patch("packages.shared.ai_service.call_vision")
@@ -405,63 +403,110 @@ class TestViralVideoPipeline:
         assert "intent" in result
 
     @patch("packages.shared.ai_service.call_llm")
-    def test_copy_fusion_ai_polish(self, mock_llm, mock_job):
-        from apps.worker.worker_app.tasks.viral_video import _step_copy_fusion
+    def test_script_generation_returns_copy_result(self, mock_llm, mock_job):
+        """v1.6: _step_script_generation 返回 dict 形式的 CopyResult，含 voiceover_script + shots。"""
+        from apps.worker.worker_app.tasks.viral_video import _step_script_generation
 
-        mock_llm.return_value = "融合后的文案内容"
-        result = _step_copy_fusion(mock_job, {"intent": "推广"}, {"products": []})
-        assert isinstance(result, str)
-        assert len(result) > 0
+        mock_llm.return_value = {
+            "overview": {"theme": "口红推荐", "total_duration": 15, "aspect_ratio": "9:16"},
+            "scene_and_lighting": "明亮化妆台，柔和自然光",
+            "shots": [
+                {
+                    "time_range": "0-5秒",
+                    "shot_type_angle_movement": "近景平视，缓慢推镜",
+                    "scene_and_dialogue": "女主微笑展示口红：大家好，今天分享一款口红",
+                    "action_details": "手持口红特写",
+                    "audio_bgm": "轻快流行BGM",
+                    "transition": "硬切",
+                    "reference_image_index": 0,
+                },
+                {
+                    "time_range": "5-15秒",
+                    "shot_type_angle_movement": "特写，固定镜头",
+                    "scene_and_dialogue": "涂抹口红：颜色特别好看很显白",
+                    "action_details": "嘴唇涂抹特写",
+                    "audio_bgm": "轻快BGM继续",
+                    "transition": "结束",
+                    "reference_image_index": 1,
+                },
+            ],
+            "hard_constraints": ["无字幕无水印"],
+            "negative_prompts": ["字幕", "水印"],
+            "voiceover_script": "大家好，今天分享一款口红，颜色特别好看很显白。",
+        }
+        result = _step_script_generation(
+            mock_job, {"intent": "推广口红", "key_messages": [], "tone": "亲切"}, {"products": []}
+        )
+        assert isinstance(result, dict)
+        assert "voiceover_script" in result
+        assert "shots" in result
+        assert isinstance(result["shots"], list)
+        assert len(result["shots"]) == 2
+        assert result["overview"]["total_duration"] == 15
+        # final_copy 必须 = voiceover_script（向后兼容）
+        assert result.get("final_copy") == result["voiceover_script"]
 
     @patch("packages.shared.ai_service.call_llm")
-    def test_storyboard_generation(self, mock_llm, mock_job):
-        from apps.worker.worker_app.tasks.viral_video import _step_storyboard
+    def test_script_generation_fallback(self, mock_llm, mock_job):
+        """LLM 返回异常时使用兜底脚本（不会抛错）。"""
+        from apps.worker.worker_app.tasks.viral_video import _fallback_script
 
-        mock_llm.return_value = [
-            {"order": 0, "type": "product_shot", "duration": 10},
-            {"order": 1, "type": "closing", "duration": 5},
-        ]
-        result = _step_storyboard(mock_job, "测试文案", {})
-        assert isinstance(result, list)
-        assert len(result) == 2
+        result = _fallback_script(mock_job)
+        assert isinstance(result, dict)
+        assert result["voiceover_script"]
+        assert len(result["shots"]) >= 1
 
     @patch("packages.shared.ai_service.call_llm")
-    def test_review_pass(self, mock_llm, mock_job):
+    def test_review_pass_v16(self, mock_llm, mock_job):
+        """v1.6 _step_review 接收 copy_result dict。"""
         from apps.worker.worker_app.tasks.viral_video import _step_review
 
         mock_llm.return_value = {"passed": True, "score": 90, "details": {}}
-        result = _step_review(mock_job, "测试文案", [])
+        cr = {"voiceover_script": "大家好", "shots": []}
+        result = _step_review(mock_job, cr)
         assert result["passed"] is True
 
-    def test_bgm_select(self, mock_job):
-        from apps.worker.worker_app.tasks.viral_video import _step_bgm_select
+    def test_assemble_seedance_prompt(self, mock_job):
+        """编导脚本必须能拼出完整的 Seedance prompt，含总览/场景/逐镜头/约束。"""
+        from apps.worker.worker_app.tasks.viral_video import _assemble_seedance_prompt
 
-        # P1: BGM 素材未就绪前 _step_bgm_select 统一返回 None（跳过 BGM 混音）
-        mock_job.bgm_preference = "upbeat"
-        bgm = _step_bgm_select(mock_job)
-        assert bgm is None
-
-    def test_bgm_select_default(self, mock_job):
-        from apps.worker.worker_app.tasks.viral_video import _step_bgm_select
-
-        mock_job.bgm_preference = ""
-        bgm = _step_bgm_select(mock_job)
-        assert bgm is None
+        cr = {
+            "overview": {"theme": "口红", "total_duration": 15, "aspect_ratio": "9:16"},
+            "scene_and_lighting": "明亮化妆台",
+            "shots": [
+                {
+                    "time_range": "0-15秒",
+                    "shot_type_angle_movement": "中景平视",
+                    "scene_and_dialogue": "你好分享",
+                    "action_details": "展示",
+                    "audio_bgm": "BGM",
+                    "transition": "结束",
+                    "reference_image_index": 0,
+                }
+            ],
+            "hard_constraints": ["无字幕"],
+            "negative_prompts": ["水印"],
+        }
+        prompt = _assemble_seedance_prompt(cr, mock_job)
+        assert "【视频总览】" in prompt
+        assert "【逐镜头时间轴】" in prompt
+        assert "【硬性约束】" in prompt
+        assert "【负面提示词】" in prompt
+        assert "0-15秒" in prompt
 
 
 # ── 端到端流水线集成测试 ────────────────────────────────────────────────
 
 
 class TestPipelineIntegration:
-    """流水线端到端集成测试（mock 外部依赖）。"""
+    """v1.6 流水线端到端集成测试（mock 外部依赖）：TTS+单次 Seedance+上传。"""
 
     @patch("apps.worker.worker_app.tasks.viral_video._step_upload")
     @patch("apps.worker.worker_app.tasks.viral_video._step_render")
-    @patch("apps.worker.worker_app.tasks.viral_video._step_bgm_select")
+    @patch("apps.worker.worker_app.tasks.viral_video._upload_tts_to_oss")
     @patch("apps.worker.worker_app.tasks.viral_video._step_tts")
     @patch("apps.worker.worker_app.tasks.viral_video._step_review")
-    @patch("apps.worker.worker_app.tasks.viral_video._step_storyboard")
-    @patch("apps.worker.worker_app.tasks.viral_video._step_copy_fusion")
+    @patch("apps.worker.worker_app.tasks.viral_video._step_script_generation")
     @patch("apps.worker.worker_app.tasks.viral_video._step_intent_parsing")
     @patch("apps.worker.worker_app.tasks.viral_video._step_video_analysis")
     @patch("apps.worker.worker_app.tasks.viral_video._step_image_analysis")
@@ -474,38 +519,46 @@ class TestPipelineIntegration:
         mock_img_analysis,
         mock_video_analysis,
         mock_intent,
-        mock_copy_fusion,
-        mock_storyboard,
+        mock_script,
         mock_review,
         mock_tts,
-        mock_bgm,
+        mock_tts_upload,
         mock_render,
         mock_upload,
     ):
-        """测试 resume 流水线能从确认状态走到完成。"""
+        """v1.6: TTS整段合成 → 上传TTS到OSS → 单次 Seedance → 上传成片。"""
         from apps.worker.worker_app.tasks.viral_video import (
             resume_viral_video_pipeline,
         )
 
-        # 构造 mock job
         job = ViralVideoJob(
             user_id="user-001",
             images=["https://img.com/1.jpg"],
             industry="美妆",
             status=ViralVideoStatus.RUNNING,
             intent_result={"intent": "推广"},
+            duration=15,
+            video_ratio="9:16",
         )
 
         mock_repo = MagicMock()
         mock_session = MagicMock()
         mock_get_repo.return_value = (mock_session, mock_repo, job)
 
-        # 设置各步骤返回值
-        mock_copy_fusion.return_value = "融合文案"
-        mock_storyboard.return_value = [{"order": 0, "duration": 10}]
+        # v1.6: 如果没有 copy_result 会现场补生成
+        mock_intent.return_value = {"intent": "推广", "key_messages": [], "tone": "亲切"}
+        mock_script.return_value = {
+            "overview": {"theme": "口红", "total_duration": 15, "aspect_ratio": "9:16"},
+            "scene_and_lighting": "明亮化妆台",
+            "shots": [],
+            "hard_constraints": [],
+            "negative_prompts": [],
+            "voiceover_script": "大家好，分享一款口红。",
+            "final_copy": "大家好，分享一款口红。",
+        }
         mock_review.return_value = {"passed": True, "score": 90}
-        mock_tts.return_value = None  # P1: TTS 返回 Path|None，mock 用 None 跳过混音
-        mock_bgm.return_value = None  # P1: BGM 未就绪前返回 None
+        mock_tts.return_value = None  # TTS 失败也能走下去（Seedance generate_audio=True 会自己合成音效）
+        mock_tts_upload.return_value = None
         mock_render.return_value = "/tmp/video.mp4"
         mock_upload.return_value = "https://oss.example.com/final.mp4"
 

tests/unit/test_viral_video_p0.py

@@ -67,49 +67,70 @@ class TestImageAnalysisField:
 # ── P0-1: storyboard 规范化 ────────────────────────────────────────
 
 
-class TestStoryboardNormalize:
-    def test_normalize_fills_defaults(self):
-        from apps.worker.worker_app.tasks.viral_video import _normalize_storyboard
+class TestScriptGenerationV16:
+    """v1.6 编导分镜脚本生成相关纯函数测试。"""
 
-        raw = [{"order": 0, "description": "镜头一"}]
-        out = _normalize_storyboard(raw, total_duration=10, n_segments=1, copy_text="文案")
-        assert len(out) == 1
-        assert out[0]["duration"] >= 3
-        assert out[0]["ken_burns"] in {"zoom_in", "zoom_out", "pan_left", "pan_right", "static"}
-        assert out[0]["type"] == "product_shot"
-        assert out[0]["text"] == ""
+    def test_fallback_script_has_required_fields(self, mock_job):
+        from apps.worker.worker_app.tasks.viral_video import _fallback_script
 
-    def test_normalize_scales_to_total_duration(self):
-        from apps.worker.worker_app.tasks.viral_video import _normalize_storyboard
+        out = _fallback_script(mock_job)
+        assert isinstance(out, dict)
+        assert "overview" in out
+        assert "shots" in out
+        assert "voiceover_script" in out
+        assert "hard_constraints" in out
+        assert "negative_prompts" in out
+        assert out["overview"]["total_duration"] == mock_job.duration
+        assert out["final_copy"] == out["voiceover_script"]
+        assert len(out["shots"]) >= 1
 
-        raw = [
-            {"order": 0, "duration": 10, "description": "a"},
-            {"order": 1, "duration": 10, "description": "b"},
-        ]
-        out = _normalize_storyboard(raw, total_duration=10, n_segments=2, copy_text="x")
-        total = sum(s["duration"] for s in out)
-        assert total == 10
+    def test_safe_json_loads_parses_fenced_code(self):
+        from apps.worker.worker_app.tasks.viral_video import _safe_json_loads
 
-    def test_fallback_storyboard(self):
-        from apps.worker.worker_app.tasks.viral_video import _fallback_storyboard
+        fenced = '```json\n{"voiceover_script": "你好", "shots": []}\n```'
+        out = _safe_json_loads(fenced)
+        assert out is not None
+        assert out["voiceover_script"] == "你好"
 
-        out = _fallback_storyboard("文案", total_duration=15, n_segments=3)
-        assert len(out) == 3
-        assert sum(s["duration"] for s in out) == 15
-        assert all(s["duration"] >= 3 for s in out)
+    def test_safe_json_loads_handles_none(self):
+        from apps.worker.worker_app.tasks.viral_video import _safe_json_loads
 
-    def test_storyboard_llm_list(self, mock_job):
-        from apps.worker.worker_app.tasks.viral_video import _step_storyboard
+        assert _safe_json_loads(None) is None
+        assert _safe_json_loads("not json") is None
 
-        with patch("packages.shared.ai_service.call_llm") as mock_llm:
-            mock_llm.return_value = [
-                {"order": 0, "description": "产品特写", "duration": 5, "text": "t1"},
-                {"order": 1, "description": "使用场景", "duration": 5, "text": "t2"},
-                {"order": 2, "description": "CTA", "duration": 5, "text": "t3"},
-            ]
-            result = _step_storyboard(mock_job, "文案", {"products": []})
-            assert len(result) == 3
-            assert all("description" in s for s in result)
+    def test_validate_normalize_fills_defaults(self, mock_job):
+        from apps.worker.worker_app.tasks.viral_video import _validate_and_normalize_script
+
+        raw = {"voiceover_script": "你好", "shots": [{"scene_and_dialogue": "测试"}]}
+        out = _validate_and_normalize_script(raw, mock_job)
+        assert out["voiceover_script"] == "你好"
+        assert len(out["shots"]) == 1
+        assert out["shots"][0]["shot_type_angle_movement"]
+        assert out["overview"]["total_duration"] == mock_job.duration
+
+    def test_assemble_seedance_prompt_contains_sections(self, mock_job):
+        from apps.worker.worker_app.tasks.viral_video import _assemble_seedance_prompt
+
+        cr = {
+            "overview": {"theme": "测试", "total_duration": 15, "aspect_ratio": "9:16"},
+            "scene_and_lighting": "明亮",
+            "shots": [
+                {
+                    "time_range": "0-15秒",
+                    "shot_type_angle_movement": "中景",
+                    "scene_and_dialogue": "你好",
+                    "action_details": "展示",
+                    "audio_bgm": "BGM",
+                    "transition": "结束",
+                    "reference_image_index": 0,
+                }
+            ],
+            "hard_constraints": ["无字幕"],
+            "negative_prompts": ["水印"],
+        }
+        p = _assemble_seedance_prompt(cr, mock_job)
+        for key in ("【视频总览】", "【场景与光线】", "【逐镜头时间轴】", "【硬性约束】", "【负面提示词】"):
+            assert key in p
 
 
 # ── P1: TTS 返回 Path|None ────────────────────────────────────────
@@ -150,12 +171,24 @@ class TestTTSPath:
 # ── P1: BGM 跳过 / MuseTalk 无 persona 跳过 ───────────────────────
 
 
-class TestBGMSkip:
-    def test_bgm_returns_none(self, mock_job):
-        from apps.worker.worker_app.tasks.viral_video import _step_bgm_select
+class TestDurationClamp:
+    """v1.6 mark_copy_generated 派生字段 + duration clamp。"""
 
-        mock_job.bgm_preference = "upbeat"
-        assert _step_bgm_select(mock_job) is None
+    def test_mark_copy_generated_derives_fields(self):
+        job = ViralVideoJob(user_id="u1", duration=15)
+        cr = {
+            "overview": {"theme": "x", "total_duration": 15, "aspect_ratio": "9:16"},
+            "scene_and_lighting": "亮",
+            "shots": [{"time_range": "0-15秒", "scene_and_dialogue": "对白"}],
+            "voiceover_script": "你好",
+            "hard_constraints": [],
+            "negative_prompts": [],
+        }
+        job.mark_copy_generated(cr)
+        assert job.copy_result is cr
+        assert job.generated_copy_text == "你好"
+        assert job.storyboard == cr["shots"]
+        assert job.effective_copy_text == "你好"
 
 
 # ── P0-1: call_video_generation 参数构造 ──────────────────────────
@@ -188,23 +221,61 @@ class TestCallVideoGeneration:
             assert kwargs["prompt"] == "测试"
             assert kwargs["image_url"] == "https://img/x.jpg"
             assert kwargs["duration"] == 5
+            assert kwargs["generate_audio"] is True
 
 
 # ── P0-1: _step_render 占位片段生成 ──────────────────────────────
 
 
-class TestPlaceholderClip:
-    def test_make_placeholder_clip(self, tmp_path):
-        import shutil
+class TestCallVideoGenerationV16:
+    """v1.6 call_video_generation 透传 reference_audios/reference_images 等参数到 client。"""
 
-        from apps.worker.worker_app.tasks.viral_video import _make_placeholder_clip, _probe_ok
+    def test_passes_reference_params_to_client(self, tmp_path):
+        from packages.shared.ai_service import call_video_generation
 
-        if not shutil.which("ffmpeg"):
-            pytest.skip("ffmpeg not available")
+        out = tmp_path / "v.mp4"
+        out.write_bytes(b"fake")
+        with patch("packages.shared.ai_service.get_doubao_client") as mock_get:
+            mock_client = MagicMock()
+            mock_client.is_available = True
+            mock_client.video_generation.return_value = str(out)
+            mock_get.return_value = mock_client
+            result = call_video_generation(
+                prompt="测试",
+                image_url="https://img/x.jpg",
+                duration=15,
+                ratio="9:16",
+                reference_images=["https://img/r1.jpg"],
+                reference_audios=["https://oss/tts.mp3"],
+                reference_videos=["https://oss/ref.mp4"],
+                generate_audio=True,
+                model="doubao-seedance-2-5-260628",
+            )
+            assert result == str(out)
+            kwargs = mock_client.video_generation.call_args.kwargs
+            # 首帧模式不传 ratio（Bug #2110）
+            assert "ratio" not in kwargs
+            assert kwargs["image_url"] == "https://img/x.jpg"
+            assert kwargs["reference_audios"] == ["https://oss/tts.mp3"]
+            assert kwargs["reference_images"] == ["https://img/r1.jpg"]
+            assert kwargs["reference_videos"] == ["https://oss/ref.mp4"]
+            assert kwargs["generate_audio"] is True
+            assert kwargs["model"] == "doubao-seedance-2-5-260628"
 
-        out = _make_placeholder_clip(tmp_path, 0, 3)
-        assert out.exists()
-        assert _probe_ok(str(out))
+    def test_ratio_passed_when_no_image(self, tmp_path):
+        from packages.shared.ai_service import call_video_generation
+
+        out = tmp_path / "v.mp4"
+        out.write_bytes(b"fake")
+        with patch("packages.shared.ai_service.get_doubao_client") as mock_get:
+            mock_client = MagicMock()
+            mock_client.is_available = True
+            mock_client.video_generation.return_value = str(out)
+            mock_get.return_value = mock_client
+            call_video_generation(prompt="测试", duration=10, ratio="16:9")
+            kwargs = mock_client.video_generation.call_args.kwargs
+            assert kwargs["ratio"] == "16:9"
+            assert kwargs["image_url"] is None
 
 
 # ── P0-1: DoubaoClient.video_generation 在不可用时返回 None ───────

tests/unit/test_viral_video_routes.py

@@ -34,7 +34,7 @@ def _make_job(job_id: str = "job-1", user_id: str = "u1", status: str = "pending
         "viral_structure": "",
         "marketing_purpose": "",
         "bgm_preference": "",
-        "duration": 30,
+        "duration": 15,
         "user_copy_text": "",
         "fusion_level": "ai_polish",
         "reference_audio_path": "",
@@ -53,9 +53,18 @@ def _make_job(job_id: str = "job-1", user_id: str = "u1", status: str = "pending
         "intent_result": None,
         "image_analysis": None,
         "storyboard": None,
+        "copy_result": None,
         "generated_copy_text": "",
+        "voice_id": "",
+        "voice_source": "",
+        "voice_mode": "global",
+        "video_ratio": "9:16",
+        "video_model": "",
         "credits_cost": 0,
         "updated_at": None,
+        "is_terminal": False,
+        "effective_copy_text": "",
+        "voiceover_script": "",
     }.items():
         setattr(job, k, kwargs.pop(k, v))
     return job
@@ -277,6 +286,71 @@ class TestGenerateCopy:
                 vv_mod.generate_copy("job-gc2", GenerateCopyRequest(), authenticated_user=user, session=session)
             assert exc.value.status_code == 409
 
+    def test_generate_copy_persists_voice_and_ratio(self):
+        """generate-copy 应把 voice_id/voice_source/video_ratio 写入 job。"""
+        from unittest.mock import patch
+
+        from app.api.routes import viral_video as vv_mod
+        from app.schemas.viral_video import GenerateCopyRequest
+
+        from packages.domain.viral_video import ViralVideoStatus
+
+        user = _auth_user("u1")
+        session = MagicMock()
+        job = _make_job(job_id="job-gc3", user_id="u1", status=ViralVideoStatus.IMAGE_ANALYZED)
+        repo = MagicMock()
+        repo.get.return_value = job
+        req = GenerateCopyRequest(
+            voice_id="cosy_voice_001",
+            voice_source="library",
+            video_ratio="16:9",
+        )
+        with (
+            patch.object(vv_mod, "_get_job_repo", return_value=repo),
+            patch.object(vv_mod.celery_app, "send_task"),
+        ):
+            vv_mod.generate_copy("job-gc3", req, authenticated_user=user, session=session)
+        assert job.voice_id == "cosy_voice_001"
+        assert job.voice_source == "library"
+        assert job.video_ratio == "16:9"
+
+
+class TestAnalyzeImagesPersist:
+    def test_analyze_images_persists_voice_and_ratio(self):
+        """analyze-images 创建任务时应带上 voice/video_ratio 字段。"""
+        from unittest.mock import patch
+
+        from app.api.routes import viral_video as vv_mod
+        from app.schemas.viral_video import AnalyzeImagesRequest
+
+        user = _auth_user("u1")
+        session = MagicMock()
+        saved = {}
+
+        class FakeRepo:
+            def save(self, job):
+                saved["job"] = job
+
+            def get(self, jid):
+                return None
+
+        req = AnalyzeImagesRequest(
+            images=["img-1"],
+            voice_id="preset_v1",
+            voice_source="preset",
+            video_ratio="1:1",
+        )
+        with (
+            patch.object(vv_mod, "_get_job_repo", return_value=FakeRepo()),
+            patch.object(vv_mod.celery_app, "send_task"),
+        ):
+            resp = vv_mod.analyze_images(req, authenticated_user=user, session=session)
+        job = saved["job"]
+        assert job.voice_id == "preset_v1"
+        assert job.voice_source == "preset"
+        assert job.video_ratio == "1:1"
+        assert resp.images == ["img-1"]
+
 
 class TestConfirmCopy:
     def test_confirm_copy_dispatches_render(self):