Merge pull request 'fix: 脚本超时优化 + 主题智能匹配' (#2233) from fix/script-generation-timeout-and-theme into develop

问题A - 超时优化:
1. DoubaoClient httpx timeout 改为显式 Timeout(connect=10, read=T, write=10, pool=5)
2. intent_parsing 单次超时 60s→20s
3. copy_review(reviewer) 添加 timeout=25s
4. 脚本 fast_timeout 150→90s, pro_timeout 150→60s
5. _step_intent_parsing 加 60s 总 deadline
6. _step_script_generation 加 180s 总 deadline
7. lite/fallback 模型与 primary 相同时跳过重复调用

问题B - 主题匹配:
1. 新增 _determine_theme() 根据图片类型+营销目的智能推断主题
2. _fallback_script 动态主题+对应口播文案
3. _empty_copy_result/_validate_and_normalize/_script_from_xml 去除硬编码好物分享
4. intent_parsing prompt 注入 marketing_purpose 和 image_category_hint
5. storyboard prompt 增加主题匹配指导

分支: fix/script-generation-timeout-and-theme

alembic/versions/099_ai_model_router_seed.py

@@ -0,0 +1,222 @@
+# -*- coding: utf-8 -*-
+"""099: AI 模型路由层 seed — 补齐缺失模型和能力配置.
+
+幂等：所有 INSERT 先检查存在性。
+- ai_models: 补齐 qwen3.7-plus, seedream, seedance, embedding, wan3.0 等
+- ai_capability_configs: 补齐 image_generation, video_generation, embedding
+- 更新已有 capability 的 lite_model_id
+"""
+
+import json
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "099_ai_model_router_seed"
+down_revision = "098_viral_video_image_analysis_v5"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+
+    # CI 环境下 ai_models 表可能尚未创建（由 ORM 自动建表，非 migration）
+    # 如果表不存在则跳过 seed，由应用启动时 ORM 建表后首次访问时生效
+    table_check = conn.execute(sa.text("SELECT to_regclass('public.ai_models')")).scalar()
+    if not table_check:
+        # ai_models 表不存在，跳过所有 seed（CI 环境）
+        return
+
+    # ── 1. 补齐 ai_models 缺失记录 ────────────────────────────────────────────
+    existing_models = {
+        row[0]
+        for row in conn.execute(
+            sa.text("SELECT model_key FROM ai_models WHERE deleted_at IS NULL")
+        ).fetchall()
+    }
+
+    # 从已有 active 记录获取 API key（复用，不硬编码）
+    dashscope_key_row = conn.execute(
+        sa.text(
+            "SELECT api_key FROM ai_models WHERE provider='dashscope' AND deleted_at IS NULL AND api_key IS NOT NULL AND api_key != '' LIMIT 1"
+        )
+    ).first()
+    dashscope_key = dashscope_key_row[0] if dashscope_key_row else ""
+
+    volcengine_key_row = conn.execute(
+        sa.text(
+            "SELECT api_key FROM ai_models WHERE provider='volcengine' AND deleted_at IS NULL AND api_key IS NOT NULL AND api_key != '' LIMIT 1"
+        )
+    ).first()
+    volcengine_key = volcengine_key_row[0] if volcengine_key_row else ""
+
+    new_models = [
+        {
+            "model_key": "qwen3.7-plus",
+            "name": "通义千问3.7 Plus（VLM 兜底）",
+            "provider": "dashscope",
+            "api_key": dashscope_key,
+            "api_base": "https://dashscope.aliyuncs.com/compatible-mode/v1",
+            "description": "阿里云百炼 Qwen3.7 Plus 多模态模型，用于 VLM 兜底分析",
+        },
+        {
+            "model_key": "doubao-seedream-5-0-flash-260915",
+            "name": "Seedream 5.0 Flash（图片生成）",
+            "provider": "volcengine",
+            "api_key": volcengine_key,
+            "api_base": "https://ark.cn-beijing.volces.com/api/v3",
+            "description": "火山引擎 Seedream 5.0 Flash 文生图模型",
+        },
+        {
+            "model_key": "doubao-seedance-2-5-260628",
+            "name": "Seedance 2.5（视频生成）",
+            "provider": "volcengine",
+            "api_key": volcengine_key,
+            "api_base": "https://ark.cn-beijing.volces.com/api/v3",
+            "description": "火山引擎 Seedance 2.5 图/文生视频模型",
+        },
+        {
+            "model_key": "doubao-embedding-vision-251215",
+            "name": "豆包多模态向量嵌入",
+            "provider": "volcengine",
+            "api_key": volcengine_key,
+            "api_base": "https://ark.cn-beijing.volces.com/api/v3",
+            "description": "火山引擎豆包多模态向量嵌入模型",
+        },
+        {
+            "model_key": "wan3.0-video",
+            "name": "Wan 3.0 视频生成",
+            "provider": "dashscope",
+            "api_key": dashscope_key,
+            "api_base": "https://dashscope.aliyuncs.com/api/v1",
+            "description": "阿里云百炼 Wan 3.0 视频生成模型",
+        },
+        {
+            "model_key": "doubao-seed-2-1-pro-260915",
+            "name": "豆包 Seed 2.1 Pro（高精度推理）",
+            "provider": "volcengine",
+            "api_key": volcengine_key,
+            "api_base": "https://ark.cn-beijing.volces.com/api/v3",
+            "description": "火山引擎豆包 Seed 2.1 Pro 深度思考+多模态",
+        },
+    ]
+
+    for m in new_models:
+        if m["model_key"] not in existing_models:
+            conn.execute(
+                sa.text(
+                    """
+                    INSERT INTO ai_models (id, name, provider, model_key, api_key, api_base, description, status, is_default, usage_today, created_at, updated_at)
+                    VALUES (gen_random_uuid()::text, :name, :provider, :model_key, :api_key, :api_base, :description, 'active', false, 0, now(), now())
+                    """
+                ),
+                m,
+            )
+
+    # ── 2. 补齐 ai_capability_configs 缺失项 ──────────────────────────────────
+    cap_table_check = conn.execute(sa.text("SELECT to_regclass('public.ai_capability_configs')")).scalar()
+    if not cap_table_check:
+        return
+
+    existing_caps = {
+        row[0]
+        for row in conn.execute(
+            sa.text("SELECT capability_key FROM ai_capability_configs")
+        ).fetchall()
+    }
+
+    def _get_model_id(model_key: str) -> str | None:
+        row = conn.execute(
+            sa.text(
+                "SELECT id FROM ai_models WHERE model_key = :key AND deleted_at IS NULL AND status = 'active' LIMIT 1"
+            ),
+            {"key": model_key},
+        ).first()
+        return row[0] if row else None
+
+    # image_generation
+    if "image_generation" not in existing_caps:
+        mid = _get_model_id("doubao-seedream-5-0-flash-260915")
+        if mid:
+            conn.execute(
+                sa.text(
+                    """
+                    INSERT INTO ai_capability_configs (id, capability_key, capability_name, primary_model_id, timeout_seconds, max_retries, concurrency, extra_params, is_enabled, created_at, updated_at)
+                    VALUES (gen_random_uuid()::text, :ck, :cn, :pm, 60, 1, 2, :ep, true, now(), now())
+                    """
+                ),
+                {
+                    "ck": "image_generation",
+                    "cn": "图片生成（Seedream）",
+                    "pm": mid,
+                    "ep": json.dumps({"size": "1K"}),
+                },
+            )
+
+    # video_generation
+    if "video_generation" not in existing_caps:
+        mid = _get_model_id("doubao-seedance-2-5-260628")
+        fb_mid = _get_model_id("wan3.0-video")
+        if mid:
+            conn.execute(
+                sa.text(
+                    """
+                    INSERT INTO ai_capability_configs (id, capability_key, capability_name, primary_model_id, fallback_model_id, timeout_seconds, max_retries, concurrency, extra_params, is_enabled, created_at, updated_at)
+                    VALUES (gen_random_uuid()::text, :ck, :cn, :pm, :fm, 600, 1, 1, :ep, true, now(), now())
+                    """
+                ),
+                {
+                    "ck": "video_generation",
+                    "cn": "视频生成（Seedance/Wan）",
+                    "pm": mid,
+                    "fm": fb_mid,
+                    "ep": json.dumps({}),
+                },
+            )
+
+    # embedding
+    if "embedding" not in existing_caps:
+        mid = _get_model_id("doubao-embedding-vision-251215")
+        if mid:
+            conn.execute(
+                sa.text(
+                    """
+                    INSERT INTO ai_capability_configs (id, capability_key, capability_name, primary_model_id, timeout_seconds, max_retries, concurrency, extra_params, is_enabled, created_at, updated_at)
+                    VALUES (gen_random_uuid()::text, :ck, :cn, :pm, 30, 2, 5, :ep, true, now(), now())
+                    """
+                ),
+                {
+                    "ck": "embedding",
+                    "cn": "向量嵌入",
+                    "pm": mid,
+                    "ep": json.dumps({}),
+                },
+            )
+
+    # ── 3. 更新 image_analysis 的 lite_model_id ─────────────────────────────
+    lite_model_id = _get_model_id("qwen3.8-flash")
+    if lite_model_id:
+        conn.execute(
+            sa.text(
+                "UPDATE ai_capability_configs SET lite_model_id = :lite WHERE capability_key = 'image_analysis' AND lite_model_id IS NULL"
+            ),
+            {"lite": lite_model_id},
+        )
+
+
+def downgrade() -> None:
+    conn = op.get_bind()
+    # 安全检查表是否存在
+    table_check = conn.execute(sa.text("SELECT to_regclass('public.ai_models')")).scalar()
+    if not table_check:
+        return
+    conn.execute(
+        sa.text("DELETE FROM ai_capability_configs WHERE capability_key IN ('image_generation', 'video_generation', 'embedding')")
+    )
+    conn.execute(
+        sa.text(
+            "DELETE FROM ai_models WHERE model_key IN ('qwen3.7-plus', 'doubao-seedream-5-0-flash-260915', 'doubao-seedance-2-5-260628', 'doubao-embedding-vision-251215', 'wan3.0-video', 'doubao-seed-2-1-pro-260915') AND deleted_at IS NULL"
+        )
+    )

alembic/versions/100_fix_capability_model_bindings.py

@@ -0,0 +1,107 @@
+# -*- coding: utf-8 -*-
+"""100: 修正已有 capability 的模型绑定.
+
+幂等：仅当 primary_model_id 当前绑定到旧模型 (doubao-seed-1-6) 时才更新，
+避免覆盖用户在后台的自定义配置。
+
+- 更新 5 个 LLM capability (intent_parsing, copy_fusion, storyboard, copy_review, asset_classify)
+  的 primary_model_id 从 doubao-seed-1-6 改为 doubao-seed-2-1-pro-260915
+- 更新 image_analysis 的 primary/lite/fallback 模型绑定
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "100_fix_capability_model_bindings"
+down_revision = "099_ai_model_router_seed"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+
+    # Check tables exist
+    table_check = conn.execute(sa.text("SELECT to_regclass('public.ai_models')")).scalar()
+    if not table_check:
+        return
+
+    config_table_check = conn.execute(sa.text("SELECT to_regclass('public.ai_capability_configs')")).scalar()
+    if not config_table_check:
+        return
+
+    # Look up model IDs by model_key (not hardcoded UUIDs)
+    pro_model_row = conn.execute(
+        sa.text(
+            "SELECT id FROM ai_models WHERE model_key = 'doubao-seed-2-1-pro-260915' AND deleted_at IS NULL LIMIT 1"
+        )
+    ).first()
+    if not pro_model_row:
+        return
+    pro_model_id = pro_model_row[0]
+
+    old_model_row = conn.execute(
+        sa.text("SELECT id FROM ai_models WHERE model_key = 'doubao-seed-1-6-250615' LIMIT 1")
+    ).first()
+    old_model_id = old_model_row[0] if old_model_row else None
+
+    llm_capabilities = [
+        "intent_parsing",
+        "copy_fusion",
+        "storyboard",
+        "copy_review",
+        "asset_classify",
+    ]
+
+    for cap_key in llm_capabilities:
+        if old_model_id:
+            conn.execute(
+                sa.text(
+                    "UPDATE ai_capability_configs SET primary_model_id = :new_id, updated_at = NOW() "
+                    "WHERE capability_key = :cap_key AND primary_model_id = :old_id"
+                ),
+                {"new_id": pro_model_id, "old_id": old_model_id, "cap_key": cap_key},
+            )
+
+    # Update image_analysis
+    qwen38_row = conn.execute(
+        sa.text("SELECT id FROM ai_models WHERE model_key = 'qwen3.8-flash' AND deleted_at IS NULL LIMIT 1")
+    ).first()
+    qwen37_row = conn.execute(
+        sa.text("SELECT id FROM ai_models WHERE model_key = 'qwen3.7-plus' AND deleted_at IS NULL LIMIT 1")
+    ).first()
+
+    if qwen38_row and qwen37_row:
+        qwen38_id = qwen38_row[0]
+        qwen37_id = qwen37_row[0]
+
+        current_ia = conn.execute(
+            sa.text(
+                "SELECT primary_model_id, lite_model_id, fallback_model_id "
+                "FROM ai_capability_configs WHERE capability_key = 'image_analysis'"
+            )
+        ).first()
+
+        if current_ia:
+            current_primary, current_lite, current_fallback = current_ia
+            updates = {}
+            if current_primary != qwen38_id:
+                updates["primary_model_id"] = qwen38_id
+            if current_lite != qwen38_id:
+                updates["lite_model_id"] = qwen38_id
+            if current_fallback != qwen37_id:
+                updates["fallback_model_id"] = qwen37_id
+
+            if updates:
+                set_clause = ", ".join([f"{k} = :{k}" for k in updates.keys()])
+                set_clause += ", updated_at = NOW()"
+                updates["cap_key"] = "image_analysis"
+                conn.execute(
+                    sa.text(f"UPDATE ai_capability_configs SET {set_clause} WHERE capability_key = :cap_key"),
+                    updates,
+                )
+
+
+def downgrade() -> None:
+    pass

alembic/versions/101_qwen_vl_plus_and_max_tokens.py

@@ -0,0 +1,153 @@
+# -*- coding: utf-8 -*-
+"""101: 补齐 qwen-vl-plus 视觉模型并修正 image_analysis 绑定与 max_tokens.
+
+背景：
+- qwen-vl-plus 做图片识别时返回 JSON 约 500-600 tokens，旧硬编码
+  max_tokens=350 导致 JSON 被截断、解析失败返回"未识别"。
+- 代码侧已移除硬编码，改由 capability 的 DB 配置决定 max_tokens。
+
+幂等：
+- qwen-vl-plus 已存在则不插入；
+- 仅当 image_analysis 当前 primary_model 不是 qwen-vl-plus 时才更新绑定，
+  避免覆盖后台手动配置。
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "101_qwen_vl_plus_and_max_tokens"
+down_revision = "100_fix_capability_model_bindings"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+
+    models_table = conn.execute(sa.text("SELECT to_regclass('public.ai_models')")).scalar()
+    if not models_table:
+        return
+
+    caps_table = conn.execute(sa.text("SELECT to_regclass('public.ai_capability_configs')")).scalar()
+    if not caps_table:
+        return
+
+    # ── c. 补全其他 capability 的 max_tokens 默认值（幂等）──────────────────
+    # 放在 image_analysis 特定逻辑之前，确保任何分支 return 都不会跳过本段。
+    # 仅在当前值为 NULL 或过小 (<100) 时更新，不覆盖已有合理配置。
+    # embedding / tts / voice_clone 不走 chat 接口，无需设置。
+    default_max_tokens = {
+        "intent_parsing": 500,
+        "copy_fusion": 2500,
+        "storyboard": 4000,
+        "copy_review": 1000,
+        "asset_classify": 500,
+        "image_generation": 500,
+        "video_generation": 500,
+    }
+    for cap_key, mt in default_max_tokens.items():
+        conn.execute(
+            sa.text(
+                "UPDATE ai_capability_configs "
+                "SET max_tokens = :mt, updated_at = now() "
+                "WHERE capability_key = :key "
+                "AND (max_tokens IS NULL OR max_tokens < 100)"
+            ),
+            {"mt": mt, "key": cap_key},
+        )
+
+    # ── a. 确保 qwen-vl-plus 模型存在 ────────────────────────────────────────
+    conn.execute(sa.text("""
+            INSERT INTO ai_models (id, name, provider, model_key, api_key, api_base,
+                                   description, status, is_default, usage_today,
+                                   created_at, updated_at)
+            SELECT gen_random_uuid()::text,
+                   '通义千问VL Plus',
+                   'dashscope',
+                   'qwen-vl-plus',
+                   COALESCE(
+                       (SELECT api_key FROM ai_models
+                        WHERE provider = 'dashscope' AND deleted_at IS NULL
+                          AND api_key IS NOT NULL AND api_key != ''
+                        LIMIT 1),
+                       ''
+                   ),
+                   'https://dashscope.aliyuncs.com/compatible-mode/v1',
+                   '阿里云视觉理解模型（图片识别/分析）',
+                   'active', false, 0, now(), now()
+            WHERE NOT EXISTS (
+                SELECT 1 FROM ai_models
+                WHERE model_key = 'qwen-vl-plus' AND deleted_at IS NULL
+            )
+            """))
+
+    qwen_vl_row = conn.execute(
+        sa.text(
+            "SELECT id FROM ai_models WHERE model_key = 'qwen-vl-plus' "
+            "AND deleted_at IS NULL AND status = 'active' LIMIT 1"
+        )
+    ).first()
+    if not qwen_vl_row:
+        return
+    qwen_vl_id = qwen_vl_row[0]
+
+    qwen37_row = conn.execute(
+        sa.text(
+            "SELECT id FROM ai_models WHERE model_key = 'qwen3.7-plus' "
+            "AND deleted_at IS NULL AND status = 'active' LIMIT 1"
+        )
+    ).first()
+    qwen37_id = qwen37_row[0] if qwen37_row else None
+
+    # ── b. 仅当当前 primary 不是 qwen-vl-plus 时修正绑定与 max_tokens ───────
+    current = conn.execute(
+        sa.text(
+            "SELECT primary_model_id, lite_model_id, fallback_model_id, max_tokens "
+            "FROM ai_capability_configs WHERE capability_key = 'image_analysis'"
+        )
+    ).first()
+
+    if current is None:
+        # capability 不存在则创建
+        conn.execute(
+            sa.text("""
+                INSERT INTO ai_capability_configs
+                    (id, capability_key, capability_name, primary_model_id,
+                     lite_model_id, fallback_model_id, timeout_seconds,
+                     max_retries, max_tokens, concurrency, extra_params,
+                     is_enabled, created_at, updated_at)
+                VALUES (gen_random_uuid()::text, 'image_analysis', '图片分析',
+                        :primary, :primary, :fallback, 30, 1, 1000, 2,
+                        '{}'::jsonb, true, now(), now())
+                """),
+            {"primary": qwen_vl_id, "fallback": qwen37_id},
+        )
+        return
+
+    current_primary = current[0]
+    if current_primary == qwen_vl_id:
+        # 已经绑定 qwen-vl-plus：视为后台/数据迁移已处理，不覆盖任何配置
+        return
+
+    set_parts = [
+        "primary_model_id = :vl_id",
+        "lite_model_id = :vl_id",
+        "max_tokens = 1000",
+        "updated_at = now()",
+    ]
+    params: dict = {"vl_id": qwen_vl_id}
+    if qwen37_id is not None:
+        set_parts.insert(2, "fallback_model_id = :qwen37_id")
+        params["qwen37_id"] = qwen37_id
+
+    conn.execute(
+        sa.text(
+            "UPDATE ai_capability_configs SET " + ", ".join(set_parts) + " WHERE capability_key = 'image_analysis'"
+        ),
+        params,
+    )
+
+
+def downgrade() -> None:
+    pass

alembic/versions/102_image_analysis_max_tokens_1500.py

@@ -0,0 +1,36 @@
+# -*- coding: utf-8 -*-
+"""102: image_analysis max_tokens 1200 -> 1500.
+
+v6 prompt 更长、字段更多，旧 max_tokens 容易截断 JSON。
+仅在 image_analysis 当前 max_tokens < 1500 时更新（幂等，不覆盖后台已调到 >=1500 的配置）。
+"""
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision = "102_image_analysis_max_tokens_1500"
+down_revision = "101_qwen_vl_plus_and_max_tokens"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+
+    caps_table = conn.execute(sa.text("SELECT to_regclass('public.ai_capability_configs')")).scalar()
+    if not caps_table:
+        return
+
+    conn.execute(
+        sa.text(
+            "UPDATE ai_capability_configs "
+            "SET max_tokens = 1500, updated_at = now() "
+            "WHERE capability_key = 'image_analysis' "
+            "AND (max_tokens IS NULL OR max_tokens < 1500)"
+        )
+    )
+
+
+def downgrade() -> None:
+    pass

alembic/versions/103_v7_prompt_and_tokens_3000.py

@@ -0,0 +1,194 @@
+# -*- coding: utf-8 -*-
+"""image_analysis v7 prompt + max_tokens 3000 + max_retries 3
+
+Revision ID: 103_v7_prompt_and_tokens_3000
+Revises: 102_image_analysis_max_tokens_1500
+Create Date: 2026-10-07
+
+变更：
+1. 插入v7精简prompt（~1KB，v6 ~4.5KB，删除few-shot/冗长规则，减少输出token占用），设为active
+2. v6停用（is_active=False），保留历史
+3. image_analysis capability: max_tokens 1500→3000，max_retries 1→3
+
+ai_capability_configs 由应用 create_all 创建，全新 alembic-only 库可能不存在，
+故第3步做 to_regclass 守卫（同 102）。
+"""
+
+from sqlalchemy import text
+
+from alembic import op
+
+revision = "103_v7_prompt_and_tokens_3000"
+down_revision = "102_image_analysis_max_tokens_1500"
+branch_labels = None
+depends_on = None
+
+V7_SYSTEM = """# 角色
+你是一位专业的图片分析师，擅长准确识别图片中的场景、人物、物体、文字、氛围。
+
+# 任务
+对用户上传的图片逐张分析，描述你看到的内容，输出JSON格式。
+
+## 技能
+
+### 技能1：判断图片类型
+判断图片属于哪种类型，type字段填对应的英文值：
+- 商品图（product）：单个或多个商品、产品包装
+- 门店场景图（store）：店铺内部、门头招牌、货架陈列
+- 人物图（person）：人物形象、穿搭造型、肖像照片
+- 风景图（scene）：风景、动物、美食、街景
+- 其他（other）：以上都不是
+
+### 技能2：描述通用信息
+不管什么图都要描述：
+- type：图片类型，填product/store/person/scene/other其中一个
+- scene：一句话描述场景，例如"理疗养生店内部，摆着多张理疗床和产品货架"
+- mood：整体氛围，2-4个词，例如"整洁专业"、"热闹温馨"
+- colors：主要颜色，最多5个，写具体颜色名（亮红色/米白色/深蓝色，不写笼统的红色蓝色）
+- visible_text：图片里看到的文字，说明什么字、在什么位置，最多5条；没看到就空数组
+- lighting：光线情况，例如"明亮柔光"、"自然光"、"室内暖黄灯"
+- composition：怎么拍的，例如"居中特写"、"中景平视"、"俯拍"
+- has_person：有没有人，true或false
+
+### 技能3：描述门店场景
+如果是门店场景图（type="store"），还要描述：
+- store_type：什么类型的店，例如"养生馆"、"便利店"、"餐饮店"、"母婴店"
+- brand_signage：招牌上写了什么字、有什么品牌标识
+- visual_elements：看到哪些显眼的东西（招牌样式、灯光、货架、商品陈列、海报、收银台等），最多8个
+- product_categories：看到哪些品类的商品，例如"饮料零食"、"养生产品"
+- promotion_elements：有没有促销活动（打折海报、满减吊旗等），没有就空数组
+- atmosphere：店内什么氛围，例如"亲民生活化"、"老字号专业感"
+- cleanliness：店内干净程度，例如"干净整洁"、"货架整齐"
+- 看到顾客或店员要描述他们在做什么，has_person填true
+
+### 技能4：描述商品
+如果是商品图（type="product"），逐个商品描述：
+- product_name：商品名称，尽量具体，例如"OMO奥妙除菌除螨洗衣液"；看不出来填null
+- brand：什么牌子，看不出来填null
+- category：类目，从以下选一个：服饰鞋包/美妆/数码/食品/家居清洁/母婴/配饰/其他
+- package_type：什么包装，例如"瓶装"、"盒装"、"罐装"、"袋装"、"多瓶装"
+- package_color：包装主要颜色，写具体色（亮红色不写红色）
+- body_shape：瓶身或包装形状，例如"圆润胖瓶"、"竖款带把手瓶身"
+- label_design：标签设计，例如"红色标签印白色品牌logo"
+- key_text_on_package：包装上最显眼的文字（品牌名、功能词、卖点词），最多5个
+- product_features：包装特征，3-6个短语，包含颜色、瓶盖、形状、标签图案
+- key_selling_points：核心卖点，1-3个短语
+
+### 技能5：描述人物
+如果是人物图（type="person"），描述：
+- person_count：几个人
+- gender：性别（男/女/无法判断）
+- age_range：年龄段（儿童/青少年/青年/中年/老年/无法判断）
+- outfit_style：穿搭风格，例如"休闲日常"、"通勤商务"、"街头潮流"
+- upper_wear：上装（颜色+款式+材质），穿裙装不填
+- lower_wear：下装（颜色+款式+版型），穿裙装不填
+- dress_wear：裙装描述，穿上下装不填
+- outerwear：外套
+- shoes：鞋子
+- bag：包袋，没有填null
+- accessories：配饰（眼镜/帽子/项链/耳环/手表/手链/围巾/腰带等），没有填空数组
+- hairstyle：发型
+- makeup：妆容，男生或看不出填null
+- expression：表情，例如"微笑看镜头"、"冷酷无表情"
+- pose：姿势动作，例如"身直立正对镜头"、"单手撩发"
+- body_type：身材，例如"纤细苗条"、"高挑身材"、"丰满匀称"
+- portrait_prompt：80-150字详细描述人物形象（后面用来AI生成肖像图），要写清年龄段、穿搭完整细节、发型发色、妆容、表情、姿势、场景、光线、风格感觉，语言要有画面感
+
+### 技能6：描述风景
+如果是风景图（type="scene"），描述：
+- scene_type：什么场景，例如"自然风景"、"城市街景"、"动物"、"美食"
+- main_subject：画面主体是什么
+- key_elements：关键元素，最多8个
+- environment_objects：周围环境物体，最多8个
+- atmosphere：整体氛围，例如"秋日慵懒氛围感"、"清新自然氧气感"
+- 有人物就描述人物特征
+
+## 限制
+- 只输出JSON，不要任何解释文字，不要markdown代码块包裹，不要写"好的""以下是分析结果"这种废话
+- 颜色写具体色调（亮红色/米白色/深蓝色/翠绿色），不写笼统词汇
+- 瓶身、包装、招牌上的文字尽量识别出来（品牌名、功能词、卖点词）
+- 多个商品、多个人物分开描述，不要合并
+- 看不出来、不确定的字段填null或空数组，布尔值填true/false，绝对不要瞎编
+- 确保JSON格式合法，所有大括号、中括号、引号正确闭合
+- 数组字段控制数量：colors最多5个，visible_text最多5条，visual_elements最多8个，accessories最多10个"""
+V7_USER = "请分析这张图片，按系统消息的JSON结构输出。"
+
+
+def _capability_table_exists(bind) -> bool:
+    return bool(bind.execute(text("SELECT to_regclass('public.ai_capability_configs')")).scalar())
+
+
+def upgrade() -> None:
+    bind = op.get_bind()
+    # 1. 停用旧的active image_analysis prompt（含v6）
+    bind.execute(
+        text(
+            "UPDATE viral_video_prompt_templates SET is_active = FALSE "
+            "WHERE prompt_type = 'image_analysis' AND is_active = TRUE"
+        )
+    )
+    # 2. 幂等插入v7（存在则更新并重新激活）
+    existing = bind.execute(
+        text("SELECT id FROM viral_video_prompt_templates " "WHERE prompt_type = 'image_analysis' AND version = 7")
+    ).fetchone()
+    if existing:
+        bind.execute(
+            text(
+                "UPDATE viral_video_prompt_templates SET is_active = TRUE, "
+                "system_prompt = :sys, user_prompt_template = :usr, "
+                "name = 'v7 精简结构化分析', updated_at = NOW() "
+                "WHERE prompt_type = 'image_analysis' AND version = 7"
+            ),
+            {"sys": V7_SYSTEM, "usr": V7_USER},
+        )
+    else:
+        bind.execute(
+            text(
+                "INSERT INTO viral_video_prompt_templates "
+                "(prompt_type, version, name, system_prompt, user_prompt_template, "
+                "is_active, created_at, updated_at) "
+                "VALUES ('image_analysis', 7, 'v7 精简结构化分析', "
+                ":sys, :usr, TRUE, NOW(), NOW())"
+            ),
+            {"sys": V7_SYSTEM, "usr": V7_USER},
+        )
+    # 3. capability max_tokens=3000、max_retries=3（表不存在则跳过）
+    if _capability_table_exists(bind):
+        bind.execute(
+            text(
+                "UPDATE ai_capability_configs SET max_tokens = 3000, "
+                "updated_at = NOW() "
+                "WHERE capability_key = 'image_analysis' AND "
+                "(max_tokens IS NULL OR max_tokens < 3000)"
+            )
+        )
+        bind.execute(
+            text(
+                "UPDATE ai_capability_configs SET max_retries = 3, updated_at = NOW() "
+                "WHERE capability_key = 'image_analysis' AND "
+                "(max_retries IS NULL OR max_retries < 3)"
+            )
+        )
+
+
+def downgrade() -> None:
+    bind = op.get_bind()
+    # 删除v7
+    bind.execute(
+        text("DELETE FROM viral_video_prompt_templates " "WHERE prompt_type = 'image_analysis' AND version = 7")
+    )
+    # 恢复v6为active
+    bind.execute(
+        text(
+            "UPDATE viral_video_prompt_templates SET is_active = TRUE "
+            "WHERE prompt_type = 'image_analysis' AND version = 6"
+        )
+    )
+    # tokens/retries回退
+    if _capability_table_exists(bind):
+        bind.execute(
+            text(
+                "UPDATE ai_capability_configs SET max_tokens = 1500, max_retries = 1, "
+                "updated_at = NOW() WHERE capability_key = 'image_analysis'"
+            )
+        )

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

@@ -280,11 +280,18 @@ def generate_copy(
         raise HTTPException(status_code=404, detail="任务不存在")
     if job.user_id != authenticated_user.user.id:
         raise HTTPException(status_code=403, detail="无权操作此任务")
-    if job.status not in (ViralVideoStatus.IMAGE_ANALYZED, ViralVideoStatus.PENDING, ViralVideoStatus.FAILED):
+    # 允许首次进入（IMAGE_ANALYZED/PENDING）、失败重试（FAILED）、文案重新生成（COPY_GENERATED/COMPLETED）
+    if job.status not in (
+        ViralVideoStatus.IMAGE_ANALYZED,
+        ViralVideoStatus.PENDING,
+        ViralVideoStatus.FAILED,
+        ViralVideoStatus.COPY_GENERATED,
+        ViralVideoStatus.COMPLETED,
+    ):
         raise HTTPException(status_code=409, detail=f"任务当前状态 {job.status} 不能生成文案")
 
-    # 允许失败任务重试：重置
-    if job.status == ViralVideoStatus.FAILED:
+    # 失败重试 / 重新生成：retry_count 自增
+    if job.status in (ViralVideoStatus.FAILED, ViralVideoStatus.COPY_GENERATED, ViralVideoStatus.COMPLETED):
         job.retry_count += 1
         job.error_msg = ""
 
@@ -341,6 +348,9 @@ def confirm_copy(
         raise HTTPException(status_code=403, detail="无权操作此任务")
     if job.status != ViralVideoStatus.COPY_GENERATED:
         raise HTTPException(status_code=409, detail=f"任务当前状态 {job.status} 不能确认文案（需 copy_generated）")
+    # #2218: 额外校验 copy_result 完整性，防止孤儿/脏数据进入渲染
+    if not isinstance(job.copy_result, dict) or not job.copy_result:
+        raise HTTPException(status_code=409, detail="文案数据缺失，请先点击「生成文案」")
 
     # 积分预扣（已扣过/重试任务跳过）
     from app.config import settings as _settings

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

@@ -1050,10 +1050,13 @@
   flex-direction: column;
   align-items: center;
   justify-content: center;
+  height: 360px;
   padding: 28px 16px;
   gap: 10px;
   background: #fff;
   border: 1px solid #e5e7eb;
+  border-radius: 10px;
+  margin-top: 8px;
 }
 .vv-copy-loading .vv-spinner {
   width: 28px;
@@ -1082,18 +1085,38 @@
 
 /* ── Storyboard (linear doc style) ── */
 .vv-storyboard {
-  padding: 6px 2px;
-  background: transparent;
-  border: none;
+  display: flex;
+  flex-direction: column;
+  height: 360px;
+  padding: 10px 12px;
+  background: #fff;
+  border: 1px solid #e5e7eb;
+  border-radius: 10px;
+  margin-top: 8px;
+  overflow: hidden;
 }
 
 .vv-sb-doc {
+  flex: 1 1 auto;
   display: flex;
   flex-direction: column;
   gap: 3px;
   color: #1f2937;
   font-size: 13px;
   line-height: 1.55;
+  overflow-y: auto;
+  padding-right: 4px;
+  margin-right: -4px;
+}
+.vv-sb-doc::-webkit-scrollbar {
+  width: 6px;
+}
+.vv-sb-doc::-webkit-scrollbar-thumb {
+  background: #d8c4ff;
+  border-radius: 3px;
+}
+.vv-sb-doc::-webkit-scrollbar-track {
+  background: transparent;
 }
 .vv-sb-h {
   margin: 6px 0 2px;
@@ -1383,12 +1406,14 @@
 /* 口播稿 —— 复用 vv-sb-field 样式，无额外需求 */
 
 .vv-sb-actions {
+  flex-shrink: 0;
   display: flex;
   align-items: center;
   gap: 8px;
   margin-top: 8px;
   padding-top: 8px;
   border-top: 1px solid #e5e7eb;
+  background: #fff;
 }
 .vv-sb-actions .vv-btn-ghost {
   padding: 6px 14px;

apps/worker/worker_app/tasks/viral_video.py

@@ -92,22 +92,30 @@ def _save_job(repo, job, session):
     session.commit()
 
 
-def _start_trust_chain_preheat(job_id: str, portrait_descriptions: list[str]) -> None:
-    """#2172/#2174 后台启动信任链预热（Seedream t2i 文生图人像），不阻塞调用方。
+def _start_trust_chain_preheat(job_id: str, products: list[dict]) -> None:
+    """#2172/#2174/#2220 后台启动信任链预热（Seedream t2i 文生图人像），不阻塞调用方。
 
-    #2174 重要：改为 t2i 文生图模式——用 VLM 分析出的人物外貌描述做 prompt，不传 reference_images，
-    产物是方舟信任模型输出，Seedance 直接放行不触发肖像审核。
-    i2i（传用户照片做 reference）产物不被信任，实测仍被 400 portrait_intercept 拦截。
+    #2220 修复：只对 has_person=True 的图（真人照片）生成 AI 人像替换，
+    场景图/商品图/门店图保持原图不变，传给 Seedance 作为 reference_image 直接使用。
 
-    预热成功后把结果写入 job.pre_trusted_images，阶段3 渲染直接使用，省掉串行等待。
-    预热失败静默（pre_trusted_images 保持 None），阶段3 会走 #2166 自动降级纯 t2v。
+    预热结果写入 job.pre_trusted_images：与 products 等长的稀疏列表，
+    人像位是 AI 图 URL，非人像位是 None（表示保留原图）。
     """
-    # 过滤有效描述：非空且不是"无人像"
-    _valid = [
-        d for d in (portrait_descriptions or []) if d and isinstance(d, str) and "无人像" not in d and len(d) >= 10
-    ]
+    # 构建人像位索引映射：person_indices[k] = products中第k个人像的位置
+    _person_indices: list[int] = []
+    _valid: list[str] = []
+    for _i, _p in enumerate(products or []):
+        if not isinstance(_p, dict):
+            continue
+        if not _p.get("has_person", False):
+            continue
+        _d = (_p.get("portrait_prompt") or "").strip()
+        if not _d or "无人像" in _d or len(_d) < 10:
+            continue
+        _person_indices.append(_i)
+        _valid.append(_d)
     if not _valid:
-        logger.info("[trust-chain][preheat] 无有效人物描述（可能是纯商品图），跳过预热 job=%s", job_id)
+        logger.info("[trust-chain][preheat] 无有效人物描述（可能是纯商品/场景图），跳过预热 job=%s", job_id)
         return
     # 判断是否是 doubao provider（DashScope/Wan 不需要信任链）
     try:
@@ -133,21 +141,33 @@ def _start_trust_chain_preheat(job_id: str, portrait_descriptions: list[str]) ->
 
             logger.info("[trust-chain][preheat] 后台t2i预热启动 job=%s n=%d", job_id, len(_valid))
             result = preheat_trust_chain(_valid, timeout=120)
-            if result and len(result) >= 1:
+            if result and len(result) == len(_valid):
                 sess2, repo2, job2 = _get_repo_and_job(job_id)
                 try:
-                    job2.pre_trusted_images = result
+                    # #2220: 构建与 products 等长的稀疏列表，人像位放AI图URL，非人像位None
+                    _imgs2 = job2.images or []
+                    _sparse: list[str | None] = [None] * max(len(_imgs2), len(products or []))
+                    for _k, _url in enumerate(result):
+                        if _k < len(_person_indices):
+                            _sparse[_person_indices[_k]] = _url
+                    job2.pre_trusted_images = _sparse
                     repo2.update(job2)
                     sess2.commit()
                     logger.info(
-                        "[trust-chain][preheat] t2i预热完成并持久化 job=%s n=%d",
+                        "[trust-chain][preheat] t2i预热完成并持久化 job=%s n_person=%d total=%d",
                         job_id,
                         len(result),
+                        len(_sparse),
                     )
                 finally:
                     sess2.close()
             else:
-                logger.info("[trust-chain][preheat] 预热失败 job=%s，阶段3现场跑兜底", job_id)
+                logger.info(
+                    "[trust-chain][preheat] 预热失败或数量不匹配 job=%s got=%s expect=%d，阶段3现场跑兜底",
+                    job_id,
+                    len(result) if result else 0,
+                    len(_valid),
+                )
         except Exception as e:
             logger.warning("[trust-chain][preheat] 预热异常 job=%s err=%s", job_id, e, exc_info=True)
 
@@ -244,8 +264,8 @@ def _recover_stale_jobs() -> int:
             "SET status='failed', error_msg='任务执行超时，请重试', updated_at=:now "
             "WHERE status='running' "
             "  AND started_at IS NOT NULL AND started_at < :cutoff_start "
-            "  AND (heartbeat_at IS NULL OR heartbeat_at < :cutoff_beat) "
-            "  AND (heartbeat_at IS NOT NULL OR updated_at < :cutoff_beat)"
+            "  AND updated_at < :cutoff_beat "
+            "  AND (heartbeat_at IS NULL OR heartbeat_at < :cutoff_beat)"
         )
         result = ssn.execute(sql, {"now": now, "cutoff_start": cutoff_start, "cutoff_beat": cutoff_beat})
         ssn.commit()
@@ -268,6 +288,7 @@ def _recover_stale_jobs() -> int:
 
 _DEFAULT_HARD_CONSTRAINTS = [
     "无字幕、无水印、无任何自动生成文字、无 logo",
+    "严格还原参考图片中的真实场景、门店环境、商品陈列、人物外貌服装特征，不得凭空生成与参考图无关的人物、场景或物品",
     "同一人物全程保持一致的五官、发型、服装、身材，不得换脸或变形",
     "口播语音必须在指定时长内自然念完，语速自然，口型与语音同步",
     "画面流畅无闪烁、无多余肢体、无扭曲变形、无穿模",
@@ -295,9 +316,9 @@ _DEFAULT_NEGATIVE_PROMPTS = [
 ]
 
 
-def _empty_copy_result(duration: int = 15, ratio: str = "9:16") -> dict:
+def _empty_copy_result(duration: int = 15, ratio: str = "9:16", theme: str = "") -> dict:
     return {
-        "overview": {"theme": "好物推荐", "total_duration": duration, "aspect_ratio": ratio},
+        "overview": {"theme": theme or "好物推荐", "total_duration": duration, "aspect_ratio": ratio},
         "scene_and_lighting": "简洁明亮的室内场景，柔和自然光，产品主体清晰",
         "shots": [],
         "hard_constraints": list(_DEFAULT_HARD_CONSTRAINTS),
@@ -323,6 +344,7 @@ def _vision_fallback(idx: int, reason: str, extra: dict | None = None) -> dict:
         "scene": "通用",
         "portrait_prompt": "无人像",
         "summary": "",
+        "has_person": False,
         "_source": reason,
     }
     if extra:
@@ -421,10 +443,14 @@ def _step_intent_parsing(job: ViralVideoJob, image_analysis: dict) -> dict:
             render_system_prompt,
             render_user_prompt,
         )
-        from packages.shared.ai_service import call_llm
+        from packages.shared.ai_router import ai_router
     except ImportError:
         return {"intent": "推广产品", "key_messages": ["产品亮点"], "tone": "专业", "suggested_title": ""}
 
+    _llm_client = ai_router.get_llm_client("intent_parsing")
+    if not _llm_client.is_available:
+        return {"intent": "推广产品", "key_messages": ["产品亮点"], "tone": "专业", "suggested_title": ""}
+
     products_summary = ""
     products = (image_analysis or {}).get("products", []) or []
     for p in products:
@@ -445,11 +471,15 @@ def _step_intent_parsing(job: ViralVideoJob, image_analysis: dict) -> dict:
 
     template = get_template("intent_parsing")
     system = render_system_prompt(template)
+    marketing_purpose = getattr(job, "marketing_purpose", "") or "未指定"
+    image_category_hint = _determine_theme(image_analysis, marketing_purpose)
     user = render_user_prompt(
         template,
         user_copy_text=job.user_copy_text or "（未提供，全由 AI 创作）",
         industry=job.industry or "未指定",
         image_analysis=products_summary or "- （无图片分析结果）",
+        marketing_purpose=marketing_purpose,
+        image_category_hint=image_category_hint,
     )
 
     def _parse(raw: str) -> dict:
@@ -473,19 +503,31 @@ def _step_intent_parsing(job: ViralVideoJob, image_analysis: dict) -> dict:
             "suggested_title": "",
         }
 
-    _s = get_shared_settings()
-    _fast = _s.doubao_fast_model
-    _pro = _s.doubao_model
-    for _m, _lbl in [(_fast, "fast"), (_pro, "pro-fallback")]:
+    # #2220: 直接用 ai_router 获取 client，不再手动提取 model_key
+    from packages.shared.ai_router import ai_router
+
+    _client_fast = ai_router.get_llm_client("intent_parsing", variant="primary")
+    _client_pro = ai_router.get_llm_client("intent_parsing", variant="lite")
+    _intent_deadline = time.time() + 60
+    _seen_models: set[str] = set()
+    for _client, _lbl in [(_client_fast, "fast"), (_client_pro, "pro-fallback")]:
+        if not _client or not _client.is_available:
+            continue
+        if _client.model in _seen_models:
+            logger.info("[爆款视频] 意图解析跳过重复模型 %s label=%s", _client.model, _lbl)
+            continue
+        _seen_models.add(_client.model)
+        if time.time() > _intent_deadline:
+            logger.warning("[爆款视频] 意图解析超过60s总预算，跳过 label=%s", _lbl)
+            break
         try:
-            logger.info("[爆款视频] 意图解析 model=%s label=%s", _m, _lbl)
-            raw = call_llm(
+            logger.info("[爆款视频] 意图解析 model=%s label=%s", _client.model, _lbl)
+            raw = _client.chat_completion(
                 [{"role": "system", "content": system}, {"role": "user", "content": user}],
                 temperature=0.4,
                 max_tokens=1024,
-                model=_m,
-                timeout=60,
-            )  # #2180: 意图解析 LLM 实测需更长响应，原25s太紧
+                timeout=20,
+            )
             if not raw:
                 continue
             parsed = _parse(raw)
@@ -521,6 +563,70 @@ def _persona_style_hint(persona_id: str) -> str:
     return "【人设风格：未指定】亲切自然、像朋友分享好物"
 
 
+def _determine_theme(image_analysis: dict | None, marketing_purpose: str = "") -> str:
+    """根据图片分析结果和营销目的，智能推断默认主题。
+
+    门店类→门店探店/到店体验；商品图→好物分享/产品种草；
+    人物图→穿搭/人物故事；场景图→场景氛围/空间体验。
+    """
+    products = (image_analysis or {}).get("products", []) or []
+    type_counts: dict[str, int] = {}
+    for p in products:
+        if not isinstance(p, dict):
+            continue
+        cat = (p.get("category") or "").strip()
+        if any(
+            k in cat
+            for k in (
+                "门店",
+                "店铺",
+                "餐饮",
+                "美容",
+                "美发",
+                "养生",
+                "健身",
+                "酒店",
+                "咖啡",
+                "奶茶",
+                "餐厅",
+                "颈肩",
+                "调理",
+            )
+        ):
+            type_counts["store"] = type_counts.get("store", 0) + 1
+        elif any(k in cat for k in ("人物", "穿搭", "人像", "服装")):
+            type_counts["person"] = type_counts.get("person", 0) + 1
+        elif any(k in cat for k in ("场景", "空间", "环境", "非产品")):
+            type_counts["scene"] = type_counts.get("scene", 0) + 1
+        elif cat and cat not in ("无法判断", "非产品图", ""):
+            type_counts["product"] = type_counts.get("product", 0) + 1
+        src = p.get("_source") or ""
+        if "store" in src:
+            type_counts["store"] = type_counts.get("store", 0) + 1
+        elif "person" in src:
+            type_counts["person"] = type_counts.get("person", 0) + 1
+
+    dominant = max(type_counts, key=type_counts.get) if type_counts else "product"
+    mp = (marketing_purpose or "").strip()
+
+    if any(k in mp for k in ("获客", "引流", "到店")):
+        if dominant == "store":
+            return "门店探店·到店体验"
+        return "门店探店·到店体验"
+    if any(k in mp for k in ("品牌", "宣传")):
+        return "品牌故事·门店体验" if dominant == "store" else "品牌故事·产品展示"
+    if any(k in mp for k in ("种草", "推荐")):
+        return "穿搭分享·人物种草" if dominant == "person" else "好物分享·产品种草"
+
+    theme_map = {
+        "store": "门店探店·到店体验",
+        "person": "穿搭分享·人物故事",
+        "scene": "空间体验·场景氛围",
+        "product": "好物分享·产品种草",
+    }
+    return theme_map.get(dominant, "好物分享·产品种草")
+
+
 def _build_products_summary(image_analysis: dict) -> str:
     """把 VLM 返回的商品分析结果拼给文案/分镜生成 prompt 用。
     优先用 summary（自然段落）；没有时用结构化字段兜底拼一段。"""
@@ -637,14 +743,36 @@ def _fallback_script(job: ViralVideoJob) -> dict:
     """脚本生成失败时的兜底脚本（极简但可用）。"""
     dur = max(5, min(30, int(getattr(job, "duration", 15) or 15)))
     ratio = getattr(job, "video_ratio", None) or "9:16"
-    base = _empty_copy_result(dur, ratio)
-    voiceover = job.user_copy_text or "你好，给大家分享一款我最近在用的好物，真的很不错，推荐你们也试试。"
+    _ia = getattr(job, "image_analysis", None) or {}
+    _mp = getattr(job, "marketing_purpose", "") or ""
+    default_theme = _determine_theme(_ia, _mp)
+    base = _empty_copy_result(dur, ratio, theme=default_theme)
+    _voiceover_map = {
+        "store": "带你探店！今天来到这家店，环境真的超棒，服务也很到位，推荐大家来体验一下。",
+        "person": "哈喽，今天给大家分享我的日常穿搭，简单舒适又好看，你们觉得怎么样？",
+        "scene": "带大家感受一下这个空间，氛围感拉满，真的很适合打卡体验。",
+        "product": "你好，给大家分享一款我最近在用的好物，真的很不错，推荐你们也试试。",
+    }
+    _products = (_ia or {}).get("products", []) or []
+    _dominant = "product"
+    for p in _products:
+        if not isinstance(p, dict):
+            continue
+        src = p.get("_source") or ""
+        cat = p.get("category") or ""
+        if "store" in src or any(k in cat for k in ("门店", "店铺", "餐饮", "美容", "颈肩", "调理")):
+            _dominant = "store"
+            break
+        elif "person" in src or any(k in cat for k in ("人物", "穿搭", "人像")):
+            _dominant = "person"
+            break
+    voiceover = job.user_copy_text or _voiceover_map.get(_dominant, _voiceover_map["product"])
     shots = [
         {
             "time_range": f"0-{dur}秒",
             "shot_type_angle_movement": "中景平视，缓慢推镜",
-            "scene_and_dialogue": "明亮室内，人物自然出镜，微笑着看向镜头。" + voiceover,
-            "action_details": "人物手持产品自然展示，表情亲切，动作流畅",
+            "scene_and_dialogue": voiceover,
+            "action_details": "自然展示，表情亲切，动作流畅",
             "audio_bgm": "轻快流行BGM",
             "transition": "结束",
             "reference_image_index": 0 if job.images else None,
@@ -654,7 +782,7 @@ def _fallback_script(job: ViralVideoJob) -> dict:
     base["voiceover_script"] = voiceover
     base["final_copy"] = voiceover
     base["suggested_copy"] = voiceover
-    base["title"] = "好物分享"
+    base["title"] = default_theme
     return base
 
 
@@ -672,12 +800,18 @@ def _validate_and_normalize_script(raw, job: ViralVideoJob) -> dict:
     ov = raw.get("overview")
     if isinstance(ov, dict):
         base["overview"] = {
-            "theme": str(ov.get("theme") or "好物分享"),
+            "theme": str(
+                ov.get("theme")
+                or _determine_theme(getattr(job, "image_analysis", None), getattr(job, "marketing_purpose", ""))
+            ),
             "total_duration": int(ov.get("total_duration") or dur),
             "aspect_ratio": str(ov.get("aspect_ratio") or ratio),
         }
     else:
-        base["overview"]["theme"] = str(raw.get("title") or "好物分享")
+        base["overview"]["theme"] = str(
+            raw.get("title")
+            or _determine_theme(getattr(job, "image_analysis", None), getattr(job, "marketing_purpose", ""))
+        )
 
     base["scene_and_lighting"] = str(raw.get("scene_and_lighting") or base["scene_and_lighting"])
 
@@ -764,7 +898,11 @@ def _script_from_xml(raw: str, job: ViralVideoJob) -> dict | None:
     base = _empty_copy_result(dur, ratio)
     if not raw:
         return None
-    base["overview"]["theme"] = xp.text_of(raw, "overview_theme") or xp.text_of(raw, "title") or "好物分享"
+    base["overview"]["theme"] = (
+        xp.text_of(raw, "overview_theme")
+        or xp.text_of(raw, "title")
+        or _determine_theme(getattr(job, "image_analysis", None), getattr(job, "marketing_purpose", ""))
+    )
     est = xp.attr_int(xp.text_of(raw, "estimated_duration"), 0)
     if est:
         base["overview"]["total_duration"] = est
@@ -825,10 +963,14 @@ def _step_script_generation(job: ViralVideoJob, intent: dict, image_analysis: di
             GLOBAL_CONSTRAINTS,
             NEGATIVE_RULES,
         )
-        from packages.shared.ai_service import call_llm
+        from packages.shared.ai_router import ai_router
     except ImportError:
         return _fallback_script(job)
 
+    _llm_client2 = ai_router.get_llm_client("storyboard")
+    if not _llm_client2.is_available:
+        return _fallback_script(job)
+
     products_summary = _build_products_summary(image_analysis)
     dur = max(5, min(30, int(getattr(job, "duration", 15) or 15)))
 
@@ -850,8 +992,11 @@ def _step_script_generation(job: ViralVideoJob, intent: dict, image_analysis: di
     system_tpl = system_tpl.replace("{global_constraints}", GLOBAL_CONSTRAINTS)
     system_tpl = system_tpl.replace("{negative_rules}", NEGATIVE_RULES)
 
+    marketing_purpose = getattr(job, "marketing_purpose", "") or "未指定"
+    image_category_hint = _determine_theme(image_analysis, marketing_purpose)
     fusion_brief = (
         f"意图：{intent_str}\n关键信息：{key_msgs}\n调性：{tone}\n"
+        f"营销目的：{marketing_purpose}\n建议主题方向：{image_category_hint}\n"
         f"用户原文：{job.user_copy_text or '（未提供）'}\n创作模式：{fusion_level}"
     )
     user = render_user_prompt(
@@ -862,13 +1007,14 @@ def _step_script_generation(job: ViralVideoJob, intent: dict, image_analysis: di
         image_analysis=products_summary,
     )
 
-    def _try_gen(model: str, temp: float, max_tok: int, label: str, tmo: int = 25):
-        logger.info("[爆款视频] 编导脚本生成 model=%s label=%s timeout=%d", model, label, tmo)
-        raw = call_llm(
+    def _try_gen(client, temp: float, max_tok: int, label: str, tmo: int = 25):
+        if not client or not client.is_available:
+            return None
+        logger.info("[爆款视频] 编导脚本生成 model=%s label=%s timeout=%d", client.model, label, tmo)
+        raw = client.chat_completion(
             [{"role": "system", "content": system_tpl}, {"role": "user", "content": user}],
             temperature=temp,
             max_tokens=max_tok,
-            model=model,
             timeout=tmo,
         )
         if not raw:
@@ -899,24 +1045,35 @@ def _step_script_generation(job: ViralVideoJob, intent: dict, image_analysis: di
         )
         return None if is_fallback else normalized
 
-    _s = get_shared_settings()
-    _fast = _s.doubao_fast_model
-    _pro = getattr(_s, "doubao_model", None) or _fast
+    # #2220: 直接用 ai_router 获取 client，不再手动提取 model_key
+    _client_fast = ai_router.get_llm_client("storyboard", variant="primary")
+    _client_pro = ai_router.get_llm_client("storyboard", variant="lite")
+    _script_fast_tmo = int(os.environ.get("VIRAL_VIDEO_SCRIPT_FAST_TIMEOUT", "90"))
+    _script_pro_tmo = int(os.environ.get("VIRAL_VIDEO_SCRIPT_PRO_TIMEOUT", "60"))
+    _script_deadline = time.time() + 180
     try:
-        # 第一次：快模型 25s
-        normalized = _try_gen(_fast, 0.8, 2500, "fast-first", tmo=90)
+        # #2233: fast_timeout=90s, pro_timeout=60s，总deadline 180s
+        normalized = _try_gen(_client_fast, 0.8, 2500, "fast-first", tmo=_script_fast_tmo)
         if normalized is not None:
             return normalized
-        # #2183: 实测pro 1500tok输出需75.8s，单次timeout提到90s
-        normalized = _try_gen(_fast, 0.6, 3200, "fast-retry", tmo=90)
+        if time.time() > _script_deadline:
+            logger.warning("[爆款视频] 编导脚本超过180s总预算，使用兜底脚本")
+            return _fallback_script(job)
+        normalized = _try_gen(_client_fast, 0.6, 3200, "fast-retry", tmo=_script_fast_tmo)
         if normalized is not None:
             return normalized
-        # 第三次：用主力模型兜底，给 120s
-        if _pro and _pro != _fast:
-            normalized = _try_gen(_pro, 0.7, 3500, "pro-fallback", tmo=120)
-            if normalized is not None:
-                return normalized
-        logger.warning("[爆款视频] 编导脚本三次都未生成合格结果，使用兜底脚本")
+        # 第三次：用 lite/pro 模型兜底，跳过与 primary 相同的模型
+        if _client_pro and _client_pro.is_available:
+            if _client_pro.model != _client_fast.model:
+                if time.time() <= _script_deadline:
+                    normalized = _try_gen(_client_pro, 0.7, 3500, "pro-fallback", tmo=_script_pro_tmo)
+                    if normalized is not None:
+                        return normalized
+                else:
+                    logger.warning("[爆款视频] 编导脚本超过180s总预算，跳过pro-fallback")
+            else:
+                logger.info("[爆款视频] pro-fallback模型与primary相同(%s)，跳过重复调用", _client_pro.model)
+        logger.warning("[爆款视频] 编导脚本均未生成合格结果，使用兜底脚本")
         return _fallback_script(job)
     except Exception as e:
         logger.warning("[爆款视频] 编导脚本生成异常: %s，使用兜底脚本", e, exc_info=True)
@@ -1236,9 +1393,12 @@ def _step_render(job: ViralVideoJob, copy_result: dict, tts_audio_url: str | Non
         if u not in all_portrait_urls:
             all_portrait_urls.append(u)
     pti = getattr(job, "pre_trusted_images", None)
-    if pti and len(pti) == len(all_portrait_urls):
-        pre_trusted = list(pti)
-        logger.info("[爆款视频] 使用信任链预热结果 n=%d，跳过现场 Seedream AI 化", len(pre_trusted))
+    # #2220: 稀疏列表模式（与images等长，None表示该位置保留原图）
+    _n_total = len(all_portrait_urls)
+    if pti and isinstance(pti, list) and len(pti) >= _n_total and _n_total > 0:
+        pre_trusted = list(pti[:_n_total])
+        _n_trusted = sum(1 for _x in pre_trusted if _x)
+        logger.info("[爆款视频] 使用信任链预热结果 person=%d total=%d，跳过现场Seedream AI化", _n_trusted, _n_total)
     elif all_portrait_urls and _mcfg.get("provider", "doubao") == "doubao":
         # #2183: 真·现场跑信任链——同步调用 Seedream t2i，拿到 AI 人像 URL 后再传 Seedance
         logger.info(
@@ -1251,32 +1411,39 @@ def _step_render(job: ViralVideoJob, copy_result: dict, tts_audio_url: str | Non
 
             _ia = getattr(job, "image_analysis", None) or {}
             _prods = (_ia.get("products") if isinstance(_ia, dict) else None) or []
-            _pdescs = []
-            if _prods:
-                _pdescs = [(pp.get("portrait_prompt") or "无人像") for pp in _prods]
-            elif isinstance(_ia, dict):
-                _pp0 = _ia.get("portrait_prompt") or "无人像"
-                if _pp0 and _pp0 != "无人像":
-                    _pdescs = [_pp0]
-            _valid = [d for d in _pdescs if d and isinstance(d, str) and "无人像" not in d and len(d) >= 10]
-            if _valid:
+            _live_person_idx: list[int] = []
+            _live_pdescs: list[str] = []
+            for _i, _pp in enumerate(_prods):
+                if not isinstance(_pp, dict) or not _pp.get("has_person", False):
+                    continue
+                _d = (_pp.get("portrait_prompt") or "").strip()
+                if not _d or "无人像" in _d or len(_d) < 10:
+                    continue
+                _live_person_idx.append(_i)
+                _live_pdescs.append(_d)
+            if _live_pdescs:
                 _t0 = time.time()
-                _live_urls = preheat_trust_chain(_valid, timeout=120)
-                if _live_urls and len(_live_urls) == len(all_portrait_urls):
-                    pre_trusted = list(_live_urls)
+                _live_urls = preheat_trust_chain(_live_pdescs, timeout=120)
+                if _live_urls and len(_live_urls) == len(_live_pdescs):
+                    # #2220: 构建稀疏列表，人像位替换AI图，非人像位保留None(ai_client里用原图)
+                    pre_trusted = [None] * len(all_portrait_urls)
+                    for _k, _u in enumerate(_live_urls):
+                        if _k < len(_live_person_idx):
+                            pre_trusted[_live_person_idx[_k]] = _u
                     logger.info(
-                        "[爆款视频] 现场信任链t2i完成 %d张 耗时%.1fs，将用AI人像传Seedance",
-                        len(pre_trusted),
+                        "[爆款视频] 现场信任链t2i完成 %d张人像AI化 耗时%.1fs（共%d张图，其余保留原图）",
+                        len(_live_urls),
                         time.time() - _t0,
+                        len(all_portrait_urls),
                     )
                 else:
                     logger.warning(
-                        "[爆款视频] 现场信任链t2i返回不匹配 urls=%s n_portraits=%d，回退原图+400降级纯t2v",
+                        "[爆款视频] 现场信任链t2i返回不匹配 urls=%s n_person=%d，人像位原图传Seedance（可能触发400拦截）",
                         _live_urls,
-                        len(all_portrait_urls),
+                        len(_live_pdescs),
                     )
             else:
-                logger.info("[爆款视频] 无有效人物描述（可能是商品图），无需现场跑信任链")
+                logger.info("[爆款视频] 无有效人物描述（商品/场景图），无需AI化，直接传原图给Seedance")
         except Exception as _te:
             logger.warning("[爆款视频] 现场跑信任链异常: %s，回退原图+400降级纯t2v", _te, exc_info=True)
 
@@ -1399,13 +1566,8 @@ def run_viral_video_pipeline(self: Task, job_id: str) -> dict:
         if job.images:
             try:
                 _products = (image_analysis or {}).get("products", []) or []
-                _portrait_descs = [(p.get("portrait_prompt") or "无人像") for p in _products] if _products else []
-                # 兼容单图结果格式（非products列表）
-                if not _portrait_descs and isinstance(image_analysis, dict):
-                    _pp = image_analysis.get("portrait_prompt") or "无人像"
-                    if _pp and _pp != "无人像":
-                        _portrait_descs = [_pp]
-                _start_trust_chain_preheat(job.id, _portrait_descs)
+                # #2220: 直接传 products 列表，由 _start_trust_chain_preheat 内部按 has_person 筛选
+                _start_trust_chain_preheat(job.id, _products)
             except Exception as _e:
                 logger.warning("[爆款视频][阶段1] 启动信任链t2i预热失败: %s", _e)
         _save_job(repo, job, session)
@@ -1577,12 +1739,8 @@ def run_viral_video_analyze(self: Task, job_id: str) -> dict:
         if job.images:
             try:
                 _products = (image_analysis or {}).get("products", []) or []
-                _portrait_descs = [(p.get("portrait_prompt") or "无人像") for p in _products] if _products else []
-                if not _portrait_descs and isinstance(image_analysis, dict):
-                    _pp = image_analysis.get("portrait_prompt") or "无人像"
-                    if _pp and _pp != "无人像":
-                        _portrait_descs = [_pp]
-                _start_trust_chain_preheat(job.id, _portrait_descs)
+                # #2220: 直接传 products 列表，由 _start_trust_chain_preheat 内部按 has_person 筛选
+                _start_trust_chain_preheat(job.id, _products)
             except Exception as _e:
                 logger.warning("[爆款视频][阶段1] 启动信任链t2i预热失败: %s", _e)
         _save_job(repo, job, session)
@@ -1670,7 +1828,7 @@ def run_viral_video_generate_copy(self: Task, job_id: str) -> dict:
         image_analysis = job.image_analysis or {"products": []}
         intent_result = _step_intent_parsing(job, image_analysis)
         job.intent_result = intent_result
-        _save_job(repo, job, session)
+        # #2218: 不在意图解析后单独落库，等 copy_result 生成后与 mark_copy_generated 一起原子写入
         _emit_progress(job_id, ViralVideoStage.INTENT_PARSING, 35.0, "意图解析完成")
 
         # 阶段：编导脚本生成（核心耗时环节，已用快模型）
@@ -1728,7 +1886,8 @@ def run_viral_video_generate_copy(self: Task, job_id: str) -> dict:
     except Retry:
         raise
     except Exception as e:
-        logger.error("[爆款视频][阶段2] 异常: %s", e, exc_info=True)
+        logger.error("[爆款视频][阶段2] 异常 job_id=%s: %s", job_id, e, exc_info=True)
+        # #2218: 阶段2任何异常都标记为 failed（由 _mark_failed_and_notify 处理），前端提示重试
         _mark_failed_and_notify(job_id, session, None, None, str(e), ViralVideoStage.SCRIPT_GENERATION)
         return {"ok": False, "job_id": job_id, "error": str(e)}
     finally:
@@ -1887,16 +2046,26 @@ def _run_render_pipeline(job_id: str, session, repo, job) -> dict:
     阶段2 generate-copy 已把 LLM 深度审核后置，这里在 TTS 前做最终审核（不通过则自动重写1次）。
     所有阶段通过 _set_stage 持久化 current_stage/phase_message。
     """
-    image_analysis = job.image_analysis or {"products": []}
 
-    # 如果没有 copy_result（旧数据/失败重试），现场补生成（意图+脚本，不走 LLM 审核，出片前会统一做）
+    # #2218: render 流程严禁补生成意图+编导脚本。copy_result 必须由 generate-copy 提前准备好；
+    # 若缺失说明 generate-copy 未完成或数据丢失，直接报错让用户重新点「生成文案」。
     copy_result = job.copy_result
+    _copy_src = "db"
     if not isinstance(copy_result, dict) or not copy_result:
-        _set_stage(job, repo, session, ViralVideoStage.SCRIPT_GENERATION, "正在补生成编导脚本...")
-        intent = job.intent_result or _step_intent_parsing(job, image_analysis)
-        copy_result = _step_script_generation(job, intent, image_analysis)
-        job.mark_copy_generated(copy_result)
-        _save_job(repo, job, session)
+        logger.error(
+            "[爆款视频][阶段3] copy_result 为空或无效，无法进入渲染流程。job_id=%s status=%s intent_len=%d，请重新触发「生成文案」",
+            job_id,
+            job.status,
+            len((job.intent_result or {}) if isinstance(job.intent_result, dict) else {}),
+        )
+        raise ValueError("文案数据缺失，请先点击「生成文案」完成文案生成后再生成视频")
+    logger.info(
+        "[爆款视频][阶段3] 进入渲染流程 job_id=%s copy_result_shots=%d copy_result_len=%d source=%s",
+        job_id,
+        len((copy_result.get("shots") or [])),
+        len(str(copy_result)),
+        _copy_src,
+    )
 
     # 出片前 LLM 深度合规审核（#2134 问题7：审核从阶段2后置到这里，不阻塞前端预览脚本）
     _set_stage(job, repo, session, ViralVideoStage.REVIEW, "正在进行出片前合规审核...")
@@ -1909,12 +2078,18 @@ def _run_render_pipeline(job_id: str, session, repo, job) -> dict:
             if isinstance(rewritten, dict) and rewritten:
                 copy_result = rewritten
             else:
-                intent = job.intent_result or _step_intent_parsing(job, image_analysis)
-                copy_result = _step_script_generation(job, intent, image_analysis)
-                _step_review(job, copy_result)
+                # #2218: 审核重写失败不再从意图解析重跑，直接报错让用户重新生成文案
+                logger.error(
+                    "[爆款视频][阶段3] 合规审核未通过且自动重写失败 job_id=%s，终止渲染",
+                    job_id,
+                )
+                raise ValueError("文案合规审核未通过，请修改文案后重试或重新生成文案")
             job.copy_result = copy_result
             job.generated_copy_text = copy_result.get("voiceover_script", "") or ""
             _save_job(repo, job, session)
+    except ValueError:
+        # #2218: 审核未通过/文案缺失的业务异常，不继续出片，向上抛出
+        raise
     except Exception as e:
         logger.warning("[爆款视频][阶段3] 合规审核异常，继续出片: %s", e)
     _emit_progress(job_id, ViralVideoStage.REVIEW, 70.0, "合规审核完成")

apps/worker/worker_app/tasks/vision/_prompt.py

@@ -3,8 +3,9 @@
 且 is_active=true），30s TTL 热加载；DB 无有效记录/异常时，fallback 到纯硬编码 JSON schema prompt。
 
 规则（简单直接，不做字符串匹配判断）：
-- DB 有 is_active=true 的 image_analysis 记录（含种子默认XML和用户修改后的版本）：
-  * system = DB.system_prompt + JSON_SCHEMA_APPEND（追加完整JSON字段schema，覆盖XML等其他输出格式要求）
+- DB 有 is_active=true 的 image_analysis 记录（含种子版本和用户修改后的版本）：
+  * system = DB.system_prompt（DB prompt 自带完整输出格式，不追加硬编码 schema，
+    避免 DB 写 XML、调用强制 json_object 造成的格式冲突）
   * user   = DB.user_prompt_template 渲染后使用；渲染后为空则用硬编码默认
 - DB 无记录/连接异常/返回空：system/user 全部用纯硬编码 JSON schema prompt
 """
@@ -73,7 +74,8 @@ _PRO_JSON_SCHEMA = (
 )
 DEFAULT_PRO_USER = "分析这张图片，返回符合schema的JSON。"
 
-# DB 配置存在时，追加在用户 system_prompt 末尾的JSON schema约束
+# 保留旧 JSON schema 追加文本作为常量（DB prompt 完全控制输出格式后不再使用，
+# 保留以便排查历史行为）。
 _FAST_JSON_APPEND = (
     "\n\n【输出格式要求】无论上文如何要求，最终你必须只返回一个合法的JSON对象，"
     "严格包含以下字段（字段值不确定时填null或空数组）：\n"
@@ -173,7 +175,6 @@ def _resolve(kind: str) -> tuple[str, str]:
 
     default_sys = _FAST_JSON_SCHEMA if kind == "fast" else _PRO_JSON_SCHEMA
     default_user = DEFAULT_FAST_USER if kind == "fast" else DEFAULT_PRO_USER
-    append = _FAST_JSON_APPEND if kind == "fast" else _PRO_JSON_APPEND
 
     sys_prompt = default_sys
     usr_prompt = default_user
@@ -182,7 +183,7 @@ def _resolve(kind: str) -> tuple[str, str]:
         if tpl is not None:
             db_sys = (getattr(tpl, "system_prompt", "") or "").strip()
             if db_sys:
-                sys_prompt = db_sys + append
+                sys_prompt = db_sys  # DB prompt自带完整输出格式，不追加硬编码schema避免冲突
                 usr_prompt = _render_user(tpl, default_user)
                 logger.info(
                     "[vision.v2] 使用DB image_analysis prompt (kind=%s version=%s sys_len=%d)",

apps/worker/worker_app/tasks/vision/assembler.py

@@ -1,5 +1,8 @@
 # -*- coding: utf-8 -*-
-"""把 fast_json VLM 输出 + OCR 文本组装为与旧 _normalize() 完全一致的 dict。
+"""把 fast_json VLM 输出 + OCR 文本组装为下游兼容的 product dict。
+
+v4 schema: DB prompt完全控制输出格式，可能是v4嵌套schema(type/products/people/store_info)
+或旧扁平schema(has_person/upper_wear/product_name/brand等)。assembler兼容两种格式。
 
 目标：下游（信任链t2i/intent_parsing/script_generation）零改动。
 必出字段：name, brand, category, appearance, packaging, text_on_package,
@@ -10,29 +13,16 @@ from __future__ import annotations
 
 from typing import Any
 
-# ---------- portrait_prompt 模板 ----------
-# 目标：60-100 字的人物穿搭描述，用于 Seedream 纯文生图。要求具体、风格化、视觉细节丰富。
-# 旧 VLM 输出格式参考："一位25岁左右的亚洲女性，身穿白色V领短袖T恤，黑色高腰阔腿裤，
-# 搭配银色项链，长发披肩，表情自信，街拍风格，阳光明媚的城市街头"
-
 
 def _join_parts(*parts: str | None) -> str:
     return "".join(p for p in parts if p)
 
 
-_AGE_PREFIX = {
-    "青年": "年轻",
-    "中年": "中年",
-    "老年": "老年",
-}
-# gender 后缀
+_AGE_PREFIX = {"青年": "年轻", "中年": "中年", "老年": "老年"}
 _GENDER_WORD = {"男": "男性", "女": "女性"}
 
 
-def _person_subject(fj: dict[str, Any]) -> str:
-    """人物主语：年轻女性 / 中年男性 / 少女 / 小男孩 / 人物 等。"""
-    gender = fj.get("gender") or ""
-    age = fj.get("age_range") or ""
+def _person_subject(gender: str, age: str) -> str:
     gw = _GENDER_WORD.get(gender, "")
     if age == "儿童":
         if gender == "女":
@@ -52,8 +42,147 @@ def _person_subject(fj: dict[str, Any]) -> str:
     return f"{prefix}人物" if prefix else "人物"
 
 
-def _build_wear_sentence(fj: dict[str, Any]) -> str:
-    """穿搭段：上装+下装/连衣裙，带颜色+材质+图案。"""
+def _build_wear_from_v4(p: dict) -> str:
+    """v4 person schema: upper_wear/upper_color/lower_wear/lower_color/dress_color"""
+    upper = p.get("upper_wear") or ""
+    upper_color = p.get("upper_color") or ""
+    lower = p.get("lower_wear") or ""
+    lower_color = p.get("lower_color") or ""
+    dress_color = p.get("dress_color") or ""
+    is_dress = ("连衣裙" in upper) or ("裙" in upper and not lower)
+    if is_dress:
+        c = dress_color or upper_color
+        return f"身穿{c}{upper}" if c else f"身穿{upper}"
+    parts = []
+    if upper:
+        up = f"{upper_color}{upper}" if upper_color else upper
+        parts.append(f"上身{up}")
+    if lower:
+        lo = f"{lower_color}{lower}" if lower_color else lower
+        parts.append(f"下身{lo}")
+    return "，".join(parts)
+
+
+def _build_portrait_prompt_from_v4(p: dict) -> str:
+    """v4 person: 直接用portrait_prompt字段；没有就拼"""
+    direct = p.get("portrait_prompt")
+    if direct and len(direct) >= 10:
+        return direct
+    subject = _person_subject(p.get("gender", ""), p.get("age_range", ""))
+    wear = _build_wear_from_v4(p)
+    acc = p.get("accessories") or []
+    if isinstance(acc, str):
+        acc = [acc]
+    acc_str = "，佩戴" + "、".join(str(a) for a in acc if a) if acc else ""
+    hair = p.get("hairstyle") or ""
+    expr = p.get("expression") or ""
+    pose = p.get("pose") or ""
+    style = p.get("outfit_style") or p.get("style") or ""
+    scene = p.get("scene") or ""
+    mood = p.get("mood") or ""
+    details = []
+    if hair:
+        details.append(hair)
+    if expr and expr not in ("自然", "平静"):
+        details.append(f"神情{expr}")
+    if pose and pose not in ("站立",):
+        details.append(pose)
+    style_parts = []
+    if style:
+        style_parts.append(style)
+    if mood:
+        style_parts.append(mood)
+    if scene and scene not in ("通用",):
+        style_parts.append(scene)
+    pieces = [f"一位{subject}"]
+    if wear:
+        pieces.append(wear)
+    if acc_str:
+        pieces.append(acc_str.lstrip("，"))
+    if details:
+        pieces.append("，".join(details))
+    pieces.append(("".join(style_parts) + "风格") if style_parts else "人像写真")
+    full = "，".join(p for p in pieces if p)
+    if len(full) < 40:
+        full += "，自然光线下人像特写，画面清晰"
+    if len(full) > 120:
+        full = full[:120].rstrip("，") + "。"
+    return full
+
+
+def _build_product_prompt_from_v4(prod: dict, top: dict) -> str:
+    """v4 product: 拼商品视觉描述prompt（用于AI生图参考）"""
+    name = prod.get("product_name") or "商品"
+    brand = prod.get("brand") or ""
+    lead = f"{brand} {name}" if brand and brand not in name else name
+    pkg_color = prod.get("package_color") or ""
+    pkg_type = prod.get("package_type") or ""
+    cap = prod.get("cap_type") or ""
+    body = prod.get("body_shape") or ""
+    features = prod.get("product_features") or []
+    sell = prod.get("key_selling_points") or []
+    colors = top.get("colors") or []
+    style = top.get("style") or ""
+    scene = top.get("scene") or ""
+    mood = top.get("mood") or ""
+
+    parts = [lead]
+    desc = []
+    if pkg_color:
+        desc.append(pkg_color)
+    if pkg_type:
+        desc.append(pkg_type)
+    if cap and len(desc) < 3:
+        desc.append(f"配{cap}")
+    if body and len(desc) < 3:
+        desc.append(body)
+    if desc:
+        parts.append("，".join(desc))
+    if features:
+        core = [str(f) for f in features[:3] if f and len(str(f)) <= 25]
+        if core:
+            parts.append("；".join(core))
+    if sell:
+        s = [str(x) for x in sell[:2] if x]
+        if s:
+            parts.append("突出" + "、".join(s))
+    cnames = []
+    for cc in colors:
+        if isinstance(cc, dict) and cc.get("name"):
+            cnames.append(cc["name"])
+        elif isinstance(cc, str):
+            cnames.append(cc)
+    cnames = cnames[:3]
+    if cnames:
+        parts.append("、".join(cnames) + "主色")
+    if style:
+        parts.append(style)
+    if mood:
+        parts.append(mood)
+    if scene and not any(k in scene for k in ("白色背景", "纯色", "通用")):
+        parts.append(scene)
+    parts.append("产品特写，画面清晰")
+    prompt = "，".join(p for p in parts if p)
+    return prompt if len(prompt) >= 10 else "产品展示图，特写镜头"
+
+
+def _is_v4_schema(fj: dict) -> bool:
+    """判断是v4嵌套schema还是旧扁平schema"""
+    return (
+        isinstance(fj.get("products"), list)
+        or fj.get("type") in ("product", "store", "person", "scene", "other")
+        or isinstance(fj.get("people"), dict)
+    )
+
+
+# ---------- 旧扁平schema兼容（保留原逻辑） ----------
+
+
+def _person_subject_old(fj: dict) -> str:
+    return _person_subject(fj.get("gender", ""), fj.get("age_range", ""))
+
+
+def _build_wear_sentence_old(fj: dict) -> str:
     upper = fj.get("upper_wear") or ""
     upper_color = fj.get("upper_color") or ""
     lower = fj.get("lower_wear") or ""
@@ -61,7 +190,6 @@ def _build_wear_sentence(fj: dict[str, Any]) -> str:
     dress_color = fj.get("dress_color") or ""
     material = fj.get("material") or ""
     pattern = fj.get("pattern") or ""
-
     is_dress = ("连衣裙" in upper) or ("裙" in upper and not lower)
     if is_dress:
         c = dress_color or upper_color
@@ -71,25 +199,22 @@ def _build_wear_sentence(fj: dict[str, Any]) -> str:
         if pattern and pattern not in wear and pattern != "纯色":
             wear += f"，{pattern}图案"
         return f"身穿{wear}"
-
-    parts: list[str] = []
+    parts = []
     if upper:
         up = f"{upper_color}{upper}" if upper_color else upper
         if material and material not in up:
             up = f"{material}{up}"
         if pattern and pattern != "纯色" and pattern not in up:
             up += f"（{pattern}）"
-        parts.append(f"上身{up}" if up else "")
+        parts.append(f"上身{up}")
     if lower:
         lo = f"{lower_color}{lower}" if lower_color else lower
-        parts.append(f"下身{lo}" if lo else "")
+        parts.append(f"下身{lo}")
     return "，".join(p for p in parts if p)
 
 
-def _build_portrait_prompt(fj: dict[str, Any]) -> str:
-    """组装最终 portrait_prompt（目标 60-100 字，用于 Seedream 纯文生图）。"""
+def _build_portrait_prompt_old(fj: dict) -> str:
     if not fj.get("has_person"):
-        # 非人像：用商品+场景+mood 拼一段
         name = fj.get("product_name") or "商品"
         brand = fj.get("brand") or ""
         colors = fj.get("colors") or []
@@ -101,7 +226,15 @@ def _build_portrait_prompt(fj: dict[str, Any]) -> str:
             pieces.append(brand)
         pieces.append(name)
         if colors:
-            pieces.append("、".join(colors[:3]) + "配色")
+            cnames = []
+            for c in colors:
+                if isinstance(c, dict):
+                    cnames.append(c.get("name", ""))
+                elif isinstance(c, str):
+                    cnames.append(c)
+            cnames = [c for c in cnames if c][:3]
+            if cnames:
+                pieces.append("、".join(cnames) + "配色")
         if style:
             pieces.append(style + "风格")
         if mood:
@@ -111,40 +244,32 @@ def _build_portrait_prompt(fj: dict[str, Any]) -> str:
         pieces.append("产品特写")
         prompt = "，".join(p for p in pieces if p)
         return prompt if len(prompt) >= 10 else "产品展示图，特写镜头"
-
-    subject = _person_subject(fj)
-    wear = _build_wear_sentence(fj)
-
+    subject = _person_subject_old(fj)
+    wear = _build_wear_sentence_old(fj)
     accessories = fj.get("accessories") or []
     if isinstance(accessories, str):
         accessories = [accessories]
-    acc_str = ""
-    if accessories:
-        acc_str = "，佩戴" + "、".join(str(a) for a in accessories if a)
-
+    acc_str = "，佩戴" + "、".join(str(a) for a in accessories if a) if accessories else ""
     hairstyle = fj.get("hairstyle") or ""
     expression = fj.get("expression") or ""
     pose = fj.get("pose") or ""
     style = fj.get("style") or ""
     scene = fj.get("scene") or ""
     mood = fj.get("mood") or ""
-
-    detail_parts: list[str] = []
+    detail_parts = []
     if hairstyle:
         detail_parts.append(hairstyle)
     if expression and expression not in ("自然", "平静"):
         detail_parts.append(f"神情{expression}")
     if pose and pose not in ("站立",):
         detail_parts.append(pose)
-
-    style_parts: list[str] = []
+    style_parts = []
     if style:
         style_parts.append(style)
     if mood:
         style_parts.append(mood)
     if scene and scene not in ("通用",):
         style_parts.append(scene)
-
     pieces = [f"一位{subject}"]
     if wear:
         pieces.append(wear)
@@ -152,53 +277,40 @@ def _build_portrait_prompt(fj: dict[str, Any]) -> str:
         pieces.append(acc_str.lstrip("，"))
     if detail_parts:
         pieces.append("，".join(detail_parts))
-    if style_parts:
-        # 风格词之间不用逗号，用空格紧凑
-        pieces.append("".join(style_parts) + "风格")
-    else:
-        pieces.append("人像写真")
-
+    pieces.append("".join(style_parts) + "风格" if style_parts else "人像写真")
     full = "，".join(p for p in pieces if p)
-    # 过短补充镜头词
     if len(full) < 40:
         full += "，自然光线下人像特写，画面清晰"
-    # 过长截断
     if len(full) > 120:
         full = full[:120].rstrip("，") + "。"
     return full
 
 
-# ---------- 商品字段 ----------
-
-
-def _infer_name(fj: dict[str, Any], ocr_texts: list[str]) -> str:
+def _infer_name_old(fj: dict, ocr_texts: list[str]) -> str:
     pname = fj.get("product_name")
     if pname and pname != "未识别":
         return str(pname)
-    # 人物图 → name 用穿搭主件
     if fj.get("has_person"):
         up = fj.get("upper_wear") or ""
         if "连衣裙" in up:
             return up
         return up or "人物穿搭"
     if ocr_texts:
-        # 商品名可能是 OCR 最长的一行（品牌/产品名）
         return max(ocr_texts, key=len)
     return "未识别"
 
 
-def _infer_brand(fj: dict[str, Any], ocr_texts: list[str]) -> str:
+def _infer_brand_old(fj: dict, ocr_texts: list[str]) -> str:
     brand = fj.get("brand")
     if brand:
         return str(brand)
-    # OCR 里短的、纯字母/汉字短串可能是 brand
     for t in ocr_texts:
         if 1 < len(t) <= 12:
             return t
     return "无法判断"
 
 
-def _infer_category(fj: dict[str, Any]) -> str:
+def _infer_category_old(fj: dict) -> str:
     cat = fj.get("category")
     if cat:
         return str(cat)
@@ -207,27 +319,19 @@ def _infer_category(fj: dict[str, Any]) -> str:
     return "非产品图"
 
 
-def _build_appearance(fj: dict[str, Any]) -> str:
-    """外观描述：颜色+款式+材质+图案 拼成一段。"""
-    parts: list[str] = []
-    for key, _label in [
-        ("upper_color", "主色"),
-        ("upper_wear", "款式"),
-        ("material", "材质"),
-        ("pattern", "图案"),
-    ]:
+def _build_appearance_old(fj: dict) -> str:
+    parts = []
+    for key in ("upper_color", "upper_wear", "material", "pattern"):
         v = fj.get(key)
         if v and v not in ("无法判断", "未知", "纯色"):
             parts.append(str(v))
     if not parts:
-        if fj.get("has_person"):
-            return "人像穿搭整体造型"
-        return "无法判断"
+        return "人像穿搭整体造型" if fj.get("has_person") else "无法判断"
     return "、".join(parts)
 
 
-def _build_key_features(fj: dict[str, Any], ocr_texts: list[str]) -> list[str]:
-    feats: list[str] = []
+def _build_key_features_old(fj: dict, ocr_texts: list[str]) -> list[str]:
+    feats = []
     for key in (
         "upper_wear",
         "lower_wear",
@@ -248,9 +352,7 @@ def _build_key_features(fj: dict[str, Any], ocr_texts: list[str]) -> list[str]:
             feats.append(v)
     if ocr_texts:
         feats.append(f"画面文字: {'/'.join(ocr_texts[:3])}")
-    # 去重
-    out: list[str] = []
-    seen: set[str] = set()
+    out, seen = [], set()
     for f in feats:
         f = f.strip()
         if f and f not in seen and len(f) <= 30:
@@ -259,27 +361,552 @@ def _build_key_features(fj: dict[str, Any], ocr_texts: list[str]) -> list[str]:
     return out[:6] if out else ["无法判断"]
 
 
-def assemble_result(
-    idx: int,
-    fast_json: dict[str, Any] | None,
-    ocr_texts: list[str],
-) -> dict[str, Any]:
-    """把 fast_json 结果 + OCR 文本组装成下游兼容的 product dict。"""
+def _flatten_colors(c) -> list[str]:
+    """colors可能是字符串数组或[{hex,name,coverage}]，统一返回名字数组"""
+    if not c:
+        return []
+    out = []
+    for item in c:
+        if isinstance(item, dict):
+            n = item.get("name")
+            if n:
+                out.append(n)
+        elif isinstance(item, str):
+            out.append(item)
+    return out
+
+
+def assemble_result(idx: int, fast_json: dict | None, ocr_texts: list[str]) -> dict[str, Any]:
     fj = fast_json or {}
     ocr_texts = ocr_texts or []
 
-    portrait_prompt = _build_portrait_prompt(fj)
-    name = _infer_name(fj, ocr_texts)
-    brand = _infer_brand(fj, ocr_texts)
-    category = _infer_category(fj)
-    appearance = _build_appearance(fj)
-    key_features = _build_key_features(fj, ocr_texts)
+    if _is_v4_schema(fj):
+        result = _assemble_v4(idx, fj, ocr_texts)
+    else:
+        result = _assemble_old(idx, fj, ocr_texts)
+    return _apply_partial_fallback(result, fj)
+
+
+def _apply_partial_fallback(result: dict[str, Any], fj: dict) -> dict[str, Any]:
+    """partial（截断修复）产物的字段兜底：用已有碎片填充空字段，
+    避免"无法判断"直接透传给下游。非partial产物原样返回。"""
+    if not fj.get("_partial"):
+        return result
+    desc = str(fj.get("description") or "").strip()
+    # 收集所有顶层标量碎片作为兜底素材
+    fragments: list[str] = []
+    for k in ("main_subject", "store_type", "scene_type", "description"):
+        v = fj.get(k)
+        if isinstance(v, str) and v.strip() and v != "无法判断":
+            fragments.append(v.strip())
+    for arr_k in ("environment_objects", "key_elements", "visual_elements"):
+        arr = fj.get(arr_k) or []
+        if isinstance(arr, list):
+            for item in arr[:3]:
+                if isinstance(item, str) and item.strip():
+                    fragments.append(item.strip())
+                elif isinstance(item, dict):
+                    tv = item.get("text") or item.get("name")
+                    if tv:
+                        fragments.append(str(tv))
+    frag_text = "；".join(fragments[:3])
+
+    if result.get("name") in ("未识别", "", None) and (desc or frag_text):
+        result["name"] = (desc or fragments[0])[:30]
+    if str(result.get("appearance", "")).startswith("无法判断"):
+        if desc:
+            result["appearance"] = desc[:200]
+        elif frag_text:
+            result["appearance"] = frag_text[:200]
+    if result.get("key_features") in (["无法判断"], []) and (desc or fragments):
+        kf = []
+        if desc:
+            kf.append(desc[:30])
+        for f in fragments[:3]:
+            if f not in kf:
+                kf.append(f[:40])
+        result["key_features"] = kf[:8]
+    if result.get("summary") in ("未识别", "", None) and (desc or fragments):
+        result["summary"] = (desc or fragments[0])[:40]
+    result["_partial"] = True
+    return result
+
+
+def _assemble_v4(idx: int, fj: dict, ocr_texts: list[str]) -> dict[str, Any]:
+    """v4嵌套schema → 下游product dict"""
+    vtype = fj.get("type") or "other"
+    products = fj.get("products") or []
     scene = fj.get("scene") or "通用"
     mood = fj.get("mood") or ""
-    packaging = "无法判断"  # 包装细节专用API无，保留占位
-    text_on_package = ocr_texts[:8]
-    summary = _build_summary(fj, name, brand, category)
+    colors = fj.get("colors") or []
+    visible_text = fj.get("visible_text") or []
+    color_names = _flatten_colors(colors)
 
+    # 合并OCR文字和visible_text
+    pkg_texts = []
+    for vt in visible_text:
+        if isinstance(vt, dict):
+            t = vt.get("text")
+            if t:
+                pkg_texts.append(str(t))
+        elif isinstance(vt, str):
+            pkg_texts.append(vt)
+    pkg_texts.extend(ocr_texts[:5])
+    # 去重
+    seen_t = set()
+    text_on_package = []
+    for t in pkg_texts:
+        t = str(t).strip()
+        if t and t not in seen_t and len(t) <= 50:
+            seen_t.add(t)
+            text_on_package.append(t)
+    text_on_package = text_on_package[:8]
+
+    has_person = fj.get("has_person", False)
+
+    # 兼容老schema：无type字段或type不在已知枚举时，按has_person/products兜底
+    _KNOWN_V4_TYPES = ("product", "store", "person", "scene", "other")
+    if vtype not in _KNOWN_V4_TYPES:
+        if has_person:
+            vtype = "person"
+        elif products:
+            vtype = "product"
+        else:
+            vtype = "other"
+
+    # 人物信息提取辅助（scene/store分支有人物时追加到key_features）
+    def _extract_person_features(source: dict) -> list[str]:
+        parts = []
+        for pk in ("outfit_style", "upper_wear", "lower_wear", "dress_wear", "outerwear", "pose", "expression"):
+            pv = source.get(pk)
+            if pv and str(pv) not in ("null", None, "无法判断") and len(str(pv)) <= 40:
+                parts.append(f"人物:{pv}")
+        for pk2 in ("shoes", "bag", "hairstyle"):
+            pv2 = source.get(pk2)
+            if pv2 and str(pv2) not in ("null", None) and len(str(pv2)) <= 40:
+                parts.append(f"人物:{pv2}")
+        return parts
+
+    # partial截断保护：声明了product但products数组没来得及输出时，
+    # 按已返回的碎片字段改路由，避免直接掉到other丢信息
+    if fj.get("_partial") and vtype == "product" and not products:
+        if any(fj.get(k) for k in ("signage_details", "store_layout", "brand_signage", "store_type")):
+            vtype = "store"
+        elif any(fj.get(k) for k in ("key_elements", "main_subject", "scene_type", "spatial_layout")):
+            vtype = "scene"
+        else:
+            vtype = "other"
+
+    # ── 人物类 ──
+    if vtype == "person":
+        # 取第一个人物信息（v5 schema人物信息在顶层）
+        person_info = fj
+        # 兼容people嵌套
+        ppl = fj.get("people")
+        if isinstance(ppl, dict) and ppl.get("has_person"):
+            person_info = {**fj, **ppl}
+            has_person = True
+
+        portrait_prompt = _build_portrait_prompt_from_v4(person_info)
+        outfit_style = person_info.get("outfit_style") or ""
+        upper = person_info.get("upper_wear") or ""
+        lower = person_info.get("lower_wear") or ""
+        dress = person_info.get("dress_wear") or ""
+        outer = person_info.get("outerwear") or ""
+        if dress:
+            name = str(dress)[:25]
+        elif outer and upper:
+            name = f"{outer}+{upper}"[:30]
+        elif upper:
+            name = (str(upper) + (f"+{lower}" if lower else ""))[:30]
+        else:
+            name = "人物穿搭"
+        brand = "无法判断"
+        category = "人物穿搭"
+        # appearance: 外套+上衣+下装/裙+鞋+包+发型+妆容
+        app_parts = []
+        for k in ("outerwear", "upper_wear", "lower_wear", "dress_wear", "shoes", "bag", "hairstyle", "makeup"):
+            v = person_info.get(k)
+            if v and v not in ("null", None, "无明显妆容"):
+                app_parts.append(str(v))
+        appearance = "；".join(app_parts) if app_parts else "人像穿搭整体造型"
+        # key_features: 服装+配饰+拍摄信息
+        kf = []
+        for k in (
+            "outfit_style",
+            "upper_wear",
+            "lower_wear",
+            "dress_wear",
+            "outerwear",
+            "shoes",
+            "bag",
+            "hairstyle",
+            "expression",
+            "pose",
+        ):
+            v = person_info.get(k)
+            if v and v not in ("null", None, "无法判断"):
+                kf.append(str(v))
+        acc = person_info.get("accessories") or []
+        if isinstance(acc, list):
+            for a in acc:
+                if a and str(a) not in kf:
+                    kf.append(str(a))
+        elif isinstance(acc, str) and acc:
+            kf.append(acc)
+        for k in ("shot_type", "camera_angle", "lighting", "atmosphere"):
+            v = person_info.get(k)
+            if v and v not in ("null", None):
+                kf.append(str(v))
+        env_obj = person_info.get("environment_objects") or []
+        if not isinstance(env_obj, list):
+            env_obj = [env_obj]
+        for o in env_obj:
+            if o and str(o) not in kf and len(str(o)) <= 40:
+                kf.append(str(o))
+        if text_on_package:
+            kf.append(f"文字:{'/'.join(text_on_package[:3])}")
+        kf = kf[:8] or ["无法判断"]
+        summary = (outfit_style + " " if outfit_style and outfit_style not in name else "") + name[:25]
+        if not summary.strip():
+            summary = "人物穿搭"
+        return {
+            "name": name[:30],
+            "brand": brand,
+            "category": category,
+            "appearance": appearance[:400],
+            "packaging": "人物形象无包装",
+            "text_on_package": text_on_package,
+            "key_features": kf,
+            "scene": scene,
+            "mood": mood,
+            "portrait_prompt": portrait_prompt[:300],
+            "summary": summary[:50],
+            "_source": "v2_fast_json_v5",
+            "has_person": True,
+        }
+
+    # 商品类
+    if vtype == "product" and products:
+        # 主商品（第一个position=main或第一个）
+        main = products[0]
+        for p in products:
+            if p.get("position") == "main":
+                main = p
+                break
+        name = main.get("product_name") or "未识别"
+        brand = main.get("brand") or "无法判断"
+        category = main.get("category") or "非产品图"
+        # appearance: 包装外观
+        app_parts = []
+        for k in ("package_color", "package_type", "cap_type", "body_shape", "label_design"):
+            v = main.get(k)
+            if v and v not in ("null", None):
+                app_parts.append(str(v))
+        appearance = "；".join(app_parts) if app_parts else "无法判断"
+        # packaging: 包装信息（直接用package_type+package_color）
+        pkg_parts = []
+        if main.get("package_type"):
+            pkg_parts.append(str(main["package_type"]))
+        if main.get("package_color"):
+            pkg_parts.append(str(main["package_color"]))
+        if main.get("cap_type"):
+            pkg_parts.append(f"配{main['cap_type']}")
+        packaging = "，".join(pkg_parts) if pkg_parts else "无法判断"
+        # key_features: product_features字段
+        feats = main.get("product_features") or []
+        if not isinstance(feats, list):
+            feats = [str(feats)]
+        kf = [str(f) for f in feats if f and len(str(f)) <= 40][:6]
+        # 补充卖点
+        sell = main.get("key_selling_points") or []
+        if isinstance(sell, list):
+            for s in sell[:2]:
+                if s and len(str(s)) <= 30 and str(s) not in kf:
+                    kf.append(f"卖点:{s}")
+        env_obj = fj.get("environment_objects") or []
+        if not isinstance(env_obj, list):
+            env_obj = [env_obj]
+        for o in env_obj:
+            if o and str(o) not in kf and len(str(o)) <= 40:
+                kf.append(str(o))
+        if text_on_package:
+            kf.append(f"文字: {'/'.join(text_on_package[:3])}")
+        kf = kf[:8] or ["无法判断"]
+        portrait_prompt = _build_product_prompt_from_v4(main, fj)
+        if brand != "无法判断" and brand not in name:
+            summary = f"{brand} {name}"
+        else:
+            summary = name
+        return {
+            "name": str(name)[:50],
+            "brand": str(brand)[:30],
+            "category": str(category)[:20],
+            "appearance": appearance[:200],
+            "packaging": packaging[:100],
+            "text_on_package": text_on_package,
+            "key_features": kf,
+            "scene": scene,
+            "mood": mood,
+            "portrait_prompt": portrait_prompt[:200],
+            "summary": str(summary)[:60],
+            "has_person": False,
+            "_source": "v2_fast_json_v4",
+        }
+
+    # 门店类
+    if vtype == "store":
+        store_type = fj.get("store_type") or "店铺"
+        # brand 多级兜底：brand_signage → visible_text招牌文字 → text_on_package短词
+        brand_raw = fj.get("brand_signage")
+        if not brand_raw or brand_raw in ("无法判断", "", None):
+            brand = None
+            # 从visible_text找招牌文字（通常是位置含招牌/门头/背景的短词）
+            for vt in visible_text:
+                vt_str = vt.get("text") if isinstance(vt, dict) else str(vt)
+                if not vt_str or len(vt_str) < 2 or len(vt_str) > 12:
+                    continue
+                loc = (vt.get("location") or "") if isinstance(vt, dict) else ""
+                if any(k in loc for k in ("招牌", "门头", "背景", "招牌墙")):
+                    brand = vt_str
+                    break
+            # 从text_on_package找2-8字的短词（非描述性）
+            if not brand:
+                _desc_words = {"干净", "整洁", "温馨", "专业", "明亮", "舒适", "宽敞", "现代", "传统", "时尚"}
+                for t in text_on_package:
+                    if 2 <= len(t) <= 8 and t not in _desc_words and not any(c in t for c in "的了是在我"):
+                        brand = t
+                        break
+            if not brand:
+                brand = "无法判断"
+        else:
+            brand = brand_raw
+        # name兜底：store_type为空时用brand
+        name = store_type if store_type != "店铺" else (brand if brand != "无法判断" else store_type)
+        category = "门店场景"
+        # appearance: store_layout + furnishings + 陈设色调
+        appearance_parts = []
+        if fj.get("store_layout"):
+            appearance_parts.append(str(fj["store_layout"]))
+        furnishings = fj.get("furnishings") or []
+        if isinstance(furnishings, dict):
+            _furn_vals = []
+            for fk in ("materials", "furniture", "shelving", "seating"):
+                fv = furnishings.get(fk)
+                if isinstance(fv, list):
+                    _furn_vals.extend(str(x) for x in fv if x)
+                elif isinstance(fv, str) and fv:
+                    _furn_vals.append(fv)
+            if _furn_vals:
+                appearance_parts.append("陈设：" + "、".join(_furn_vals[:4]))
+        elif isinstance(furnishings, list) and furnishings:
+            appearance_parts.append("陈设：" + "、".join(str(f) for f in furnishings[:4] if f))
+        if fj.get("cleanliness"):
+            appearance_parts.append(str(fj["cleanliness"]))
+        appearance = "；".join(appearance_parts) if appearance_parts else "门店环境"
+        # key_features: 招牌细节+海报+外部物品+视觉元素+陈列商品+环境物件
+        kf = []
+        for arr_key in (
+            "signage_details",
+            "signage_posters",
+            "exterior_items",
+            "visual_elements",
+            "products_on_display",
+            "environment_objects",
+        ):
+            arr = fj.get(arr_key) or []
+            if not isinstance(arr, list):
+                arr = [arr]
+            for item in arr:
+                if item and str(item) not in kf and len(str(item)) <= 40:
+                    kf.append(str(item))
+        prods_vis = fj.get("product_categories_visible") or []
+        if isinstance(prods_vis, list):
+            for c in prods_vis[:3]:
+                if c and str(c) not in kf:
+                    kf.append(str(c))
+        promo = fj.get("promotion_elements") or []
+        if isinstance(promo, list) and promo:
+            kf.append("促销活动:" + "、".join(str(p) for p in promo[:2]))
+        if has_person:
+            for pf in _extract_person_features(fj):
+                if pf not in kf:
+                    kf.append(pf)
+        if text_on_package:
+            kf.append("文字:" + "/".join(text_on_package[:3]))
+        kf = kf[:8] or ["门店场景"]
+        atmosphere = fj.get("atmosphere") or fj.get("mood") or mood
+        # portrait_prompt: 品牌+store_type+主色+atmosphere+scene+核心视觉元素
+        pieces = []
+        if brand != "无法判断":
+            pieces.append(brand)
+        pieces.append(store_type)
+        if color_names:
+            pieces.append("、".join(color_names[:3]) + "配色")
+        if atmosphere:
+            pieces.append(atmosphere)
+        if scene and scene != "通用":
+            pieces.append(scene)
+        visual = fj.get("visual_elements") or []
+        if isinstance(visual, list) and visual:
+            pieces.append("、".join(str(v) for v in visual[:3] if v))
+        pieces.append("门店实拍")
+        portrait_prompt = "，".join(p for p in pieces if p)
+        if len(portrait_prompt) > 200:
+            portrait_prompt = portrait_prompt[:200].rstrip("，")
+        summary = f"{brand} {store_type}" if brand != "无法判断" else f"{store_type}场景"
+        return {
+            "name": name[:30],
+            "brand": str(brand)[:30],
+            "category": category,
+            "appearance": appearance[:300],
+            "packaging": "门店场景无包装",
+            "text_on_package": text_on_package,
+            "key_features": kf,
+            "scene": scene,
+            "mood": atmosphere,
+            "portrait_prompt": portrait_prompt[:200],
+            "summary": summary[:40],
+            "has_person": bool(fj.get("has_person", False)),
+            "_source": "v2_fast_json_v6_store",
+        }
+
+    # 场景类（纯场景图，无产品/人物）
+    if vtype == "scene":
+        main_subject = fj.get("main_subject") or ""
+        scene_type = fj.get("scene_type") or "场景图"
+        name = main_subject or scene_type
+        brand = "无法判断"
+        category = scene_type or "场景图"
+        # appearance: 空间布局 + 关键元素
+        app_parts = []
+        if fj.get("spatial_layout"):
+            app_parts.append(str(fj["spatial_layout"]))
+        key_elements = fj.get("key_elements") or []
+        if not isinstance(key_elements, list):
+            key_elements = [key_elements]
+        if key_elements:
+            app_parts.append("关键元素：" + "、".join(str(k) for k in key_elements[:4] if k))
+        if fj.get("lighting"):
+            app_parts.append("光线：" + str(fj["lighting"]))
+        appearance = "；".join(app_parts) if app_parts else "场景环境"
+        # key_features: key_elements + environment_objects（去重，最多8）
+        kf = []
+        for k in key_elements:
+            if k and str(k) not in kf and len(str(k)) <= 40:
+                kf.append(str(k))
+        env_obj = fj.get("environment_objects") or []
+        if not isinstance(env_obj, list):
+            env_obj = [env_obj]
+        for o in env_obj:
+            if o and str(o) not in kf and len(str(o)) <= 40:
+                kf.append(str(o))
+        if has_person:
+            for pf in _extract_person_features(fj):
+                if pf not in kf:
+                    kf.append(pf)
+        if text_on_package:
+            kf.append("文字:" + "/".join(text_on_package[:3]))
+        kf = kf[:8] or ["场景元素"]
+        scene_name = fj.get("scene") or scene_type
+        scene_mood = fj.get("mood") or mood
+        packaging = "场景无包装"
+        # portrait_prompt: main_subject + key_elements + lighting + atmosphere + composition + style
+        pieces = []
+        if main_subject:
+            pieces.append(main_subject)
+        if key_elements:
+            pieces.append("、".join(str(k) for k in key_elements[:3] if k))
+        if fj.get("lighting"):
+            pieces.append(str(fj["lighting"]) + "光线")
+        if scene_mood:
+            pieces.append(scene_mood + "氛围")
+        if fj.get("composition"):
+            pieces.append(str(fj["composition"]))
+        if fj.get("style"):
+            pieces.append(str(fj["style"]))
+        pieces.append("场景实拍，摄影级画质")
+        portrait_prompt = "，".join(p for p in pieces if p)
+        if len(portrait_prompt) > 200:
+            portrait_prompt = portrait_prompt[:200].rstrip("，")
+        summary = main_subject or scene_type
+        return {
+            "name": str(name)[:30],
+            "brand": brand,
+            "category": str(category)[:20],
+            "appearance": appearance[:300],
+            "packaging": packaging,
+            "text_on_package": text_on_package,
+            "key_features": kf,
+            "scene": scene_name,
+            "mood": scene_mood,
+            "portrait_prompt": portrait_prompt[:200],
+            "summary": str(summary)[:40],
+            "has_person": bool(fj.get("has_person", False)),
+            "_source": "v2_fast_json_v6_scene",
+        }
+
+    # other 兜底：当 description 为空时，用 environment_objects/scene 拼基本描述
+    desc = fj.get("description") or ""
+    if not desc.strip():
+        env_obj = fj.get("environment_objects") or []
+        if not isinstance(env_obj, list):
+            env_obj = [env_obj]
+        env_names = [str(e) for e in env_obj if e and len(str(e)) <= 25][:5]
+        scene_text = fj.get("scene") or ""
+        if env_names:
+            desc = f"{scene_text}场景中" + "、".join(env_names) if scene_text else "、".join(env_names)
+        elif scene_text and scene_text != "通用":
+            desc = f"{scene_text}场景"
+        else:
+            desc = "未识别"
+    kf_other = []
+    if desc != "未识别":
+        kf_other.append(desc[:30])
+    env_obj = fj.get("environment_objects") or []
+    if not isinstance(env_obj, list):
+        env_obj = [env_obj]
+    for o in env_obj:
+        if o and str(o) not in kf_other and len(str(o)) <= 40:
+            kf_other.append(str(o))
+    kf_other = kf_other[:8] or ["无法判断"]
+    return {
+        "name": desc[:30],
+        "brand": "无法判断",
+        "category": "非产品图",
+        "appearance": desc[:200],
+        "packaging": "无法判断",
+        "text_on_package": text_on_package,
+        "key_features": kf_other,
+        "scene": scene,
+        "mood": mood,
+        "portrait_prompt": f"{scene}，{mood}氛围，{desc}"[:200],
+        "summary": desc[:40],
+        "has_person": False,
+        "_source": "v2_fast_json_v4_other",
+    }
+
+
+def _assemble_old(idx: int, fj: dict, ocr_texts: list[str]) -> dict[str, Any]:
+    """旧扁平schema（兼容存量prompt或pro兜底输出）"""
+    portrait_prompt = _build_portrait_prompt_old(fj)
+    name = _infer_name_old(fj, ocr_texts)
+    brand = _infer_brand_old(fj, ocr_texts)
+    category = _infer_category_old(fj)
+    appearance = _build_appearance_old(fj)
+    key_features = _build_key_features_old(fj, ocr_texts)
+    scene = fj.get("scene") or "通用"
+    mood = fj.get("mood") or ""
+    packaging = "无法判断"
+    text_on_package = ocr_texts[:8]
+    if fj.get("has_person"):
+        up = fj.get("upper_wear") or "穿搭"
+        style = fj.get("style") or ""
+        summary = f"{style}{up}" if style and style not in up else up
+    elif brand != "无法判断" and name != brand:
+        summary = f"{brand} {name}"
+    else:
+        summary = name
     return {
         "name": name,
         "brand": brand,
@@ -293,15 +920,5 @@ def assemble_result(
         "portrait_prompt": portrait_prompt,
         "summary": summary,
         "_source": "v2_fast_json",
+        "has_person": bool(fj.get("has_person", False)),
     }
-
-
-def _build_summary(fj: dict, name: str, brand: str, category: str) -> str:
-    if fj.get("has_person"):
-        up = fj.get("upper_wear") or "穿搭"
-        style = fj.get("style") or ""
-        base = f"{style}{up}" if style and style not in up else up
-        return base
-    if brand != "无法判断" and name != brand:
-        return f"{brand} {name}"
-    return name

apps/worker/worker_app/tasks/vision/fast_path.py

@@ -23,10 +23,10 @@ logger = logging.getLogger(__name__)
 
 # 超时（可通过环境变量覆盖）
 _IMG_WORKERS = int(os.environ.get("VISION_V2_IMG_WORKERS", "8"))
-_FAST_TIMEOUT = float(os.environ.get("VISION_V2_FAST_TIMEOUT", "12"))
-_FAST_JSON_TIMEOUT = float(os.environ.get("VISION_V2_FAST_JSON_TIMEOUT", "12"))
+_FAST_TIMEOUT = float(os.environ.get("VISION_V2_FAST_TIMEOUT", "20"))
+_FAST_JSON_TIMEOUT = float(os.environ.get("VISION_V2_FAST_JSON_TIMEOUT", "20"))
 _OCR_TIMEOUT = float(os.environ.get("VISION_V2_OCR_TIMEOUT", "6"))
-_PRO_TIMEOUT = float(os.environ.get("VISION_V2_PRO_TIMEOUT", "25"))
+_PRO_TIMEOUT = float(os.environ.get("VISION_V2_PRO_TIMEOUT", "45"))
 
 _FALLBACK_RESULT = {
     "name": "未识别",
@@ -58,27 +58,29 @@ def analyze_image_v2(idx: int, img_url: str) -> dict[str, Any]:
 
     fj_result: dict[str, Any] | None = None
     ocr_result: list[str] = []
-    with ThreadPoolExecutor(max_workers=2) as pool:
-        f_fj = pool.submit(vlm_fast_json.call_fast_json, img_url, timeout=_FAST_JSON_TIMEOUT)
-        f_ocr = pool.submit(ocr_volc.call_ocr, img_url, timeout=_OCR_TIMEOUT)
-        try:
-            for fut in as_completed([f_fj, f_ocr], timeout=_FAST_TIMEOUT):
-                try:
-                    res = fut.result(timeout=1)
-                except Exception as e:
-                    logger.warning("[vision.v2] 图片 #%d 子任务异常: %s", idx, e)
-                    continue
-                if fut is f_fj and isinstance(res, dict):
-                    fj_result = res
-                elif fut is f_ocr and isinstance(res, list):
-                    ocr_result = res
-        except TimeoutError:
-            for f in (f_fj, f_ocr):
-                if not f.done():
-                    f.cancel()
-            logger.warning("[vision.v2] 图片 #%d fast路径超时(%.0fs)，走pro兜底", idx, _FAST_TIMEOUT)
-
-    fast_elapsed = time.time() - t0
+    fast_elapsed = 0.0
+    pool = ThreadPoolExecutor(max_workers=2)
+    f_fj = pool.submit(vlm_fast_json.call_fast_json, img_url, timeout=_FAST_JSON_TIMEOUT)
+    f_ocr = pool.submit(ocr_volc.call_ocr, img_url, timeout=_OCR_TIMEOUT)
+    try:
+        for fut in as_completed([f_fj, f_ocr], timeout=_FAST_TIMEOUT):
+            try:
+                res = fut.result(timeout=1)
+            except Exception as e:
+                logger.warning("[vision.v2] 图片 #%d 子任务异常: %s", idx, e)
+                continue
+            if fut is f_fj and isinstance(res, dict):
+                fj_result = res
+            elif fut is f_ocr and isinstance(res, list):
+                ocr_result = res
+    except TimeoutError:
+        for f in (f_fj, f_ocr):
+            if not f.done():
+                f.cancel()
+        logger.warning("[vision.v2] 图片 #%d fast路径超时(%.0fs)，走pro兜底", idx, _FAST_TIMEOUT)
+    finally:
+        fast_elapsed = time.time() - t0
+        pool.shutdown(wait=False)  # 不等待未完成的线程，避免计时膨胀
 
     if fj_result:
         assembled = assembler.assemble_result(idx, fj_result, ocr_result)

apps/worker/worker_app/tasks/vision/json_utils.py

@@ -0,0 +1,138 @@
+# -*- coding: utf-8 -*-
+"""VLM 返回文本的稳健 JSON 提取工具。
+
+背景：复杂门店图 VLM 输出经常被 max_tokens 截断（finish_reason=length），
+json.loads 失败后整个结果被丢弃，导致"未识别"。本工具提供：
+1. markdown 代码块剥离（含只开不闭的截断场景）
+2. 最外层 { } 切片
+3. 非法控制字符清理
+4. 直接 json.loads
+5. 截断 JSON 括号/引号栈补全修复
+6. 尾部逐字符截断重试（去除最后一个不完整 token 后修复）
+
+成功返回 dict；截断修复产物带 _partial=True 标记；彻底失败返回 None。
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import re
+
+logger = logging.getLogger(__name__)
+
+_CODE_FENCE_RE = re.compile(r"^```(?:json)?\s*\n?(.*?)\n?```\s*$", re.DOTALL)
+
+
+def _strip_code_fence(s: str) -> str:
+    s = s.strip()
+    m = _CODE_FENCE_RE.match(s)
+    if m:
+        return m.group(1).strip()
+    # 兼容开头 ```json 但结尾无 ```（截断场景）
+    if s.startswith("```"):
+        lines = s.split("\n")
+        if lines and lines[0].startswith("```"):
+            lines = lines[1:]
+        s = "\n".join(lines).strip()
+    return s
+
+
+def _repair_truncated_json(text: str) -> str:
+    """尝试补全被截断的JSON：维护 bracket/quote 栈，在末尾补闭合符。"""
+    stack: list[str] = []
+    in_string = False
+    escape = False
+    for ch in text:
+        if escape:
+            escape = False
+            continue
+        if ch == "\\" and in_string:
+            escape = True
+            continue
+        if ch == '"':
+            in_string = not in_string
+            continue
+        if in_string:
+            continue
+        if ch in "{[":
+            stack.append(ch)
+        elif ch == "}":
+            if stack and stack[-1] == "{":
+                stack.pop()
+        elif ch == "]":
+            if stack and stack[-1] == "[":
+                stack.pop()
+    repair = ""
+    if in_string:
+        repair += '"'
+    for opener in reversed(stack):
+        repair += "}" if opener == "{" else "]"
+    if repair:
+        logger.info(
+            "[json_utils] 截断JSON修复: 补全%d个闭合符 in_string=%s",
+            len(repair),
+            in_string,
+        )
+    return text + repair
+
+
+def _clean_invalid_chars(text: str) -> str:
+    """清理JSON中非法的控制字符（tab/newline 之外的 0x00-0x1f 段）。"""
+    return re.sub(r"[\x00-\x08\x0b\x0c\x0e-\x1f]", "", text)
+
+
+def extract_json_object(text: str) -> dict | None:
+    """从VLM返回文本中稳健提取JSON对象。
+
+    返回 dict 或 None。成功的 dict 可能带 _partial=True 标记，
+    表示原始文本被截断、经括号补全后得到的产物。
+    """
+    if not text or not isinstance(text, str):
+        return None
+    # 1. 剥离 markdown
+    text = _strip_code_fence(text)
+    # 2. 找最外层 { }
+    lpos = text.find("{")
+    if lpos < 0:
+        return None
+    rpos = text.rfind("}")
+    if rpos > lpos:
+        text = text[lpos : rpos + 1]
+    else:
+        # 截断场景：无任何闭合 }，取到末尾交给修复器
+        text = text[lpos:]
+    # 3. 清理非法控制字符
+    text = _clean_invalid_chars(text)
+    # 4. 直接 loads
+    try:
+        obj = json.loads(text)
+        return obj if isinstance(obj, dict) else None
+    except json.JSONDecodeError:
+        pass
+    # 5. 尝试截断修复
+    repaired = _repair_truncated_json(text)
+    try:
+        obj = json.loads(repaired)
+        if isinstance(obj, dict):
+            obj["_partial"] = True
+            return obj
+    except json.JSONDecodeError:
+        pass
+    # 6. 尾部逐字符截断重试（去除最后一个不完整 token）
+    for _ in range(50):
+        last_comma = repaired.rfind(",")
+        last_brace = max(repaired.rfind("}"), repaired.rfind("]"))
+        cut = max(last_comma, last_brace)
+        if cut < 10:
+            break
+        repaired = repaired[: cut + 1]
+        repaired = _repair_truncated_json(repaired)
+        try:
+            obj = json.loads(repaired)
+            if isinstance(obj, dict):
+                obj["_partial"] = True
+                return obj
+        except json.JSONDecodeError:
+            continue
+    return None

apps/worker/worker_app/tasks/vision/vlm_fallback.py

@@ -1,114 +1,27 @@
 # -*- coding: utf-8 -*-
-"""V2 兜底路径：qwen3.7-plus（阿里云百炼/DashScope）单图调用。
+"""V2 兜底路径：image_analysis（默认 qwen-vl-plus 视觉模型，fallback qwen3.7-plus / DashScope）单图调用。
 
 fast_json 超时/返回非 JSON/识别为空时，本路径单次调用兜底。
 设计要点：
-- 直接 httpx 直连 DashScope，不走 ai_client
-- enable_thinking=false + response_format=json_object
+- 通过 ai_router.get_vision_client() 获取 DoubaoClient 实例，不再自己拼 httpx 请求
+- enable_thinking=False + response_format=json_object
 - system prompt 优先读后台 viral_video_prompt_templates 配置，DB不可用时fallback到硬编码JSON schema
-- timeout=25s
-- API Key 从环境变量 DASHSCOPE_API_KEY 读取
-- 返回 dict 字段与旧 _normalize() 兼容，下游零改动
+- max_tokens 不传，使用 client 中 capability 的 DB 配置（避免硬编码截断 JSON）
+- timeout=30s
+- 返回 dict 统一走 assembler.assemble_result 组装，与 fast 路径输出格式完全一致
 """
 
 from __future__ import annotations
 
-import json
 import logging
-import os
 import time
 from typing import Any
 
-from . import _prompt
+from . import _prompt, assembler
 
 logger = logging.getLogger(__name__)
 
-_BASE_URL = "https://dashscope.aliyuncs.com/compatible-mode/v1"
-_PRO_MODEL = "qwen3.7-plus"
-_DEFAULT_TIMEOUT = 25
-_DEFAULT_MAX_TOKENS = 800
-
-
-def _api_key() -> str | None:
-    return os.environ.get("DASHSCOPE_API_KEY")
-
-
-def _assemble_pp(obj: dict[str, Any]) -> str:
-    """从 JSON 字段组装 portrait_prompt（60-100字人物穿搭描述，给 Seedream t2i 用）。"""
-    if not obj.get("has_person"):
-        name = obj.get("product_name") or "商品"
-        brand = obj.get("brand") or ""
-        kf = obj.get("key_features") or []
-        scene = obj.get("scene") or ""
-        mood = obj.get("mood") or ""
-        outfit = obj.get("outfit") or ""
-        if outfit:
-            return outfit
-        pieces = []
-        if brand:
-            pieces.append(brand)
-        pieces.append(str(name))
-        if isinstance(kf, list):
-            pieces.extend(str(x) for x in kf[:2] if x)
-        if mood:
-            pieces.append(str(mood) + "氛围")
-        if scene:
-            pieces.append(str(scene) + "场景")
-        pieces.append("产品特写")
-        p = "，".join(x for x in pieces if x)
-        return p if len(p) >= 10 else "产品展示图，特写镜头"
-
-    parts: list[str] = []
-    gender = obj.get("gender") or ""
-    age = obj.get("age_range") or ""
-    subj = ""
-    if age == "儿童":
-        subj = "小女孩" if gender == "女" else ("小男孩" if gender == "男" else "儿童")
-    elif age == "青少年":
-        subj = "少女" if gender == "女" else ("少年" if gender == "男" else "青少年")
-    else:
-        prefix_map = {"青年": "年轻", "中年": "中年", "老年": "老年"}
-        gw = {"男": "男性", "女": "女性"}.get(gender, "")
-        prefix = prefix_map.get(age, "")
-        subj = (prefix + gw) if (prefix or gw) else "人物"
-    parts.append(f"一位{subj}")
-
-    outfit = obj.get("outfit") or ""
-    if outfit:
-        parts.append(f"身着{outfit}")
-
-    hair = obj.get("hair") or ""
-    if hair:
-        parts.append(str(hair))
-
-    pose = obj.get("pose") or ""
-    expr = obj.get("expression") or ""
-    det = []
-    if expr and expr not in ("自然", "平静"):
-        det.append(f"神情{expr}")
-    if pose and pose not in ("站立",):
-        det.append(str(pose))
-    if det:
-        parts.append("，".join(det))
-
-    style_parts = []
-    mood = obj.get("mood") or ""
-    scene = obj.get("scene") or ""
-    if mood:
-        style_parts.append(str(mood))
-    if scene and scene != "通用":
-        style_parts.append(str(scene))
-    if style_parts:
-        parts.append("".join(style_parts) + "风格")
-    else:
-        parts.append("人像写真")
-
-    full = "，".join(p for p in parts if p)
-    if len(full) < 40:
-        full += "，自然光线下人像特写，画面清晰"
-    if len(full) > 120:
-        full = full[:120].rstrip("，") + "。"
-    return full
+_DEFAULT_TIMEOUT = 45
 
 
 def call_pro_vlm(
@@ -116,109 +29,87 @@ def call_pro_vlm(
     idx: int,
     *,
     timeout: int = _DEFAULT_TIMEOUT,
+    max_tokens: int | None = None,
 ) -> dict[str, Any] | None:
+    """max_tokens 默认 None：不显式传参，使用 client 内 capability 的 DB 配置。"""
     t0 = time.time()
-    import httpx
 
-    api_key = _api_key()
-    if not api_key:
-        logger.warning("[vision.v2] pro DASHSCOPE_API_KEY 未配置，跳过")
+    try:
+        from packages.shared.ai_router import ai_router
+
+        client = ai_router.get_vision_client("image_analysis", variant="fallback")
+        if not client or not client.is_available:
+            logger.warning("[vision.v2] pro vision client 不可用，跳过")
+            return None
+    except Exception as e:
+        logger.warning("[vision.v2] ai_router 获取失败: %s", e)
         return None
 
     system_prompt, user_prompt = _prompt.resolve_pro_prompt()
 
-    payload: dict[str, Any] = {
-        "model": _PRO_MODEL,
-        "messages": [
-            {"role": "system", "content": system_prompt},
-            {
-                "role": "user",
-                "content": [
-                    {"type": "image_url", "image_url": {"url": img_url}},
-                    {"type": "text", "text": user_prompt},
-                ],
-            },
-        ],
-        "temperature": 0.3,
-        "max_tokens": _DEFAULT_MAX_TOKENS,
-        "stream": False,
-        "enable_thinking": False,
-        "response_format": {"type": "json_object"},
-    }
+    messages = [
+        {"role": "system", "content": system_prompt},
+        {
+            "role": "user",
+            "content": [
+                {"type": "image_url", "image_url": {"url": img_url}},
+                {"type": "text", "text": user_prompt},
+            ],
+        },
+    ]
+
     try:
-        r = httpx.post(
-            f"{_BASE_URL}/chat/completions",
-            headers={"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"},
-            json=payload,
-            timeout=timeout,
-        )
-        elapsed = time.time() - t0
-        if r.status_code != 200:
-            logger.warning("[vision.v2] pro HTTP %d elapsed=%.1fs body=%s", r.status_code, elapsed, r.text[:200])
-            return None
-        data = r.json()
-        raw = (data.get("choices") or [{}])[0].get("message", {}).get("content")
-        if not raw:
-            logger.warning("[vision.v2] pro 返回空 elapsed=%.1fs", elapsed)
-            return None
-        usage = data.get("usage") or {}
-        reasoning_tokens = usage.get("reasoning_tokens", 0)
-        ctd = usage.get("completion_tokens_details") or {}
-        if not reasoning_tokens:
-            reasoning_tokens = ctd.get("reasoning_tokens", 0)
-        logger.info(
-            "[vision.v2] pro 完成 model=%s elapsed=%.1fs in=%d out=%d reasoning=%d",
-            _PRO_MODEL,
-            elapsed,
-            usage.get("prompt_tokens", 0),
-            usage.get("completion_tokens", 0),
-            reasoning_tokens,
-        )
-        s = raw.strip()
-        if s.startswith("```"):
-            lines = s.split("\n")
-            if lines and lines[0].startswith("```"):
-                lines = lines[1:]
-            if lines and lines[-1].strip().startswith("```"):
-                lines = lines[:-1]
-            s = "\n".join(lines).strip()
-        l, rr = s.find("{"), s.rfind("}")
-        if l >= 0 and rr > l:
-            s = s[l : rr + 1]
-        try:
-            obj = json.loads(s)
-        except json.JSONDecodeError:
-            logger.warning("[vision.v2] pro JSON 解析失败 head=%s", raw[:200])
-            return None
-        if not isinstance(obj, dict):
-            return None
-
-        pp = _assemble_pp(obj)
-        kf = obj.get("key_features")
-        if not isinstance(kf, list):
-            kf = [str(kf)] if kf else ["无法判断"]
-        else:
-            kf = [str(x) for x in kf if x] or ["无法判断"]
-
-        name = obj.get("product_name") or "未识别"
-        if obj.get("has_person") and (not name or name == "未识别"):
-            name = obj.get("outfit") or "人物穿搭"
-        brand = obj.get("brand") or "无法判断"
-        category = obj.get("category") or ("服饰" if obj.get("has_person") else "非产品图")
-        return {
-            "name": str(name),
-            "brand": str(brand),
-            "category": str(category),
-            "appearance": str(obj.get("outfit") or "无法判断"),
-            "packaging": "无法判断",
-            "text_on_package": [],
-            "key_features": kf[:6],
-            "scene": str(obj.get("scene") or "通用"),
-            "mood": str(obj.get("mood") or ""),
-            "portrait_prompt": pp,
-            "summary": str(name),
-            "_source": "vlm_pro",
+        call_kwargs: dict[str, Any] = {
+            "messages": messages,
+            "images": None,  # 图片已在 messages 中
+            "temperature": 0.3,
+            "timeout": timeout,
+            "enable_thinking": False,
+            "response_format": {"type": "json_object"},
         }
+        # pro fallback：显式4000 tokens给复杂门店图留足空间
+        call_kwargs["max_tokens"] = max_tokens if max_tokens is not None else 4000
+
+        from .json_utils import extract_json_object
+
+        raw = None
+        obj = None
+        for _outer in range(2):
+            kw = dict(call_kwargs)
+            if _outer == 1:
+                kw.pop("response_format", None)
+                msgs2 = [dict(messages[0]), dict(messages[1])]
+                cont = [dict(c) for c in list(msgs2[1]["content"])]
+                cont[-1] = {"type": "text", "text": user_prompt + "\n严格只输出JSON对象，不要解释或markdown。"}
+                msgs2[1] = {"role": "user", "content": cont}
+                kw["messages"] = msgs2
+            raw = client.vision_completion(**kw)
+            if not raw:
+                logger.warning("[vision.v2] pro 返回空 outer=%s", _outer)
+                continue
+            obj = extract_json_object(raw)
+            if obj is not None:
+                break
+            logger.warning("[vision.v2] pro 非JSON(100字) outer=%s: %s", _outer, raw[:100])
+
+        elapsed = time.time() - t0
+        if obj is None:
+            logger.warning("[vision.v2] pro 两次均未得到JSON elapsed=%.1fs", elapsed)
+            return None
+        if obj.get("_partial"):
+            logger.warning("[vision.v2] pro 返回截断JSON(partial) elapsed=%.1fs", elapsed)
+        logger.info(
+            "[vision.v2] pro 完成 model=%s elapsed=%.1fs type=%s",
+            client.model,
+            elapsed,
+            obj.get("type"),
+        )
+
+        # 通过assembler统一组装，兼容v4嵌套schema和旧扁平schema
+        result = assembler.assemble_result(idx, obj, [])
+        result["_source"] = "vlm_pro"
+        result["_fallback_used"] = True
+        return result
     except Exception as e:
         elapsed = time.time() - t0
         logger.warning("[vision.v2] pro 异常 elapsed=%.1fs err=%s", elapsed, e, exc_info=True)

apps/worker/worker_app/tasks/vision/vlm_fast_json.py

@@ -1,22 +1,20 @@
 # -*- coding: utf-8 -*-
-"""V2 快速路径：qwen3.8-flash（阿里云百炼/DashScope）强约束 JSON-only 调用。
+"""V2 快速路径：image_analysis capability（默认 qwen-vl-plus 视觉模型 / DashScope）强约束 JSON-only 调用。
 
 目标：替代"人体属性/商品检测/图像标签"三个火山不存在的专用云端 API。
 设计要点：
-- 直接用 httpx 发最小 payload 到 DashScope OpenAI 兼容 endpoint，不走 ai_client 包装
-- enable_thinking=false 关闭推理链（reasoning 是延迟主因）
+- 通过 ai_router.get_vision_client() 获取 DoubaoClient 实例，不再自己拼 httpx 请求
+- enable_thinking=False 关闭推理链（reasoning 是延迟主因）
 - response_format=json_object 强约束JSON输出
 - system prompt 优先读后台 viral_video_prompt_templates 配置，DB不可用时fallback到硬编码JSON schema
-- max_tokens=350、temperature=0.1（稳定输出 JSON）
-- timeout=12s（失败由外层走 pro 兜底）
-- API Key 从环境变量 DASHSCOPE_API_KEY 读取
+- max_tokens 不传，使用 client 中 capability 的 DB 配置（避免硬编码截断 JSON）
+- temperature=0.1（稳定输出 JSON）
+- timeout=15s（失败由外层走 pro 兜底）
 """
 
 from __future__ import annotations
 
-import json
 import logging
-import os
 import time
 from typing import Any
 
@@ -24,124 +22,99 @@ from . import _prompt
 
 logger = logging.getLogger(__name__)
 
-# DashScope OpenAI 兼容 endpoint
-_BASE_URL = "https://dashscope.aliyuncs.com/compatible-mode/v1"
-_FAST_MODEL = "qwen3.8-flash"
-_DEFAULT_TIMEOUT = 12
-_DEFAULT_MAX_TOKENS = 350
-
-
-def _api_key() -> str | None:
-    return os.environ.get("DASHSCOPE_API_KEY")
-
-
-def _strip_code_fence(s: str) -> str:
-    s = s.strip()
-    if s.startswith("```"):
-        lines = s.split("\n")
-        if lines and lines[0].startswith("```"):
-            lines = lines[1:]
-        if lines and lines[-1].strip().startswith("```"):
-            lines = lines[:-1]
-        s = "\n".join(lines).strip()
-    return s
+_DEFAULT_TIMEOUT = 20
 
 
 def call_fast_json(
     img_url: str,
     *,
     timeout: int = _DEFAULT_TIMEOUT,
-    max_tokens: int = _DEFAULT_MAX_TOKENS,
+    max_tokens: int | None = None,
 ) -> dict[str, Any] | None:
-    """调用 qwen3.8-flash 返回结构化 dict；失败/非 JSON 返回 None。"""
-    t0 = time.time()
-    import httpx
+    """调用 vision client 返回结构化 dict；失败/非 JSON 返回 None。
 
-    api_key = _api_key()
-    if not api_key:
-        logger.warning("[vision.v2] DASHSCOPE_API_KEY 未配置，跳过 fast_json")
+    max_tokens 默认 None：不显式传参，使用 client 内 capability 的 DB 配置；
+    显式传入时作为覆盖。
+    """
+    t0 = time.time()
+
+    try:
+        from packages.shared.ai_router import ai_router
+
+        client = ai_router.get_vision_client("image_analysis", variant="primary")
+        if not client or not client.is_available:
+            logger.warning("[vision.v2] vision client 不可用，跳过 fast_json")
+            return None
+    except Exception as e:
+        logger.warning("[vision.v2] ai_router 获取失败: %s", e)
         return None
 
     system_prompt, user_prompt = _prompt.resolve_fast_prompt()
 
-    url = f"{_BASE_URL}/chat/completions"
-    payload: dict[str, Any] = {
-        "model": _FAST_MODEL,
-        "messages": [
-            {"role": "system", "content": system_prompt},
-            {
-                "role": "user",
-                "content": [
-                    {"type": "image_url", "image_url": {"url": img_url}},
-                    {"type": "text", "text": user_prompt},
-                ],
-            },
-        ],
-        "temperature": 0.1,
-        "max_tokens": max_tokens,
-        "stream": False,
-        "enable_thinking": False,
-        "response_format": {"type": "json_object"},
-    }
+    messages = [
+        {"role": "system", "content": system_prompt},
+        {
+            "role": "user",
+            "content": [
+                {"type": "image_url", "image_url": {"url": img_url}},
+                {"type": "text", "text": user_prompt},
+            ],
+        },
+    ]
+
     try:
-        resp = httpx.post(
-            url,
-            headers={"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"},
-            json=payload,
-            timeout=timeout,
-        )
-        elapsed = time.time() - t0
-        if resp.status_code == 400 and "enable_thinking" in resp.text[:300].lower():
-            logger.warning("[vision.v2] fast_json HTTP 400 thinking 参数不兼容，重试 elapsed=%.1fs", elapsed)
-            payload.pop("enable_thinking", None)
-            resp = httpx.post(
-                url,
-                headers={"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"},
-                json=payload,
-                timeout=timeout,
-            )
-            elapsed = time.time() - t0
-        if resp.status_code != 200:
+        call_kwargs: dict[str, Any] = {
+            "messages": messages,
+            "images": None,  # 图片已在 messages 中
+            "temperature": 0.1,
+            "timeout": timeout,
+            "enable_thinking": False,
+            "response_format": {"type": "json_object"},
+        }
+        if max_tokens is not None:
+            call_kwargs["max_tokens"] = max_tokens
+
+        # 双重防护：第1次正常调用；第2次去掉json_object强约束（部分模型在该约束下
+        # 反而幻觉），并加严格指令。解析全部走 json_utils，截断partial产物可用。
+        from .json_utils import extract_json_object
+
+        raw = None
+        obj = None
+        for _outer in range(2):
+            kw = dict(call_kwargs)
+            if _outer == 1:
+                kw.pop("response_format", None)
+                msgs2 = [dict(messages[0]), dict(messages[1])]
+                cont = list(msgs2[1]["content"])
+                cont = [dict(c) for c in cont]
+                cont[-1] = {"type": "text", "text": user_prompt + "\n严格只输出JSON对象，不要解释或markdown。"}
+                msgs2[1] = {"role": "user", "content": cont}
+                kw["messages"] = msgs2
+            raw = client.vision_completion(**kw)
+            if not raw:
+                logger.warning("[vision.v2] fast_json 返回空 outer=%s", _outer)
+                continue
+            obj = extract_json_object(raw)
+            if obj is not None:
+                break
             logger.warning(
-                "[vision.v2] fast_json HTTP %d elapsed=%.1fs body=%s", resp.status_code, elapsed, resp.text[:200]
+                "[vision.v2] fast_json 非JSON(100字) outer=%s: %s",
+                _outer,
+                raw[:100],
             )
+
+        elapsed = time.time() - t0
+        if obj is None:
+            logger.warning("[vision.v2] fast_json 两次均未得到JSON elapsed=%.1fs", elapsed)
             return None
-        data = resp.json()
-        raw = (data.get("choices") or [{}])[0].get("message", {}).get("content")
-        if not raw:
-            logger.warning("[vision.v2] fast_json 返回空 elapsed=%.1fs", elapsed)
-            return None
-        usage = data.get("usage") or {}
-        reasoning_tokens = usage.get("reasoning_tokens", 0)
-        ctd = usage.get("completion_tokens_details") or {}
-        if not reasoning_tokens:
-            reasoning_tokens = ctd.get("reasoning_tokens", 0)
+        if obj.get("_partial"):
+            logger.warning("[vision.v2] fast_json 返回截断JSON(partial) elapsed=%.1fs", elapsed)
         logger.info(
-            "[vision.v2] fast_json 完成 model=%s elapsed=%.1fs in=%d out=%d reasoning=%d",
-            _FAST_MODEL,
-            elapsed,
-            usage.get("prompt_tokens", 0),
-            usage.get("completion_tokens", 0),
-            reasoning_tokens,
-        )
-        text = _strip_code_fence(raw)
-        lpos, r = text.find("{"), text.rfind("}")
-        if lpos >= 0 and r > lpos:
-            text = text[lpos : r + 1]
-        try:
-            obj = json.loads(text)
-        except json.JSONDecodeError:
-            logger.warning("[vision.v2] fast_json JSON 解析失败 elapsed=%.1fs head=%s", elapsed, raw[:200])
-            return None
-        if not isinstance(obj, dict):
-            logger.warning("[vision.v2] fast_json 非 dict: %s", type(obj))
-            return None
-        logger.info(
-            "[vision.v2] fast_json 完成 elapsed=%.1fs has_person=%s has_product=%s category=%s",
+            "[vision.v2] fast_json 完成 model=%s elapsed=%.1fs has_person=%s type=%s",
+            client.model,
             elapsed,
             obj.get("has_person"),
-            obj.get("has_product"),
-            obj.get("category"),
+            obj.get("type"),
         )
         return obj
     except Exception as e:

packages/application/cosyvoice_service.py

@@ -351,12 +351,25 @@ class CosyVoiceService:
                 用于私有 bucket 下，将裸 URL 转为预签名 URL，
                 确保 CosyVoice 服务器能下载参考音频.
         """
+        # 优先从 ai_router 获取 DB 配置
+        _router_key, _router_url, _router_model = "", "", ""
+        try:
+            from packages.shared.ai_router import ai_router
+
+            tts_client = ai_router.get_tts_client("tts")
+            if tts_client and tts_client.is_available:
+                _router_key = tts_client.api_key
+                _router_url = tts_client.base_url
+                _router_model = tts_client.model
+        except Exception:
+            pass
+
         settings = get_shared_settings()
 
-        self._api_key = api_key or settings.cosyvoice_api_key
-        self._base_url = base_url or settings.cosyvoice_base_url
-        self._model = model or settings.cosyvoice_model
-        self._clone_model = clone_model or getattr(settings, "cosyvoice_clone_model", "voice-enrollment")
+        self._api_key = api_key or _router_key or settings.cosyvoice_api_key
+        self._base_url = base_url or _router_url or settings.cosyvoice_base_url
+        self._model = model or _router_model or settings.cosyvoice_model
+        self._clone_model = clone_model or getattr(settings, "cosyvoice_clone_model", "")
         self._audio_url_signer = audio_url_signer
 
         # base_url 规范化：去掉末尾的路径残留（兼容旧版配置）

packages/application/viral_video/prompts.py

@@ -110,10 +110,12 @@ _INTENT_SYSTEM = f"""你负责理解用户的营销意图。用户给的文案
 
 _INTENT_USER = """用户原始文案：{user_copy_text}
 所属行业：{industry}
+营销目的：{marketing_purpose}
 图片分析结果（供参考）：
 {image_analysis}
+图片类型推断：{image_category_hint}
 
-请理解用户意图，按标签格式输出。"""
+请理解用户意图，按标签格式输出。注意：theme和emotion_tone应与图片类型和营销目的匹配——门店类图片偏向"门店探店/到店体验"，商品图偏向"好物分享/产品种草"，人物图偏向"穿搭/人物故事"。"""
 
 _INTENT_EXAMPLE = """<intent_summary>一款厨房去油污神器，喷一喷油污就掉</intent_summary>
 <core_messages>
@@ -176,7 +178,7 @@ _FUSION_EXAMPLE = """<title>厨房重油污，别再用洗洁精硬擦了</title
 <segment duration_sec="4" image_index="0">39块钱625ml，厨房重油污的可以试一瓶</segment>
 </script_segments>
 <voiceover_script>这油污我真的忍很久了，用洗洁精擦半天都没用。后来换了这个大公鸡头油污净，喷上等几分钟，一擦就干净。39块钱625ml，厨房重油污的可以试一瓶。</voiceover_script>
-<overview_theme>厨房油污清洁好物分享</overview_theme>
+<overview_theme>厨房好物分享·产品种草</overview_theme>
 <scene_and_lighting>简洁明亮的厨房台面场景，自然光从窗户洒入，色调温暖柔和，突出产品白色瓶身与去油污对比效果。</scene_and_lighting>
 <word_count>58</word_count>
 <estimated_duration>13</estimated_duration>"""
@@ -215,6 +217,8 @@ _STORYBOARD_USER = """目标时长：{duration}秒
 图片分析结果：
 {image_analysis}
 
+重要：overview_theme 必须与图片实际内容和营销目的匹配。门店/餐饮/服务类图片用"门店探店·到店体验"；商品图用"好物分享·产品种草"；人物图用"穿搭分享·人物故事"；场景图用"空间体验·场景氛围"。不要对所有图片都使用"好物分享"。
+
 请按标签格式输出分镜。"""
 
 _STORYBOARD_EXAMPLE = """<clips>

packages/application/viral_video/reviewer.py

@@ -55,9 +55,14 @@ _LOCATIONS = ["title", "hook", "body_points", "cta", "script_segments"]
 class Reviewer:
     def __init__(self, client=None):
         if client is None:
-            from packages.shared.ai_client import get_doubao_client
+            try:
+                from packages.shared.ai_router import ai_router
 
-            client = get_doubao_client()
+                client = ai_router.get_llm_client("copy_review")
+            except Exception:
+                from packages.shared.ai_client import get_doubao_client
+
+                client = get_doubao_client()
         self.client = client
 
     # ── 审核 ────────────────────────────────────────────────────────────
@@ -96,6 +101,7 @@ class Reviewer:
             ],
             temperature=0.2,
             max_tokens=1024,
+            timeout=25,
         )
         if not raw:
             return None
@@ -242,6 +248,7 @@ class Reviewer:
             ],
             temperature=0.5,
             max_tokens=2048,
+            timeout=25,
         )
         if not raw:
             return self._rule_fix(fusion, review)

packages/config/base.py

@@ -80,54 +80,46 @@ class SharedSettings(BaseSettings):
 
     # ── CosyVoice (阿里云百炼语音合成) ───────────────────────────────────
     cosyvoice_api_key: str = ""
-    cosyvoice_base_url: str = "https://dashscope.aliyuncs.com/api/v1"
-    cosyvoice_model: str = "cosyvoice-v3-flash"
-    cosyvoice_voice: str = "longxiaochun_v3"  # 默认音色（v3 系列系统音色带 _v3 后缀）
+    cosyvoice_base_url: str = ""
+    cosyvoice_model: str = ""
+    cosyvoice_voice: str = "longxiaochun_v3"
     cosyvoice_sample_rate: int = 22050
-    cosyvoice_format: str = "mp3"  # 输出格式：mp3/wav/pcm
+    cosyvoice_format: str = "mp3"
     # 音色克隆模型名（固定为 voice-enrollment）
-    cosyvoice_clone_model: str = "voice-enrollment"
+    cosyvoice_clone_model: str = ""
 
     # ── 豆包大模型（火山引擎方舟） ────────────────────────────────────────
+    # AI模型路由化：model/base_url 默认值清空，由 DB ai_models/ai_capability_configs 配置驱动。
+    # 环境变量仍可覆盖（兼容旧部署）；无任何配置时 ai_router fallback 提供最终默认值。
     doubao_api_key: str = ""
-    doubao_model: str = "doubao-seed-2-1-pro-260915"  # 推理模型（Seed 2.1 Pro，深度思考+多模态；原 seed-1-6 已下线）
-    doubao_fast_model: str = (
-        "doubao-seed-2-1-pro-260915"  # #2181: lite方舟侧100%超时，默认fast_model也走pro；方舟恢复lite后通过ENV DOUBAO_FAST_MODEL切回
-    )
-    doubao_base_url: str = "https://ark.cn-beijing.volces.com/api/v3"
-    doubao_timeout: int = 45  # #2180: 方舟LLM高峰期响应6-8s，原30s太紧提到45s
-    doubao_max_retries: int = 1  # #2180: timeout调大后一次调用就够，1次重试防偶发抖动；避免6次重试叠加到351s
-    doubao_vision_model: str = (
-        "doubao-seed-2-1-pro-260915"  # 高精度视觉（Seed 2.1 Pro 原生多模态；原 vision-pro-250328 已下线）
-    )
-    doubao_vision_lite_model: str = (
-        "doubao-seed-2-1-lite-260915"  # 快速视觉（Seed 2.1 Lite 原生多模态；原 vision-lite-250315 不可用）
-    )
-    doubao_vision_use_lite: bool = True  # #2188: lite恢复稳定，爆款视频默认lite-first提速（20-30s）
-    doubao_embedding_model: str = "doubao-embedding-vision-251215"  # 多模态向量化（原 large-text-240915 已 Retiring）
-    doubao_video_model: str = "doubao-seedance-2-5-260628"
-    doubao_video_timeout: int = 600  # 视频生成轮询总超时（秒）
-    doubao_video_poll_interval: int = 10  # 轮询间隔（秒）
-    doubao_image_model: str = (
-        "doubao-seedream-5-0-flash-260915"  # #2173: 信任链 Seedream 改 flash 模型（实测 pro 46.5s→flash 13s；pro AI化图仍被Seedance拦截）
-    )
-    doubao_image_size: str = "1K"  # #2173: 1K 已足够做 Seedance 参考图，2K 在 flash 下也 22s，1K 13s
-    doubao_image_timeout: int = 60  # #2173: flash+1K 通常15s内，给60s余量
-    doubao_trust_chain_enabled: bool = (
-        True  # #2173: 信任链总开关；若Seedream产物仍被Seedance拦截，可配 False 关闭直接t2v降级
-    )
+    doubao_model: str = ""
+    doubao_fast_model: str = ""
+    doubao_base_url: str = ""
+    doubao_timeout: int = 45
+    doubao_max_retries: int = 3
+    doubao_vision_model: str = ""
+    doubao_vision_lite_model: str = ""
+    doubao_vision_use_lite: bool = True
+    doubao_embedding_model: str = ""
+    doubao_video_model: str = ""
+    doubao_video_timeout: int = 600
+    doubao_video_poll_interval: int = 10
+    doubao_image_model: str = ""
+    doubao_image_size: str = "1K"
+    doubao_image_timeout: int = 60
+    doubao_trust_chain_enabled: bool = True
 
     # ── DashScope (阿里云百炼 Wan 3.0 等) ─────────────────────────────────
     dashscope_api_key: str = ""
-    dashscope_base_url: str = "https://dashscope.aliyuncs.com/api/v1"
-    dashscope_video_timeout: int = 900  # Wan 视频任务轮询总超时（秒）
+    dashscope_base_url: str = ""
+    dashscope_video_timeout: int = 900
     dashscope_video_poll_interval: int = 10
 
     # ── MediaKit (火山引擎 AI 媒体工具) ──────────────────────────────────
     mediakit_api_key: str = ""
-    mediakit_base_url: str = "https://mediakit.cn-beijing.volces.com/api/v1"
+    mediakit_base_url: str = ""
     mediakit_timeout: int = 60
-    mediakit_cover_enabled: bool = False  # 封面抽帧是否走MediaKit(默认false走本地ffmpeg+cv2，<2s完成)
+    mediakit_cover_enabled: bool = False
 
     # ── 积分/会员系统 (#1895) ────────────────────────────────────────────
     # 积分系统总开关（产品要求 #1895：暂停积分系统但保留全部代码/表/接口）。

packages/domain/viral_video.py

@@ -191,13 +191,45 @@ class ViralVideoJob:
         self.updated_at = datetime.now(timezone.utc)
 
     def resume_from_image_analyzed(self, **kwargs) -> None:
-        if self.status not in (ViralVideoStatus.IMAGE_ANALYZED, ViralVideoStatus.PENDING):
+        """阶段2入口：允许从 IMAGE_ANALYZED/PENDING 首次进入，也允许从 COPY_GENERATED/COMPLETED/FAILED 重新生成文案。
+
+        重新生成时清空上一轮文案产物（copy_result/intent_result/storyboard/generated_copy_text），
+        并重置 completed_at/result_video_url/error_msg，确保前端轮询能看到新的阶段2进度。
+        """
+        _allowed = (
+            ViralVideoStatus.IMAGE_ANALYZED,
+            ViralVideoStatus.PENDING,
+            ViralVideoStatus.COPY_GENERATED,
+            ViralVideoStatus.COMPLETED,
+            ViralVideoStatus.FAILED,
+        )
+        if self.status not in _allowed:
             raise ValueError(f"Cannot resume from {self.status} to copy-gen")
+        _is_regen = self.status in (
+            ViralVideoStatus.COPY_GENERATED,
+            ViralVideoStatus.COMPLETED,
+            ViralVideoStatus.FAILED,
+        )
         for k, v in kwargs.items():
             if hasattr(self, k) and v not in (None, "", []):
                 setattr(self, k, v)
+        if _is_regen:
+            # 清空上一轮文案/视频产物，避免前端拿到旧数据
+            self.intent_result = None
+            self.copy_result = None
+            self.storyboard = None
+            self.generated_copy_text = ""
+            self.result_video_url = ""
+            self.current_stage = ""
+            self.phase_message = ""
+            self.error_msg = ""
+            self.completed_at = None
+            self.heartbeat_at = None
+        _now = datetime.now(timezone.utc)
+        self.started_at = _now
+        self.heartbeat_at = _now
         self.status = ViralVideoStatus.RUNNING
-        self.updated_at = datetime.now(timezone.utc)
+        self.updated_at = _now
 
     def resume_from_copy_generated(self, edited_copy: str | None = None) -> None:
         """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。"""
@@ -206,14 +238,20 @@ class ViralVideoJob:
         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
+        _now = datetime.now(timezone.utc)
+        self.started_at = _now
+        self.heartbeat_at = _now
         self.status = ViralVideoStatus.RUNNING
-        self.updated_at = datetime.now(timezone.utc)
+        self.updated_at = _now
 
     def resume_from_confirm(self) -> None:
         if self.status != ViralVideoStatus.WAIT_USER_CONFIRM:
             raise ValueError(f"Cannot resume from {self.status}")
+        _now = datetime.now(timezone.utc)
+        self.started_at = _now
+        self.heartbeat_at = _now
         self.status = ViralVideoStatus.RUNNING
-        self.updated_at = datetime.now(timezone.utc)
+        self.updated_at = _now
 
     def mark_completed(self, video_url: str) -> None:
         self.status = ViralVideoStatus.COMPLETED

packages/shared/ai_client.py

@@ -170,14 +170,35 @@ class DoubaoClient:
     未配置 API Key 时 is_available 为 False，调用方应降级处理。
     """
 
-    def __init__(self) -> None:
+    def __init__(
+        self,
+        api_key: str = "",
+        base_url: str = "",
+        model: str = "",
+        timeout: int = 0,
+        max_retries: int = 0,
+        max_tokens: int | None = None,
+        temperature: float | None = None,
+        extra_params: dict | None = None,
+        provider: str = "volcengine",
+    ) -> None:
         settings = get_shared_settings()
-        self.api_key: str = settings.doubao_api_key
-        self.model: str = settings.doubao_model
-        self.base_url: str = settings.doubao_base_url.rstrip("/")
-        self.timeout: int = settings.doubao_timeout
-        self.max_retries: int = settings.doubao_max_retries
+        self.provider: str = provider
+        if provider == "dashscope":
+            self.api_key: str = api_key or getattr(settings, "dashscope_api_key", "")
+            self.model: str = model or getattr(settings, "dashscope_model", "")
+            self.base_url: str = (base_url or getattr(settings, "dashscope_base_url", "")).rstrip("/")
+        else:  # volcengine (default)
+            self.api_key = api_key or settings.doubao_api_key
+            self.model = model or settings.doubao_model
+            self.base_url = (base_url or settings.doubao_base_url).rstrip("/")
+        self.timeout: int = timeout or settings.doubao_timeout
+        self.max_retries: int = max_retries or settings.doubao_max_retries
+        self.max_tokens: int | None = max_tokens
+        self.temperature: float | None = temperature
+        self.extra_params: dict = extra_params or {}
         self.vision_model: str = settings.doubao_vision_model
+        self.last_finish_reason: str = ""
         self.vision_lite_model: str = settings.doubao_vision_lite_model
         self.fast_model: str = settings.doubao_fast_model
         self.embedding_model: str = settings.doubao_embedding_model
@@ -190,6 +211,13 @@ class DoubaoClient:
         # 最近一次图片生成的详细错误，供上层读取
         self.last_image_error: dict = {}
 
+    def _resolve_timeout(self, timeout) -> "httpx.Timeout":
+        """将整数超时转为 httpx.Timeout，区分 connect/read/write/pool，避免 read 卡到 TCP 120s 默认值."""
+        if isinstance(timeout, httpx.Timeout):
+            return timeout
+        t = int(timeout) if timeout else 60
+        return httpx.Timeout(connect=10, read=max(t, 10), write=10, pool=5)
+
     def embed_text(self, text: str, timeout: int | None = None) -> list[float] | None:
         """调用豆包文本 Embedding API，返回浮点向量；失败返回 None。"""
         if not self.is_available or not text or not text.strip():
@@ -240,16 +268,17 @@ class DoubaoClient:
         self,
         messages: list[dict[str, str]],
         temperature: float = 0.7,
-        max_tokens: int = 1024,
+        max_tokens: int | None = None,
         model: str | None = None,
         timeout: int | None = None,
+        **kwargs,
     ) -> Optional[str]:
         """调用 Chat Completion 接口.
 
         Args:
             messages: 对话消息列表，[{"role": "user"/"system"/"assistant", "content": "..."}]
             temperature: 采样温度，0-2，默认0.7
-            max_tokens: 最大生成token数，默认1024
+            max_tokens: 最大生成token数，默认 None（使用实例 self.max_tokens DB 配置，兜底 1024）
 
         Returns:
             模型返回的文本内容，失败返回 None
@@ -262,18 +291,24 @@ class DoubaoClient:
             "Authorization": f"Bearer {self.api_key}",
             "Content-Type": "application/json",
         }
+        effective_max_tokens = max_tokens if max_tokens is not None else (self.max_tokens or 1024)
         payload: dict[str, Any] = {
             "model": model or self.model,
             "messages": messages,
             "temperature": temperature,
-            "max_tokens": max_tokens,
+            "max_tokens": effective_max_tokens,
         }
+        # 合并实例级额外参数和调用方传入的额外参数
+        if self.extra_params:
+            payload.update(self.extra_params)
+        if kwargs:
+            payload.update(kwargs)
 
         last_error: Optional[Exception] = None
         _t0 = time.time()
         for attempt in range(self.max_retries + 1):
             try:
-                _req_timeout = timeout if timeout is not None else self.timeout
+                _req_timeout = self._resolve_timeout(timeout if timeout is not None else self.timeout)
                 response = httpx.post(
                     url,
                     headers=headers,
@@ -282,7 +317,24 @@ class DoubaoClient:
                 )
                 response.raise_for_status()
                 data = response.json()
+                finish_reason = (data.get("choices") or [{}])[0].get("finish_reason", "")
+                if finish_reason == "length" and attempt < self.max_retries:
+                    # 输出被 max_tokens 截断：2.0x 扩容后重试（计入 max_retries，不额外增加）
+                    old_max = int(payload["max_tokens"])
+                    new_max = int(old_max * 2)
+                    payload["max_tokens"] = new_max
+                    wait = 0.5 * (2**attempt)
+                    logger.warning(
+                        "输出被max_tokens截断(%d)，扩容到%d后重试 (第%d/%d次)",
+                        old_max,
+                        new_max,
+                        attempt + 1,
+                        self.max_retries + 1,
+                    )
+                    time.sleep(wait)
+                    continue
                 content = data["choices"][0]["message"]["content"]
+                self.last_finish_reason = finish_reason
                 _elapsed = time.time() - _t0
                 logger.info(
                     "[doubao] chat_completion 完成 model=%s tokens_in=%d tokens_out=%d elapsed=%.1fs attempt=%d timeout=%d",
@@ -291,6 +343,7 @@ class DoubaoClient:
                     data.get("usage", {}).get("completion_tokens", 0),
                     _elapsed,
                     attempt + 1,
+                    _req_timeout,
                 )
                 return content.strip()
             except Exception as e:
@@ -314,20 +367,21 @@ class DoubaoClient:
         self,
         messages: list[dict],
         images: list[str] | None = None,
-        max_tokens: int = 2048,
+        max_tokens: int | None = None,
         temperature: float = 0.3,
         timeout: int | None = None,
         model: str | None = None,
+        **kwargs,
     ) -> Optional[str]:
         """调用豆包视觉理解 API（OpenAI 兼容多模态格式）.
 
         将 images 附加到最后一条 user message 的 content 中，
-        使用 vision_model（默认 doubao-1-5-vision-pro-250328）。
+        使用构造函数传入的 self.model（DB capability 绑定的视觉模型，默认 qwen-vl-plus）。
 
         Args:
             messages: 对话消息列表。最后一条 user message 会被注入图片内容。
             images: 图片列表，支持 base64 data URI 或 HTTP(S) URL。
-            max_tokens: 最大生成 token 数，默认 2048。
+            max_tokens: 最大生成 token 数，默认 None（使用实例 self.max_tokens DB 配置，兜底 2048）。
             temperature: 采样温度，默认 0.3（视觉任务偏低更稳定）。
             timeout: 单次请求超时秒数，不传则使用默认 self.timeout。
 
@@ -367,14 +421,19 @@ class DoubaoClient:
             "Authorization": f"Bearer {self.api_key}",
             "Content-Type": "application/json",
         }
+        effective_max_tokens = max_tokens if max_tokens is not None else (self.max_tokens or 2048)
         payload: dict[str, Any] = {
-            "model": model or self.vision_model,
+            "model": model or self.model,
             "messages": vision_messages,
             "temperature": temperature,
-            "max_tokens": max_tokens,
+            "max_tokens": effective_max_tokens,
         }
+        if self.extra_params:
+            payload.update(self.extra_params)
+        if kwargs:
+            payload.update(kwargs)
 
-        req_timeout = timeout or self.timeout
+        req_timeout = self._resolve_timeout(timeout or self.timeout)
         last_error: Optional[Exception] = None
         _t0 = time.time()
         for attempt in range(self.max_retries + 1):
@@ -387,7 +446,24 @@ class DoubaoClient:
                 )
                 response.raise_for_status()
                 data = response.json()
+                finish_reason = (data.get("choices") or [{}])[0].get("finish_reason", "")
+                if finish_reason == "length" and attempt < self.max_retries:
+                    # 视觉输出被 max_tokens 截断：2.0x 扩容后重试（计入 max_retries）
+                    old_max = int(payload["max_tokens"])
+                    new_max = int(old_max * 2)
+                    payload["max_tokens"] = new_max
+                    wait = 0.5 * (2**attempt)
+                    logger.warning(
+                        "视觉输出被max_tokens截断(%d)，扩容到%d后重试 (第%d/%d次)",
+                        old_max,
+                        new_max,
+                        attempt + 1,
+                        self.max_retries + 1,
+                    )
+                    time.sleep(wait)
+                    continue
                 content = data["choices"][0]["message"]["content"]
+                self.last_finish_reason = finish_reason
                 _elapsed = time.time() - _t0
                 logger.info(
                     "[doubao] vision_completion 完成 model=%s tokens_in=%d tokens_out=%d elapsed=%.1fs attempt=%d",
@@ -589,28 +665,32 @@ class DoubaoClient:
         # 信任链只作用于 doubao provider；DashScope(Wan) 保持原行为。
         trust_chain_applied = False
         if provider == "doubao" and getattr(self, "trust_chain_enabled", True) and pre_trusted_images:
+            # #2220: 稀疏列表模式——pre_trusted_images 与 raw_portrait_urls 等长，
+            # None 位保留原图，非 None 位用 AI 人像替换。
             raw_portrait_urls: list[str] = []
             if image_url:
                 raw_portrait_urls.append(image_url)
             for u in ref_imgs:
                 if u not in raw_portrait_urls:
                     raw_portrait_urls.append(u)
-            trusted_urls: list[str] = []
-            if len(pre_trusted_images) >= 1:
-                trusted_urls = list(pre_trusted_images)
+            _n_trusted = sum(1 for _x in pre_trusted_images if _x)
+            if _n_trusted >= 1 and len(pre_trusted_images) >= len(raw_portrait_urls):
+                merged: list[str] = []
+                for _i, _orig in enumerate(raw_portrait_urls):
+                    _ai = pre_trusted_images[_i] if _i < len(pre_trusted_images) else None
+                    merged.append(str(_ai) if _ai else _orig)
                 trust_chain_applied = True
                 logger.info(
-                    "[trust-chain] 使用预热t2i结果 %d 张，替换原参考图走 reference_image 模式（原n=%d）",
-                    len(trusted_urls),
+                    "[trust-chain] 稀疏替换 %d/%d 张为AI人像（场景/商品图保留原图），走reference_image模式",
+                    _n_trusted,
                     len(raw_portrait_urls),
                 )
-            if trust_chain_applied and trusted_urls:
-                # 替换：原 image_url 用第一张 AI 图，ref_imgs 用剩余
-                if image_url and trusted_urls:
-                    image_url = trusted_urls[0]
-                    ref_imgs = trusted_urls[1:] if len(trusted_urls) > 1 else []
+                # 替换：image_url 用第一张（可能是AI或原图），ref_imgs 用剩余
+                if image_url and merged:
+                    image_url = merged[0]
+                    ref_imgs = merged[1:] if len(merged) > 1 else []
                 else:
-                    ref_imgs = trusted_urls
+                    ref_imgs = merged
         # ─────────────────────────────────────────────────────────────────
 
         # 判断任务模式：

packages/shared/ai_config_version.py

@@ -0,0 +1,62 @@
+"""AI 配置版本号管理 — Redis 通知机制.
+
+admin 后台修改 ai_models / ai_capability_configs 后调用 bump_version()，
+SaaS 端 AIRouter 每次取配置前比对版本号，变了才重新查 DB。
+
+Redis key: xiaoxia:ai_config:version = 时间戳字符串
+"""
+
+from __future__ import annotations
+
+import logging
+import time
+from typing import Optional
+
+logger = logging.getLogger(__name__)
+
+_REDIS_KEY = "xiaoxia:ai_config:version"
+
+
+def _get_redis_client():
+    """获取 Redis 客户端（复用 Celery broker 连接）."""
+    try:
+        import redis as _redis
+
+        from packages.shared.config import get_shared_settings
+
+        settings = get_shared_settings()
+        redis_url = getattr(settings, "redis_url", None) or getattr(
+            settings, "celery_broker_url", "redis://localhost:6379/0"
+        )
+        return _redis.Redis.from_url(redis_url, decode_responses=True, socket_timeout=2)
+    except Exception as e:
+        logger.warning("AI config version: Redis 客户端初始化失败: %s", e)
+        return None
+
+
+def bump_version() -> str:
+    """写入新版本号（当前时间戳），返回版本号字符串。失败返回空串。"""
+    r = _get_redis_client()
+    if r is None:
+        logger.warning("AI config bump_version: Redis 不可用，跳过版本号更新")
+        return ""
+    try:
+        ver = str(int(time.time() * 1000))
+        r.set(_REDIS_KEY, ver)
+        logger.info("AI config version bumped to %s", ver)
+        return ver
+    except Exception as e:
+        logger.warning("AI config bump_version 失败: %s", e)
+        return ""
+
+
+def get_version() -> Optional[str]:
+    """读取当前版本号。Redis 不可用或异常返回 None。"""
+    r = _get_redis_client()
+    if r is None:
+        return None
+    try:
+        return r.get(_REDIS_KEY)
+    except Exception as e:
+        logger.warning("AI config get_version 失败: %s", e)
+        return None

packages/shared/ai_router.py

@@ -0,0 +1,516 @@
+"""AI 模型路由层 — 统一模型配置读取与客户端构建.
+
+业务代码通过 AIRouter 获取客户端，不再硬编码 model/api_key/base_url。
+配置来源：DB ai_capability_configs JOIN ai_models → Redis 版本号缓存 → SharedSettings fallback。
+
+使用方式：
+    from packages.shared.ai_router import ai_router
+
+    client = ai_router.get_llm_client("intent_parsing")
+    result = client.chat_completion(messages=[...])
+"""
+
+from __future__ import annotations
+
+import logging
+import threading
+from dataclasses import dataclass
+
+from packages.shared.config import get_shared_settings
+
+logger = logging.getLogger(__name__)
+
+
+# ── 配置数据类 ──────────────────────────────────────────────────────────────
+
+
+@dataclass(frozen=True)
+class ModelConfig:
+    """单个 AI 模型配置（来自 ai_models 表）"""
+
+    id: str
+    name: str
+    provider: str
+    model_key: str
+    api_key: str
+    api_base: str
+    api_version: str | None
+    status: str
+
+
+@dataclass(frozen=True)
+class CapabilityConfig:
+    """业务能力配置（来自 ai_capability_configs JOIN ai_models）"""
+
+    capability_key: str
+    capability_name: str
+    primary_model: ModelConfig | None
+    lite_model: ModelConfig | None
+    fallback_model: ModelConfig | None
+    timeout_seconds: int
+    max_retries: int
+    max_tokens: int | None
+    temperature: float | None
+    concurrency: int
+    extra_params: dict
+    is_enabled: bool
+
+
+# ── 简单包装类（TTS / ImageGen / VideoGen）──────────────────────────────────
+
+
+class TTSClient:
+    """TTS 客户端（简单配置持有者，实际调用由 CosyVoiceService 完成）"""
+
+    def __init__(
+        self,
+        provider: str,
+        api_key: str,
+        base_url: str,
+        model: str,
+        timeout: int = 60,
+        extra_params: dict | None = None,
+    ):
+        self.provider = provider
+        self.api_key = api_key
+        self.base_url = base_url
+        self.model = model
+        self.timeout = timeout
+        self.extra_params = extra_params or {}
+
+    @property
+    def is_available(self) -> bool:
+        return bool(self.api_key and self.base_url and self.model)
+
+
+class ImageGenClient:
+    """图片生成客户端（简单配置持有者）"""
+
+    def __init__(
+        self,
+        provider: str,
+        api_key: str,
+        base_url: str,
+        model: str,
+        timeout: int = 60,
+        extra_params: dict | None = None,
+    ):
+        self.provider = provider
+        self.api_key = api_key
+        self.base_url = base_url
+        self.model = model
+        self.timeout = timeout
+        self.extra_params = extra_params or {}
+
+    @property
+    def is_available(self) -> bool:
+        return bool(self.api_key and self.base_url and self.model)
+
+
+class VideoGenClient:
+    """视频生成客户端（简单配置持有者）"""
+
+    def __init__(
+        self,
+        provider: str,
+        api_key: str,
+        base_url: str,
+        model: str,
+        timeout: int = 600,
+        extra_params: dict | None = None,
+    ):
+        self.provider = provider
+        self.api_key = api_key
+        self.base_url = base_url
+        self.model = model
+        self.timeout = timeout
+        self.extra_params = extra_params or {}
+
+    @property
+    def is_available(self) -> bool:
+        return bool(self.api_key and self.base_url and self.model)
+
+
+# ── DB Session 获取 ─────────────────────────────────────────────────────────
+
+
+def _get_session():
+    """获取 DB session，兼容 api / worker / 独立脚本场景"""
+    from packages.adapters.sqlalchemy_impl.session import SessionLocal
+
+    if SessionLocal is not None:
+        return SessionLocal()
+
+    try:
+        from worker_app.db import SessionLocal as WorkerSL
+
+        if WorkerSL is not None:
+            return WorkerSL()
+    except ImportError:
+        pass
+
+    try:
+        from app.db import SessionLocal as ApiSL
+
+        if ApiSL is not None:
+            return ApiSL()
+    except ImportError:
+        pass
+
+    return None
+
+
+# ── 核心路由类 ──────────────────────────────────────────────────────────────
+
+
+class AIRouter:
+    """AI 模型路由器 — 统一配置读取与客户端构建.
+
+    缓存策略：
+    1. 本地内存缓存 {capability_key: CapabilityConfig}
+    2. 每次读取前比对 Redis 版本号，变了则清缓存重新查 DB
+    3. DB 无配置 / Redis 不可用 → fallback 到 SharedSettings 环境变量
+    """
+
+    def __init__(self):
+        self._cache: dict[str, CapabilityConfig] = {}
+        self._local_ver: str | None = None
+        self._lock = threading.Lock()
+
+    def _check_version(self) -> bool:
+        """检查 Redis 版本号，变了返回 True（需要刷新缓存）"""
+        from packages.shared.ai_config_version import get_version
+
+        current_ver = get_version()
+        if current_ver is None:
+            return False
+        if self._local_ver != current_ver:
+            return True
+        return False
+
+    def _load_from_db(self, capability_key: str) -> CapabilityConfig | None:
+        """从 DB 加载配置（ai_capability_configs JOIN ai_models）"""
+        session = _get_session()
+        if session is None:
+            logger.warning("AI Router: 无法获取 DB session")
+            return None
+        try:
+            from sqlalchemy import text
+
+            sql = text("""
+                SELECT
+                    cc.capability_key, cc.capability_name, cc.timeout_seconds,
+                    cc.max_retries, cc.max_tokens, cc.temperature,
+                    cc.concurrency, cc.extra_params, cc.is_enabled,
+                    pm.id AS pm_id, pm.name AS pm_name, pm.provider AS pm_provider,
+                    pm.model_key AS pm_model_key, pm.api_key AS pm_api_key,
+                    pm.api_base AS pm_api_base, pm.api_version AS pm_api_version,
+                    pm.status AS pm_status,
+                    lm.id AS lm_id, lm.name AS lm_name, lm.provider AS lm_provider,
+                    lm.model_key AS lm_model_key, lm.api_key AS lm_api_key,
+                    lm.api_base AS lm_api_base, lm.api_version AS lm_api_version,
+                    lm.status AS lm_status,
+                    fm.id AS fm_id, fm.name AS fm_name, fm.provider AS fm_provider,
+                    fm.model_key AS fm_model_key, fm.api_key AS fm_api_key,
+                    fm.api_base AS fm_api_base, fm.api_version AS fm_api_version,
+                    fm.status AS fm_status
+                FROM ai_capability_configs cc
+                LEFT JOIN ai_models pm ON cc.primary_model_id = pm.id AND pm.deleted_at IS NULL
+                LEFT JOIN ai_models lm ON cc.lite_model_id = lm.id AND lm.deleted_at IS NULL
+                LEFT JOIN ai_models fm ON cc.fallback_model_id = fm.id AND fm.deleted_at IS NULL
+                WHERE cc.capability_key = :key AND cc.is_enabled = true
+            """)
+            row = session.execute(sql, {"key": capability_key}).first()
+            if not row:
+                return None
+
+            def _to_model(prefix: str) -> ModelConfig | None:
+                mid = getattr(row, f"{prefix}_id", None)
+                if not mid:
+                    return None
+                return ModelConfig(
+                    id=mid,
+                    name=getattr(row, f"{prefix}_name", "") or "",
+                    provider=getattr(row, f"{prefix}_provider", "") or "",
+                    model_key=getattr(row, f"{prefix}_model_key", "") or "",
+                    api_key=getattr(row, f"{prefix}_api_key", "") or "",
+                    api_base=getattr(row, f"{prefix}_api_base", "") or "",
+                    api_version=getattr(row, f"{prefix}_api_version", None),
+                    status=getattr(row, f"{prefix}_status", "active") or "active",
+                )
+
+            return CapabilityConfig(
+                capability_key=row.capability_key,
+                capability_name=row.capability_name,
+                primary_model=_to_model("pm"),
+                lite_model=_to_model("lm"),
+                fallback_model=_to_model("fm"),
+                timeout_seconds=row.timeout_seconds or 30,
+                max_retries=row.max_retries or 1,
+                max_tokens=row.max_tokens,
+                temperature=row.temperature,
+                concurrency=row.concurrency or 2,
+                extra_params=row.extra_params or {},
+                is_enabled=row.is_enabled,
+            )
+        except Exception as e:
+            logger.warning("AI Router: DB 查询失败 (key=%s): %s", capability_key, e)
+            return None
+        finally:
+            session.close()
+
+    def get_capability(self, key: str) -> CapabilityConfig | None:
+        """获取业务能力配置（带缓存）"""
+        with self._lock:
+            if self._check_version():
+                self._cache.clear()
+                from packages.shared.ai_config_version import get_version
+
+                self._local_ver = get_version()
+
+            if key in self._cache:
+                return self._cache[key]
+
+            config = self._load_from_db(key)
+            if config:
+                self._cache[key] = config
+            return config
+
+    def _get_model_or_fallback(self, cap: CapabilityConfig, variant: str = "primary") -> ModelConfig | None:
+        """按 variant 选择模型，不存在则降级。
+
+        - primary: primary → fallback
+        - lite: lite → primary
+        - fallback: fallback → primary（修复点：此前 fallback variant 被忽略，错误地使用了 primary 模型）
+        """
+        if variant == "fallback":
+            if cap.fallback_model:
+                return cap.fallback_model
+            if cap.primary_model:
+                return cap.primary_model
+        elif variant == "lite":
+            if cap.lite_model:
+                return cap.lite_model
+            if cap.primary_model:
+                return cap.primary_model
+        else:  # primary
+            if cap.primary_model:
+                return cap.primary_model
+            if cap.fallback_model:
+                return cap.fallback_model
+        return None
+
+    # ── 构建客户端 ─────────────────────────────────────────────────────────
+
+    def _build_llm_client(self, model: ModelConfig, cap: CapabilityConfig):
+        """构建 LLM 客户端 — 返回 DoubaoClient 实例"""
+        from packages.shared.ai_client import DoubaoClient
+
+        return DoubaoClient(
+            api_key=model.api_key,
+            base_url=model.api_base,
+            model=model.model_key,
+            timeout=cap.timeout_seconds,
+            max_retries=cap.max_retries,
+            max_tokens=cap.max_tokens,
+            temperature=cap.temperature,
+            extra_params=cap.extra_params,
+            provider=model.provider,
+        )
+
+    def _build_vision_client(self, model: ModelConfig, cap: CapabilityConfig):
+        """构建 VLM 客户端 — 返回 DoubaoClient 实例（DoubaoClient 已支持 vision_completion）"""
+        from packages.shared.ai_client import DoubaoClient
+
+        return DoubaoClient(
+            api_key=model.api_key,
+            base_url=model.api_base,
+            model=model.model_key,
+            timeout=cap.timeout_seconds,
+            max_retries=cap.max_retries,
+            max_tokens=cap.max_tokens,
+            temperature=cap.temperature,
+            extra_params=cap.extra_params,
+            provider=model.provider,
+        )
+
+    def _build_tts_client(self, model: ModelConfig, cap: CapabilityConfig) -> TTSClient:
+        return TTSClient(
+            provider=model.provider,
+            api_key=model.api_key,
+            base_url=model.api_base,
+            model=model.model_key,
+            timeout=cap.timeout_seconds,
+            extra_params=cap.extra_params,
+        )
+
+    def _build_image_gen_client(self, model: ModelConfig, cap: CapabilityConfig) -> ImageGenClient:
+        return ImageGenClient(
+            provider=model.provider,
+            api_key=model.api_key,
+            base_url=model.api_base,
+            model=model.model_key,
+            timeout=cap.timeout_seconds,
+            extra_params=cap.extra_params,
+        )
+
+    def _build_video_gen_client(self, model: ModelConfig, cap: CapabilityConfig) -> VideoGenClient:
+        return VideoGenClient(
+            provider=model.provider,
+            api_key=model.api_key,
+            base_url=model.api_base,
+            model=model.model_key,
+            timeout=cap.timeout_seconds,
+            extra_params=cap.extra_params,
+        )
+
+    # ── 公开接口 ────────────────────────────────────────────────────────────
+
+    def get_llm_client(self, key: str, variant: str = "primary"):
+        """获取 LLM 客户端（返回 DoubaoClient 实例）"""
+        cap = self.get_capability(key)
+        if cap and cap.is_enabled:
+            model = self._get_model_or_fallback(cap, variant)
+            if model and model.api_key:
+                return self._build_llm_client(model, cap)
+
+        return self._fallback_llm_client(key)
+
+    def get_vision_client(self, key: str, variant: str = "primary"):
+        """获取 VLM 客户端（返回 DoubaoClient 实例）"""
+        cap = self.get_capability(key)
+        if cap and cap.is_enabled:
+            model = self._get_model_or_fallback(cap, variant)
+            if model and model.api_key:
+                return self._build_vision_client(model, cap)
+
+        return self._fallback_vision_client(key)
+
+    def get_tts_client(self, key: str = "tts") -> TTSClient | None:
+        """获取 TTS 客户端"""
+        cap = self.get_capability(key)
+        if cap and cap.is_enabled and cap.primary_model and cap.primary_model.api_key:
+            return self._build_tts_client(cap.primary_model, cap)
+
+        return self._fallback_tts_client()
+
+    def get_image_gen_client(self, key: str = "image_generation") -> ImageGenClient | None:
+        """获取图片生成客户端"""
+        cap = self.get_capability(key)
+        if cap and cap.is_enabled and cap.primary_model and cap.primary_model.api_key:
+            return self._build_image_gen_client(cap.primary_model, cap)
+
+        return self._fallback_image_gen_client()
+
+    def get_video_gen_client(self, key: str = "video_generation") -> VideoGenClient | None:
+        """获取视频生成客户端"""
+        cap = self.get_capability(key)
+        if cap and cap.is_enabled and cap.primary_model and cap.primary_model.api_key:
+            return self._build_video_gen_client(cap.primary_model, cap)
+
+        return self._fallback_video_gen_client()
+
+    # ── Fallback 方法（读 SharedSettings 环境变量）──────────────────────────
+
+    def _fallback_llm_client(self, key: str):
+        """Fallback LLM 客户端 — 从 settings 读取配置，不硬编码"""
+        settings = get_shared_settings()
+        model_map = {
+            "intent_parsing": (settings.doubao_fast_model, settings.doubao_base_url, settings.doubao_api_key),
+            "copy_fusion": (settings.doubao_fast_model, settings.doubao_base_url, settings.doubao_api_key),
+            "storyboard": (settings.doubao_model, settings.doubao_base_url, settings.doubao_api_key),
+            "copy_review": (settings.doubao_model, settings.doubao_base_url, settings.doubao_api_key),
+            "asset_classify": (settings.doubao_model, settings.doubao_base_url, settings.doubao_api_key),
+        }
+        if key in model_map:
+            model_id, base_url, api_key = model_map[key]
+        else:
+            model_id = settings.doubao_model
+            base_url = settings.doubao_base_url
+            api_key = settings.doubao_api_key
+
+        if not api_key:
+            return None
+
+        from packages.shared.ai_client import DoubaoClient
+
+        return DoubaoClient(
+            provider="volcengine",
+            api_key=api_key,
+            base_url=base_url,
+            model=model_id,
+            timeout=settings.doubao_timeout,
+            max_retries=settings.doubao_max_retries,
+        )
+
+    def _fallback_vision_client(self, key: str):
+        """Fallback VLM 客户端 — 从 settings 读取 dashscope 配置，不硬编码"""
+        settings = get_shared_settings()
+        api_key = getattr(settings, "dashscope_api_key", "")
+        if not api_key:
+            return None
+        base_url = getattr(settings, "dashscope_base_url", "") or ""
+        model = getattr(settings, "dashscope_model", "") or getattr(settings, "doubao_vision_model", "")
+
+        from packages.shared.ai_client import DoubaoClient
+
+        return DoubaoClient(
+            provider="dashscope",
+            api_key=api_key,
+            base_url=base_url,
+            model=model,
+            timeout=15,
+        )
+
+    def _fallback_tts_client(self) -> TTSClient | None:
+        settings = get_shared_settings()
+        api_key = getattr(settings, "cosyvoice_api_key", "")
+        if not api_key:
+            return None
+        base_url = getattr(settings, "cosyvoice_base_url", "")
+        model = getattr(settings, "cosyvoice_model", "")
+
+        return TTSClient(provider="dashscope", api_key=api_key, base_url=base_url, model=model)
+
+    def _fallback_image_gen_client(self) -> ImageGenClient | None:
+        settings = get_shared_settings()
+        api_key = getattr(settings, "doubao_api_key", "")
+        if not api_key:
+            return None
+        base_url = getattr(settings, "doubao_base_url", "")
+        model = getattr(settings, "doubao_image_model", "")
+
+        return ImageGenClient(
+            provider="volcengine",
+            api_key=api_key,
+            base_url=base_url,
+            model=model,
+            timeout=getattr(settings, "doubao_image_timeout", 60),
+        )
+
+    def _fallback_video_gen_client(self) -> VideoGenClient | None:
+        settings = get_shared_settings()
+        api_key = getattr(settings, "doubao_api_key", "")
+        if not api_key:
+            return None
+        base_url = getattr(settings, "doubao_base_url", "")
+        model = getattr(settings, "doubao_video_model", "")
+
+        return VideoGenClient(
+            provider="volcengine",
+            api_key=api_key,
+            base_url=base_url,
+            model=model,
+            timeout=getattr(settings, "doubao_video_timeout", 600),
+        )
+
+    def invalidate(self):
+        """清空本地缓存"""
+        with self._lock:
+            self._cache.clear()
+            self._local_ver = None
+
+
+# ── 全局单例 ──────────────────────────────────────────────────────────────
+
+ai_router = AIRouter()

tests/unit/test_ai_router.py

@@ -0,0 +1,395 @@
+"""AI Router 单元测试 — 23 cases covering routing/cache/fallback/client construction."""
+
+from __future__ import annotations
+
+import sys
+import unittest
+from dataclasses import dataclass
+from typing import Optional
+from unittest.mock import MagicMock, patch
+
+# ── Pre-mock heavy import chain to avoid pulling in full app ──
+_mock_config = MagicMock()
+_mock_settings = MagicMock()
+_mock_settings.doubao_model = "doubao-seed-2-1-pro-260915"
+_mock_settings.doubao_fast_model = "doubao-seed-2-1-pro-260915"
+_mock_settings.doubao_base_url = "https://ark.cn-beijing.volces.com/api/v3"
+_mock_settings.doubao_api_key = "test-key"
+_mock_settings.doubao_timeout = 45
+_mock_settings.doubao_max_retries = 1
+_mock_settings.doubao_image_model = "doubao-seedream-5-0-flash-260915"
+_mock_settings.doubao_image_timeout = 60
+_mock_settings.doubao_video_model = "doubao-seedance-2-5-260628"
+_mock_settings.doubao_video_timeout = 600
+_mock_settings.dashscope_api_key = "ds-key"
+_mock_settings.cosyvoice_api_key = "cv-key"
+_mock_settings.cosyvoice_base_url = "https://dashscope.aliyuncs.com/api/v1"
+_mock_settings.cosyvoice_model = "cosyvoice-v3-flash"
+_mock_settings.redis_url = "redis://localhost:6379/0"
+_mock_settings.celery_broker_url = "redis://localhost:6379/0"
+_mock_config.get_shared_settings.return_value = _mock_settings
+
+# Prevent the full packages.shared from loading
+for mod_name in list(sys.modules.keys()):
+    if "packages.shared" in mod_name and "ai_router" not in mod_name and "ai_config_version" not in mod_name:
+        pass  # don't remove, just prevent new imports
+
+# Direct import of our modules (bypassing __init__.py)
+import importlib.util
+import os
+
+
+def _load_module_from_file(name, path):
+    spec = importlib.util.spec_from_file_location(name, path)
+    mod = importlib.util.module_from_spec(spec)
+    sys.modules[name] = mod
+    spec.loader.exec_module(mod)
+    return mod
+
+
+# Load ai_config_version
+_ai_config_version = _load_module_from_file(
+    "packages.shared.ai_config_version",
+    os.path.join(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))), "packages", "shared", "ai_config_version.py"),
+)
+# Patch get_shared_settings in the loaded module
+_ai_config_version.get_shared_settings = lambda: _mock_settings
+
+# Load ai_router - needs packages.shared.config to be available
+sys.modules["packages.shared.config"] = MagicMock()
+sys.modules["packages.shared.config"].get_shared_settings = lambda: _mock_settings
+
+# Mock packages.shared.ai_client to avoid triggering packages.shared.__init__ chain
+# (which fails on Python 3.10 due to datetime.UTC import in packages.domain)
+_mock_ai_client = MagicMock()
+
+class _FakeDoubaoClient:
+    """Fake DoubaoClient for testing - mimics the real interface."""
+    def __init__(self, api_key="", base_url="", model="", timeout=0, max_retries=0,
+                 max_tokens=None, temperature=None, extra_params=None, provider="volcengine"):
+        self.api_key = api_key
+        self.base_url = base_url
+        self.model = model
+        self.timeout = timeout
+        self.max_retries = max_retries
+        self.max_tokens = max_tokens
+        self.temperature = temperature
+        self.extra_params = extra_params or {}
+        self.provider = provider
+        self.vision_model = model
+
+    @property
+    def is_available(self):
+        return bool(self.api_key)
+
+    def chat_completion(self, messages, **kwargs):
+        return None
+
+    def vision_completion(self, messages, **kwargs):
+        return None
+
+_mock_ai_client.DoubaoClient = _FakeDoubaoClient
+sys.modules["packages.shared.ai_client"] = _mock_ai_client
+
+_ai_router = _load_module_from_file(
+    "packages.shared.ai_router",
+    os.path.join(os.path.dirname(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))), "packages", "shared", "ai_router.py"),
+)
+
+
+class TestAIConfigVersion(unittest.TestCase):
+    """Redis 版本号机制测试"""
+
+    @patch.object(_ai_config_version, "_get_redis_client")
+    def test_bump_version_success(self, mock_redis_fn):
+        mock_r = MagicMock()
+        mock_r.set.return_value = True
+        mock_redis_fn.return_value = mock_r
+        ver = _ai_config_version.bump_version()
+        self.assertTrue(ver)
+        self.assertTrue(ver.isdigit())
+        mock_r.set.assert_called_once()
+
+    @patch.object(_ai_config_version, "_get_redis_client")
+    def test_bump_version_redis_unavailable(self, mock_redis_fn):
+        mock_redis_fn.return_value = None
+        ver = _ai_config_version.bump_version()
+        self.assertEqual(ver, "")
+
+    @patch.object(_ai_config_version, "_get_redis_client")
+    def test_get_version_success(self, mock_redis_fn):
+        mock_r = MagicMock()
+        mock_r.get.return_value = "1234567890"
+        mock_redis_fn.return_value = mock_r
+        ver = _ai_config_version.get_version()
+        self.assertEqual(ver, "1234567890")
+
+    @patch.object(_ai_config_version, "_get_redis_client")
+    def test_get_version_redis_down(self, mock_redis_fn):
+        mock_redis_fn.return_value = None
+        ver = _ai_config_version.get_version()
+        self.assertIsNone(ver)
+
+    @patch.object(_ai_config_version, "_get_redis_client")
+    def test_get_version_exception(self, mock_redis_fn):
+        mock_r = MagicMock()
+        mock_r.get.side_effect = Exception("connection refused")
+        mock_redis_fn.return_value = mock_r
+        ver = _ai_config_version.get_version()
+        self.assertIsNone(ver)
+
+
+class TestAIRouter(unittest.TestCase):
+    """AIRouter 路由/缓存/fallback 测试"""
+
+    def setUp(self):
+        self.router = _ai_router.AIRouter()
+
+    @patch.object(_ai_config_version, "get_version", return_value=None)
+    def test_get_capability_db_unavailable(self, mock_ver):
+        with patch.object(_ai_router, "_get_session", return_value=None):
+            cap = self.router.get_capability("intent_parsing")
+            self.assertIsNone(cap)
+
+    @patch.object(_ai_config_version, "get_version", return_value=None)
+    def test_get_capability_from_db(self, mock_ver):
+        mock_session = MagicMock()
+        mock_row = MagicMock()
+        mock_row.capability_key = "intent_parsing"
+        mock_row.capability_name = "文案意图解析"
+        mock_row.timeout_seconds = 45
+        mock_row.max_retries = 1
+        mock_row.max_tokens = None
+        mock_row.temperature = None
+        mock_row.concurrency = 2
+        mock_row.extra_params = {}
+        mock_row.is_enabled = True
+        mock_row.pm_id = "model-1"
+        mock_row.pm_name = "豆包"
+        mock_row.pm_provider = "volcengine"
+        mock_row.pm_model_key = "doubao-seed-1-6-250615"
+        mock_row.pm_api_key = "test-key"
+        mock_row.pm_api_base = "https://ark.test.com"
+        mock_row.pm_api_version = None
+        mock_row.pm_status = "active"
+        mock_row.lm_id = None
+        mock_row.fm_id = None
+        mock_session.execute.return_value.first.return_value = mock_row
+
+        with patch.object(_ai_router, "_get_session", return_value=mock_session):
+            cap = self.router.get_capability("intent_parsing")
+            self.assertIsNotNone(cap)
+            self.assertEqual(cap.capability_key, "intent_parsing")
+            self.assertEqual(cap.primary_model.model_key, "doubao-seed-1-6-250615")
+
+    @patch.object(_ai_config_version, "get_version", side_effect=[None, "v2"])
+    def test_cache_invalidation_on_version_change(self, mock_ver):
+        with patch.object(self.router, "_load_from_db", return_value=None):
+            self.router.get_capability("test_key")
+        self.router._local_ver = "v1"
+        self.assertTrue(self.router._check_version())
+
+    @patch.object(_ai_config_version, "get_version", return_value="same_ver")
+    def test_cache_hit_same_version(self, mock_ver):
+        model = _ai_router.ModelConfig(
+            id="m1", name="test", provider="volcengine", model_key="test-model",
+            api_key="key", api_base="https://test.com", api_version=None, status="active",
+        )
+        cap = _ai_router.CapabilityConfig(
+            capability_key="test", capability_name="test", primary_model=model,
+            lite_model=None, fallback_model=None, timeout_seconds=30,
+            max_retries=1, max_tokens=None, temperature=None, concurrency=2,
+            extra_params={}, is_enabled=True,
+        )
+        self.router._cache["test"] = cap
+        self.router._local_ver = "same_ver"
+        result = self.router.get_capability("test")
+        self.assertEqual(result, cap)
+
+    def test_invalidate_clears_cache(self):
+        self.router._cache["x"] = MagicMock()
+        self.router._local_ver = "v1"
+        self.router.invalidate()
+        self.assertEqual(len(self.router._cache), 0)
+        self.assertIsNone(self.router._local_ver)
+
+    @patch.object(_ai_router, "_get_session", return_value=None)
+    @patch.object(_ai_config_version, "get_version", return_value=None)
+    def test_get_llm_client_fallback(self, mock_ver, mock_session):
+        _ai_router.get_shared_settings = lambda: _mock_settings
+        client = self.router.get_llm_client("intent_parsing")
+        self.assertIsNotNone(client)
+        self.assertEqual(client.model, "doubao-seed-2-1-pro-260915")
+        self.assertEqual(client.api_key, "test-key")
+
+    @patch.object(_ai_config_version, "get_version", return_value=None)
+    def test_get_llm_client_from_db(self, mock_ver):
+        model = _ai_router.ModelConfig(
+            id="m1", name="test", provider="dashscope", model_key="qwen3.8-flash",
+            api_key="db-key", api_base="https://dashscope.test.com", api_version=None, status="active",
+        )
+        cap = _ai_router.CapabilityConfig(
+            capability_key="image_analysis", capability_name="图片分析",
+            primary_model=model, lite_model=None, fallback_model=None,
+            timeout_seconds=15, max_retries=1, max_tokens=350, temperature=0.1,
+            concurrency=2, extra_params={}, is_enabled=True,
+        )
+        with patch.object(self.router, "get_capability", return_value=cap):
+            client = self.router.get_llm_client("image_analysis")
+            self.assertIsNotNone(client)
+            self.assertEqual(client.model, "qwen3.8-flash")
+            self.assertEqual(client.provider, "dashscope")
+
+    @patch.object(_ai_config_version, "get_version", return_value=None)
+    def test_get_vision_client(self, mock_ver):
+        model = _ai_router.ModelConfig(
+            id="m1", name="test", provider="dashscope", model_key="qwen3.8-flash",
+            api_key="key", api_base="https://dashscope.test.com", api_version=None, status="active",
+        )
+        cap = _ai_router.CapabilityConfig(
+            capability_key="image_analysis", capability_name="图片分析",
+            primary_model=model, lite_model=None, fallback_model=None,
+            timeout_seconds=15, max_retries=1, max_tokens=None, temperature=None,
+            concurrency=2, extra_params={}, is_enabled=True,
+        )
+        with patch.object(self.router, "get_capability", return_value=cap):
+            client = self.router.get_vision_client("image_analysis")
+            self.assertIsNotNone(client)
+            # #2220: vision client is now DoubaoClient with vision_completion
+            self.assertTrue(hasattr(client, "vision_completion"))
+
+    @patch.object(_ai_config_version, "get_version", return_value=None)
+    def test_get_tts_client(self, mock_ver):
+        model = _ai_router.ModelConfig(
+            id="m1", name="test", provider="dashscope", model_key="cosyvoice-v3-flash",
+            api_key="key", api_base="https://dashscope.test.com", api_version=None, status="active",
+        )
+        cap = _ai_router.CapabilityConfig(
+            capability_key="tts", capability_name="语音合成",
+            primary_model=model, lite_model=None, fallback_model=None,
+            timeout_seconds=60, max_retries=1, max_tokens=None, temperature=None,
+            concurrency=2, extra_params={}, is_enabled=True,
+        )
+        with patch.object(self.router, "get_capability", return_value=cap):
+            client = self.router.get_tts_client()
+            self.assertIsNotNone(client)
+            self.assertEqual(client.model, "cosyvoice-v3-flash")
+
+    @patch.object(_ai_config_version, "get_version", return_value=None)
+    def test_get_image_gen_client(self, mock_ver):
+        model = _ai_router.ModelConfig(
+            id="m1", name="test", provider="volcengine", model_key="seedream-5.0-flash",
+            api_key="key", api_base="https://ark.test.com", api_version=None, status="active",
+        )
+        cap = _ai_router.CapabilityConfig(
+            capability_key="image_generation", capability_name="图片生成",
+            primary_model=model, lite_model=None, fallback_model=None,
+            timeout_seconds=60, max_retries=1, max_tokens=None, temperature=None,
+            concurrency=2, extra_params={"size": "1K"}, is_enabled=True,
+        )
+        with patch.object(self.router, "get_capability", return_value=cap):
+            client = self.router.get_image_gen_client()
+            self.assertIsNotNone(client)
+            self.assertEqual(client.model, "seedream-5.0-flash")
+
+    @patch.object(_ai_config_version, "get_version", return_value=None)
+    def test_get_video_gen_client(self, mock_ver):
+        model = _ai_router.ModelConfig(
+            id="m1", name="test", provider="volcengine", model_key="seedance-2.5",
+            api_key="key", api_base="https://ark.test.com", api_version=None, status="active",
+        )
+        cap = _ai_router.CapabilityConfig(
+            capability_key="video_generation", capability_name="视频生成",
+            primary_model=model, lite_model=None, fallback_model=None,
+            timeout_seconds=600, max_retries=1, max_tokens=None, temperature=None,
+            concurrency=1, extra_params={}, is_enabled=True,
+        )
+        with patch.object(self.router, "get_capability", return_value=cap):
+            client = self.router.get_video_gen_client()
+            self.assertIsNotNone(client)
+            self.assertEqual(client.model, "seedance-2.5")
+
+    @patch.object(_ai_config_version, "get_version", return_value=None)
+    def test_lite_variant_preference(self, mock_ver):
+        primary = _ai_router.ModelConfig(id="p1", name="pro", provider="volcengine", model_key="pro-model", api_key="k", api_base="u", api_version=None, status="active")
+        lite = _ai_router.ModelConfig(id="l1", name="lite", provider="volcengine", model_key="lite-model", api_key="k", api_base="u", api_version=None, status="active")
+        cap = _ai_router.CapabilityConfig(
+            capability_key="image_analysis", capability_name="图片分析",
+            primary_model=primary, lite_model=lite, fallback_model=None,
+            timeout_seconds=15, max_retries=1, max_tokens=None, temperature=None,
+            concurrency=2, extra_params={}, is_enabled=True,
+        )
+        model = self.router._get_model_or_fallback(cap, "lite")
+        self.assertEqual(model.model_key, "lite-model")
+        model_primary = self.router._get_model_or_fallback(cap, "primary")
+        self.assertEqual(model_primary.model_key, "pro-model")
+
+    @patch.object(_ai_config_version, "get_version", return_value=None)
+    def test_disabled_capability_returns_fallback(self, mock_ver):
+        cap = _ai_router.CapabilityConfig(
+            capability_key="test", capability_name="test",
+            primary_model=None, lite_model=None, fallback_model=None,
+            timeout_seconds=30, max_retries=1, max_tokens=None, temperature=None,
+            concurrency=2, extra_params={}, is_enabled=False,
+        )
+        _ai_router.get_shared_settings = lambda: _mock_settings
+        with patch.object(self.router, "get_capability", return_value=cap):
+            client = self.router.get_llm_client("test")
+            self.assertIsNotNone(client)
+            self.assertEqual(client.model, "doubao-seed-2-1-pro-260915")
+
+    @patch.object(_ai_config_version, "get_version", return_value=None)
+    def test_fallback_chain_primary_none(self, mock_ver):
+        """primary_model 为 None 时 fallback 到 fallback_model"""
+        fb = _ai_router.ModelConfig(id="f1", name="fb", provider="volcengine", model_key="fb-model", api_key="k", api_base="u", api_version=None, status="active")
+        cap = _ai_router.CapabilityConfig(
+            capability_key="test", capability_name="test",
+            primary_model=None, lite_model=None, fallback_model=fb,
+            timeout_seconds=30, max_retries=1, max_tokens=None, temperature=None,
+            concurrency=2, extra_params={}, is_enabled=True,
+        )
+        model = self.router._get_model_or_fallback(cap, "primary")
+        self.assertEqual(model.model_key, "fb-model")
+
+
+class TestModelConfig(unittest.TestCase):
+    """数据类测试"""
+
+    def test_model_config_frozen(self):
+        m = _ai_router.ModelConfig(id="1", name="t", provider="p", model_key="k", api_key="a", api_base="b", api_version=None, status="active")
+        with self.assertRaises(AttributeError):
+            m.model_key = "new"
+
+    def test_capability_config_frozen(self):
+        c = _ai_router.CapabilityConfig(
+            capability_key="k", capability_name="n", primary_model=None,
+            lite_model=None, fallback_model=None, timeout_seconds=30,
+            max_retries=1, max_tokens=None, temperature=None, concurrency=2,
+            extra_params={}, is_enabled=True,
+        )
+        with self.assertRaises(AttributeError):
+            c.is_enabled = False
+
+
+class TestClientAvailability(unittest.TestCase):
+    """客户端可用性测试"""
+
+    def test_tts_client_available(self):
+        c = _ai_router.TTSClient(provider="p", api_key="k", base_url="u", model="m")
+        self.assertTrue(c.is_available)
+
+    def test_tts_client_unavailable_no_model(self):
+        c = _ai_router.TTSClient(provider="p", api_key="k", base_url="u", model="")
+        self.assertFalse(c.is_available)
+
+    def test_image_gen_client_unavailable_no_url(self):
+        c = _ai_router.ImageGenClient(provider="p", api_key="k", base_url="", model="m")
+        self.assertFalse(c.is_available)
+
+    def test_video_gen_client_available(self):
+        c = _ai_router.VideoGenClient(provider="p", api_key="k", base_url="u", model="m")
+        self.assertTrue(c.is_available)
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/unit/test_api_settings.py

@@ -73,15 +73,15 @@ class TestSharedSettingsDefaults:
 
     def test_default_cosyvoice_settings(self):
         s = SharedSettings()
-        assert s.cosyvoice_model == "cosyvoice-v3-flash"
+        assert s.cosyvoice_model == ""  # 零硬编码：默认值已清空
         assert s.cosyvoice_format == "mp3"
         assert s.cosyvoice_sample_rate == 22050
 
     def test_default_doubao_settings(self):
         s = SharedSettings()
-        assert "doubao" in s.doubao_model
+        assert s.doubao_model == ""  # 零硬编码：默认值已清空
         assert s.doubao_timeout == 45  # #2180 默认提到45s
-        assert s.doubao_max_retries == 1
+        assert s.doubao_max_retries == 3
 
 
 class TestAPISettingsDefaults:
@@ -321,7 +321,7 @@ class TestWorkerSettingsDefaults:
         assert s.database_url  # 继承自SharedSettings
         assert s.redis_url
         assert s.oss_endpoint
-        assert s.cosyvoice_model == "cosyvoice-v3-flash"
+        assert s.cosyvoice_model == ""  # 零硬编码：默认值已清空
 
 
 class TestGetWorkerSettings:

tests/unit/test_config_base.py

@@ -102,17 +102,17 @@ class TestSharedSettingsDefaults:
     def test_default_cosyvoice_config(self):
         """CosyVoice 默认配置"""
         s = self._make_settings()
-        assert s.cosyvoice_model == "cosyvoice-v3-flash"
+        assert s.cosyvoice_model == ""  # 零硬编码：默认值已清空
         assert s.cosyvoice_sample_rate == 22050
         assert s.cosyvoice_format == "mp3"
-        assert s.cosyvoice_clone_model == "voice-enrollment"
+        assert s.cosyvoice_clone_model == ""  # 零硬编码：默认值已清空
 
     def test_default_doubao_config(self):
         """豆包默认配置"""
         s = self._make_settings()
         assert s.doubao_timeout == 45  # #2180 默认提到45s
-        assert s.doubao_max_retries == 1
-        assert "volces.com" in s.doubao_base_url
+        assert s.doubao_max_retries == 3
+        assert s.doubao_base_url == ""  # 零硬编码：默认值已清空
 
     def test_default_empty_api_keys(self):
         """API Key 默认空字符串"""

tests/unit/test_cosyvoice_service.py

@@ -27,7 +27,10 @@ def mock_client() -> MagicMock:
 @pytest.fixture
 def service(mock_client: MagicMock) -> CosyVoiceService:
     """Create CosyVoiceService with mocked HTTP client and config."""
-    with patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings:
+    with (
+        patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings,
+        patch("packages.shared.ai_router.ai_router") as mock_router,
+    ):
         settings = MagicMock()
         settings.cosyvoice_api_key = "sk-test-12345678"
         settings.cosyvoice_base_url = "https://dashscope.aliyuncs.com/api/v1"
@@ -37,6 +40,7 @@ def service(mock_client: MagicMock) -> CosyVoiceService:
         settings.cosyvoice_voice = "longxiaochun_v3"
         settings.cosyvoice_clone_model = "voice-enrollment"
         mock_settings.return_value = settings
+        mock_router.get_tts_client.return_value = None  # ai_router returns None in tests
         svc = CosyVoiceService(http_client=mock_client)
         svc.CLONE_POLL_INTERVAL = 0.001  # 加速测试
         svc.RETRY_BACKOFF = 0.001
@@ -48,7 +52,10 @@ class TestInitConfig:
 
     def test_base_url_with_old_text2audio_path_gets_normalized(self, mock_client: MagicMock) -> None:
         """旧版 base_url 带 text2audio 路径应自动修正."""
-        with patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings:
+        with (
+            patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings,
+            patch("packages.shared.ai_router.ai_router") as mock_router,
+        ):
             settings = MagicMock()
             settings.cosyvoice_api_key = "sk-test"
             settings.cosyvoice_base_url = "https://dashscope.aliyuncs.com/api/v1/services/aigc/text2audio"
@@ -58,12 +65,16 @@ class TestInitConfig:
             settings.cosyvoice_voice = "test"
             settings.cosyvoice_clone_model = "voice-enrollment"
             mock_settings.return_value = settings
+            mock_router.get_tts_client.return_value = None
             svc = CosyVoiceService(http_client=mock_client)
             assert svc._base_url == "https://dashscope.aliyuncs.com/api/v1"
 
     def test_custom_params_override_config(self, mock_client: MagicMock) -> None:
         """显式传入参数覆盖配置."""
-        with patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings:
+        with (
+            patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings,
+            patch("packages.shared.ai_router.ai_router") as mock_router,
+        ):
             settings = MagicMock()
             settings.cosyvoice_api_key = "sk-config"
             settings.cosyvoice_base_url = "https://config.example.com"
@@ -73,6 +84,7 @@ class TestInitConfig:
             settings.cosyvoice_voice = "test"
             settings.cosyvoice_clone_model = "voice-enrollment"
             mock_settings.return_value = settings
+            mock_router.get_tts_client.return_value = None
             svc = CosyVoiceService(
                 api_key="sk-custom",
                 base_url="https://custom.example.com/api/v1",
@@ -87,7 +99,10 @@ class TestInitConfig:
 
     def test_context_manager(self, mock_client: MagicMock) -> None:
         """上下文管理器正常工作."""
-        with patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings:
+        with (
+            patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings,
+            patch("packages.shared.ai_router.ai_router") as mock_router,
+        ):
             settings = MagicMock()
 
             settings.cosyvoice_api_key = "sk-test"
@@ -105,6 +120,7 @@ class TestInitConfig:
             settings.cosyvoice_clone_model = "voice-enrollment"
 
             mock_settings.return_value = settings
+            mock_router.get_tts_client.return_value = None
             svc = CosyVoiceService(http_client=mock_client)
             with svc as s:
                 assert s is svc
@@ -113,7 +129,10 @@ class TestInitConfig:
 
     def test_owns_client_gets_closed(self) -> None:
         """自有client在close时被关闭."""
-        with patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings:
+        with (
+            patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings,
+            patch("packages.shared.ai_router.ai_router") as mock_router,
+        ):
             settings = MagicMock()
 
             settings.cosyvoice_api_key = "sk-test"
@@ -131,6 +150,7 @@ class TestInitConfig:
             settings.cosyvoice_clone_model = "voice-enrollment"
 
             mock_settings.return_value = settings
+            mock_router.get_tts_client.return_value = None
             with patch("packages.application.cosyvoice_service.httpx.Client") as mock_cls:
                 mock_instance = MagicMock()
                 mock_cls.return_value = mock_instance
@@ -173,7 +193,10 @@ class TestSubmitCloneTask:
 
     def test_no_api_key_raises_auth_error(self, mock_client: MagicMock) -> None:
         """无API Key抛认证错误."""
-        with patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings:
+        with (
+            patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings,
+            patch("packages.shared.ai_router.ai_router") as mock_router,
+        ):
             settings = MagicMock()
 
             settings.cosyvoice_api_key = ""
@@ -191,6 +214,7 @@ class TestSubmitCloneTask:
             settings.cosyvoice_clone_model = "voice-enrollment"
 
             mock_settings.return_value = settings
+            mock_router.get_tts_client.return_value = None
             svc = CosyVoiceService(http_client=mock_client)
             with pytest.raises(CosyVoiceAuthError, match="API Key 未配置"):
                 svc.submit_clone_task(audio_url="https://example.com/audio.mp3")
@@ -258,7 +282,10 @@ class TestSubmitCloneTask:
 
     def test_audio_url_signer_is_called(self, mock_client: MagicMock) -> None:
         """配置了audio_url_signer时会被调用预签名."""
-        with patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings:
+        with (
+            patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings,
+            patch("packages.shared.ai_router.ai_router") as mock_router,
+        ):
             settings = MagicMock()
 
             settings.cosyvoice_api_key = "sk-test"
@@ -276,6 +303,7 @@ class TestSubmitCloneTask:
             settings.cosyvoice_clone_model = "voice-enrollment"
 
             mock_settings.return_value = settings
+            mock_router.get_tts_client.return_value = None
             signer = MagicMock(return_value="https://signed.example.com/audio.mp3?token=xxx")
             svc = CosyVoiceService(http_client=mock_client, audio_url_signer=signer)
 
@@ -297,7 +325,10 @@ class TestSubmitCloneTask:
 
     def test_signer_failure_falls_back_to_original_url(self, mock_client: MagicMock) -> None:
         """预签名失败时回退到原始URL."""
-        with patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings:
+        with (
+            patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings,
+            patch("packages.shared.ai_router.ai_router") as mock_router,
+        ):
             settings = MagicMock()
 
             settings.cosyvoice_api_key = "sk-test"
@@ -315,6 +346,7 @@ class TestSubmitCloneTask:
             settings.cosyvoice_clone_model = "voice-enrollment"
 
             mock_settings.return_value = settings
+            mock_router.get_tts_client.return_value = None
             signer = MagicMock(side_effect=RuntimeError("sign failed"))
             svc = CosyVoiceService(http_client=mock_client, audio_url_signer=signer)
 
@@ -375,7 +407,10 @@ class TestQueryVoiceStatus:
 
     def test_no_api_key_raises(self, mock_client: MagicMock) -> None:
         """无API Key抛认证错误."""
-        with patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings:
+        with (
+            patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings,
+            patch("packages.shared.ai_router.ai_router") as mock_router,
+        ):
             settings = MagicMock()
 
             settings.cosyvoice_api_key = ""
@@ -393,6 +428,7 @@ class TestQueryVoiceStatus:
             settings.cosyvoice_clone_model = "voice-enrollment"
 
             mock_settings.return_value = settings
+            mock_router.get_tts_client.return_value = None
             svc = CosyVoiceService(http_client=mock_client)
             with pytest.raises(CosyVoiceAuthError):
                 svc.query_voice_status("v1")
@@ -589,7 +625,10 @@ class TestSubmitSynthesizeTask:
 
     def test_no_api_key_raises(self, mock_client: MagicMock) -> None:
         """无API Key抛认证错误."""
-        with patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings:
+        with (
+            patch("packages.application.cosyvoice_service.get_shared_settings") as mock_settings,
+            patch("packages.shared.ai_router.ai_router") as mock_router,
+        ):
             settings = MagicMock()
 
             settings.cosyvoice_api_key = ""
@@ -607,6 +646,7 @@ class TestSubmitSynthesizeTask:
             settings.cosyvoice_clone_model = "voice-enrollment"
 
             mock_settings.return_value = settings
+            mock_router.get_tts_client.return_value = None
             svc = CosyVoiceService(http_client=mock_client)
             with pytest.raises(CosyVoiceAuthError):
                 svc.submit_synthesize_task(text="你好", voice_id="v1")

tests/unit/test_domain_entities_extended.py

@@ -521,3 +521,81 @@ class TestIngestJob:
             storage_key="k",
         )
         assert job.error_message == ""
+
+
+class TestViralVideoResumeForRegenerate:
+    """#2222: resume_from_image_analyzed 应支持 COPY_GENERATED/COMPLETED/FAILED 重新生成文案。"""
+
+    def test_regen_from_copy_generated_clears_old_copy(self):
+        from datetime import datetime, timezone
+
+        from packages.domain.viral_video import ViralVideoJob, ViralVideoStatus
+
+        job = ViralVideoJob(user_id="u1", images=["img1"])
+        # 模拟已经生成过文案和视频
+        job.status = ViralVideoStatus.COPY_GENERATED
+        job.copy_result = {"shots": [{"x": 1}], "voiceover_script": "旧文案"}
+        job.intent_result = {"intent": "旧意图"}
+        job.storyboard = [{"x": 1}]
+        job.generated_copy_text = "旧文案"
+        job.result_video_url = "http://old.mp4"
+        job.completed_at = datetime(2026, 10, 6, tzinfo=timezone.utc)
+        job.error_msg = ""
+        job.current_stage = "tts_generation"
+        job.phase_message = "TTS完成"
+
+        # 重新生成
+        job.resume_from_image_analyzed()
+
+        assert job.status == ViralVideoStatus.RUNNING
+        assert job.copy_result is None
+        assert job.intent_result is None
+        assert job.storyboard is None
+        assert job.generated_copy_text == ""
+        assert job.result_video_url == ""
+        assert job.completed_at is None
+        assert job.error_msg == ""
+        assert job.current_stage == ""
+        assert job.phase_message == ""
+
+    def test_regen_from_completed_clears_old_copy(self):
+        from packages.domain.viral_video import ViralVideoJob, ViralVideoStatus
+
+        job = ViralVideoJob(user_id="u1", images=["img1"])
+        job.status = ViralVideoStatus.COMPLETED
+        job.copy_result = {"shots": [], "voiceover_script": "xx"}
+        job.intent_result = {"intent": "x"}
+        job.result_video_url = "http://v.mp4"
+
+        job.resume_from_image_analyzed()
+
+        assert job.status == ViralVideoStatus.RUNNING
+        assert job.copy_result is None
+        assert job.intent_result is None
+        assert job.result_video_url == ""
+
+    def test_first_call_from_image_analyzed_keeps_fields(self):
+        """首次进入（IMAGE_ANALYZED）不应清空任何已有的字段。"""
+        from packages.domain.viral_video import ViralVideoJob, ViralVideoStatus
+
+        job = ViralVideoJob(user_id="u1", images=["img1"])
+        job.status = ViralVideoStatus.IMAGE_ANALYZED
+        job.image_analysis = {"products": []}
+        job.industry = "美妆"
+
+        job.resume_from_image_analyzed()
+
+        assert job.status == ViralVideoStatus.RUNNING
+        assert job.image_analysis == {"products": []}
+        assert job.industry == "美妆"
+
+    def test_wait_user_confirm_rejected(self):
+        """wait_user_confirm 中间状态应被拒绝（前端正在编辑/确认文案）。"""
+        import pytest
+
+        from packages.domain.viral_video import ViralVideoJob, ViralVideoStatus
+
+        job = ViralVideoJob(user_id="u1", images=["img1"])
+        job.status = ViralVideoStatus.WAIT_USER_CONFIRM
+        with pytest.raises(ValueError, match="Cannot resume"):
+            job.resume_from_image_analyzed()

tests/unit/test_tts_service_factory.py

@@ -3,7 +3,7 @@
 from __future__ import annotations
 
 import os
-from unittest.mock import patch
+from unittest.mock import MagicMock, patch
 
 import pytest
 
@@ -108,8 +108,14 @@ class TestGetTtsService:
 
     def test_empty_env_falls_back_to_auto_detect(self):
         """环境变量为空时自动检测."""
-        with patch.dict(os.environ, {"TTS_PROVIDER": ""}):
+        with (
+            patch.dict(os.environ, {"TTS_PROVIDER": ""}),
+            patch("packages.shared.config.get_shared_settings") as mock_settings,
+        ):
             # 没有 cosyvoice_api_key 时应该用 mock
+            settings = MagicMock()
+            settings.cosyvoice_api_key = ""
+            mock_settings.return_value = settings
             service = get_tts_service(None)
             assert service.provider_name == "mock"
 

tests/unit/test_viral_video_routes.py

@@ -431,7 +431,7 @@ class TestGenerateCopy:
         assert resp.id == "job-gc"
 
     def test_generate_copy_rejects_wrong_status(self):
-        """任务在 copy_generated/completed 时不能再 generate-copy（状态保护）。"""
+        """wait_user_confirm 等中间状态不允许调用 generate-copy（状态保护）。"""
         import pytest
         from app.api.routes import viral_video as vv_mod
         from app.schemas.viral_video import GenerateCopyRequest
@@ -441,7 +441,8 @@ class TestGenerateCopy:
 
         user = _auth_user("u1")
         session = MagicMock()
-        job = _make_job(job_id="job-gc2", user_id="u1", status=ViralVideoStatus.COPY_GENERATED)
+        # wait_user_confirm 属于前端在编辑/确认文案的中间状态，应拒绝重新触发生成
+        job = _make_job(job_id="job-gc2", user_id="u1", status=ViralVideoStatus.WAIT_USER_CONFIRM)
         repo = MagicMock()
         repo.get.return_value = job
 
@@ -450,6 +451,31 @@ class TestGenerateCopy:
                 vv_mod.generate_copy("job-gc2", GenerateCopyRequest(), authenticated_user=user, session=session)
             assert exc.value.status_code == 409
 
+    def test_generate_copy_allows_regenerate_from_copy_generated(self):
+        """#2222: COPY_GENERATED/COMPLETED 状态下点「重新生成文案」应放行入队，不返回 409。"""
+        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()
+        for regen_status in (ViralVideoStatus.COPY_GENERATED, ViralVideoStatus.COMPLETED):
+            job = _make_job(job_id=f"job-regen-{regen_status}", user_id="u1", status=regen_status)
+            repo = MagicMock()
+            repo.get.return_value = job
+            with (
+                patch.object(vv_mod, "_get_job_repo", return_value=repo),
+                patch.object(vv_mod.celery_app, "send_task") as mock_send,
+            ):
+                resp = vv_mod.generate_copy(f"job-regen-{regen_status}", GenerateCopyRequest(), authenticated_user=user, session=session)
+            mock_send.assert_called_once()
+            job.resume_from_image_analyzed.assert_called()
+            assert job.retry_count >= 1
+            assert resp.id == f"job-regen-{regen_status}"
+
     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

tests/unit/test_vision_v4_prompt.py

@@ -0,0 +1,291 @@
+# -*- coding: utf-8 -*-
+"""vision v4 prompt / assembler 单元测试：
+
+- assembler 正确识别 v4 嵌套 schema 与旧扁平 schema
+- v4 product/person/store/other 四类输出组装出下游必出字段
+- 旧扁平 schema 行为不变
+- _prompt._resolve：DB 有 active prompt 时原样使用（不追加硬编码 schema）；
+  DB 无记录时回落到硬编码 JSON schema
+"""
+
+from __future__ import annotations
+
+import types
+from typing import Any
+
+import pytest
+from worker_app.tasks.vision import _prompt, assembler
+
+REQUIRED_KEYS = {
+    "name",
+    "brand",
+    "category",
+    "appearance",
+    "packaging",
+    "text_on_package",
+    "key_features",
+    "scene",
+    "mood",
+    "portrait_prompt",
+    "summary",
+    "_source",
+}
+
+
+# ---------- schema 识别 ----------
+
+
+def test_is_v4_schema_products_list() -> None:
+    assert assembler._is_v4_schema({"type": "product", "products": []})
+
+
+def test_is_v4_schema_type_only() -> None:
+    assert assembler._is_v4_schema({"type": "person"})
+
+
+def test_is_v4_schema_people_dict() -> None:
+    assert assembler._is_v4_schema({"people": {"has_person": True}})
+
+
+def test_is_not_v4_schema_flat() -> None:
+    assert not assembler._is_v4_schema({"has_person": True, "upper_wear": "T恤"})
+
+
+# ---------- v4 product ----------
+
+V4_PRODUCT: dict[str, Any] = {
+    "type": "product",
+    "scene": "白色背景产品图",
+    "mood": "清新专业",
+    "style": "商业产品摄影",
+    "colors": [{"hex": "#E60012", "name": "亮红色", "coverage": 0.6}],
+    "visible_text": [{"text": "OMO奥妙除菌除螨", "position": "瓶身正面"}],
+    "products": [
+        {
+            "product_name": "OMO奥妙除菌除螨洗衣液",
+            "brand": "OMO奥妙",
+            "category": "洗护",
+            "package_type": "瓶装",
+            "package_color": "亮红色瓶身",
+            "cap_type": "透明翻盖式按压瓶口",
+            "body_shape": "带侧面握持把手的竖款瓶身",
+            "label_design": "瓶身印十字盾牌图案",
+            "product_features": ["亮红色瓶装", "按压式瓶口", "十字盾牌标签"],
+            "key_selling_points": ["天然除菌除螨"],
+            "position": "main",
+        }
+    ],
+    "has_person": False,
+}
+
+
+def test_assemble_v4_product_fields() -> None:
+    r = assembler.assemble_result(0, V4_PRODUCT, ["OMO奥妙"])
+    assert REQUIRED_KEYS <= set(r.keys())
+    assert r["name"] == "OMO奥妙除菌除螨洗衣液"
+    assert r["brand"] == "OMO奥妙"
+    assert r["category"] == "洗护"
+    assert "瓶装" in r["packaging"]
+    assert isinstance(r["key_features"], list) and r["key_features"]
+    assert any("除菌" in str(t) for t in r["text_on_package"])
+    assert len(r["portrait_prompt"]) >= 10
+    assert r["_source"] == "v2_fast_json_v4"
+
+
+def test_assemble_v4_product_multi_selects_main() -> None:
+    fj = {
+        "type": "product",
+        "products": [
+            {"product_name": "次要商品", "brand": "B"},
+            {"product_name": "主商品", "brand": "A", "position": "main"},
+        ],
+    }
+    r = assembler.assemble_result(1, fj, [])
+    assert r["name"] == "主商品"
+
+
+# ---------- v4 person ----------
+
+V4_PERSON: dict[str, Any] = {
+    "type": "person",
+    "scene": "户外街拍",
+    "mood": "自信",
+    "style": "街拍",
+    "colors": [],
+    "visible_text": [],
+    "has_person": True,
+    "gender": "女",
+    "age_range": "青年",
+    "upper_wear": "白色V领短袖T恤",
+    "upper_color": "白色",
+    "lower_wear": "黑色高腰阔腿裤",
+    "lower_color": "黑色",
+    "dress_color": None,
+    "accessories": ["银色项链"],
+    "hairstyle": "黑色长直发",
+    "expression": "自信",
+    "pose": "侧身站立",
+    "outfit_style": "休闲日常",
+    "portrait_prompt": (
+        "一位年轻女性，身穿白色V领短袖T恤、黑色高腰阔腿裤，佩戴银色项链，"
+        "黑色长直发，神情自信，侧身站立，休闲日常风格，城市街拍场景"
+    ),
+    "products": [],
+}
+
+
+def test_assemble_v4_person() -> None:
+    r = assembler.assemble_result(0, V4_PERSON, [])
+    assert REQUIRED_KEYS <= set(r.keys())
+    assert r["category"] == "人物穿搭"
+    assert r["_source"] == "v2_fast_json_v5"
+    assert "T恤" in r["name"]
+    assert "年轻女性" in r["portrait_prompt"]
+    assert "项链" in r["portrait_prompt"]
+    assert isinstance(r["key_features"], list) and len(r["key_features"]) <= 8
+
+
+def test_assemble_v4_person_people_nested() -> None:
+    fj = {"type": "person", "people": {**V4_PERSON, "has_person": True}}
+    r = assembler.assemble_result(0, fj, [])
+    assert r["category"] == "人物穿搭"
+    assert "年轻女性" in r["portrait_prompt"]
+
+
+# ---------- v4 store ----------
+
+
+def test_assemble_v4_store() -> None:
+    fj = {
+        "type": "store",
+        "scene": "便利店内部",
+        "mood": "日常便民",
+        "style": "门店实拍",
+        "store_type": "社区便利店",
+        "store_layout": "纵深货架布局",
+        "brand_signage": "全家FamilyMart",
+        "visual_elements": ["红白主色调", "促销海报"],
+        "product_categories_visible": ["饮料", "零食"],
+        "promotion_elements": ["第二件半价海报"],
+        "atmosphere": "亲民生活化",
+        "has_person": False,
+    }
+    r = assembler.assemble_result(0, fj, [])
+    assert REQUIRED_KEYS <= set(r.keys())
+    assert r["name"] == "社区便利店"
+    assert r["brand"] == "全家FamilyMart"
+    assert r["category"] == "门店场景"
+    assert any("饮料" in str(f) for f in r["key_features"])
+    assert "门店实拍" in r["portrait_prompt"]
+
+
+# ---------- v4 other ----------
+
+
+def test_assemble_v4_other() -> None:
+    fj = {"type": "other", "description": "海边日落风景", "scene": "海边", "mood": "宁静"}
+    r = assembler.assemble_result(0, fj, [])
+    assert REQUIRED_KEYS <= set(r.keys())
+    assert r["name"] == "海边日落风景"
+    assert r["category"] == "非产品图"
+
+
+# ---------- 旧扁平 schema 兼容 ----------
+
+
+def test_assemble_old_flat_person() -> None:
+    fj = {
+        "has_person": True,
+        "gender": "男",
+        "age_range": "中年",
+        "upper_wear": "西装",
+        "upper_color": "深灰色",
+        "lower_wear": "西裤",
+        "lower_color": "黑色",
+        "accessories": ["手表"],
+        "hairstyle": "短发",
+        "expression": "严肃",
+        "scene": "办公室",
+        "style": "商务",
+        "mood": "专业",
+    }
+    r = assembler.assemble_result(0, fj, [])
+    assert REQUIRED_KEYS <= set(r.keys())
+    assert "中年男性" in r["portrait_prompt"]
+    assert r["_source"] == "v2_fast_json"
+
+
+def test_assemble_old_flat_product() -> None:
+    fj = {
+        "has_person": False,
+        "product_name": "口红",
+        "brand": "Dior",
+        "category": "美妆",
+        "colors": ["红色"],
+        "scene": "通用",
+        "style": "商业",
+        "mood": "高级",
+    }
+    r = assembler.assemble_result(0, fj, ["Dior"])
+    assert r["name"] == "口红"
+    assert r["brand"] == "Dior"
+    assert r["text_on_package"] == ["Dior"]
+
+
+def test_assemble_none_input() -> None:
+    r = assembler.assemble_result(0, None, [])
+    assert REQUIRED_KEYS <= set(r.keys())
+
+
+# ---------- _prompt 解析 ----------
+
+
+@pytest.fixture(autouse=True)
+def _clear_prompt_cache() -> Any:
+    _prompt.invalidate_cache()
+    yield
+    _prompt.invalidate_cache()
+
+
+def _fake_tpl(system_prompt: str = "v4 system prompt 只返回JSON") -> Any:
+    return types.SimpleNamespace(
+        system_prompt=system_prompt,
+        user_prompt_template="分析 {image_count} 张图",
+        version=4,
+    )
+
+
+def test_resolve_uses_db_prompt_without_append(monkeypatch: pytest.MonkeyPatch) -> None:
+    monkeypatch.setattr(_prompt, "_load_db_template", lambda: _fake_tpl("DB_V4_PROMPT_XYZ"))
+    sys_prompt, user_prompt = _prompt.resolve_fast_prompt()
+    assert sys_prompt == "DB_V4_PROMPT_XYZ"
+    assert "DB_V4_PROMPT_XYZ" not in _prompt._FAST_JSON_APPEND  # sanity: 旧append是另一段文本
+    assert "分析 1 张图" in user_prompt
+
+
+def test_resolve_pro_uses_db_prompt_without_append(monkeypatch: pytest.MonkeyPatch) -> None:
+    monkeypatch.setattr(_prompt, "_load_db_template", lambda: _fake_tpl("DB_V4_PRO_PROMPT"))
+    sys_prompt, _ = _prompt.resolve_pro_prompt()
+    assert sys_prompt == "DB_V4_PRO_PROMPT"
+    assert "【输出格式要求】" not in sys_prompt
+
+
+def test_resolve_falls_back_when_no_db(monkeypatch: pytest.MonkeyPatch) -> None:
+    monkeypatch.setattr(_prompt, "_load_db_template", lambda: None)
+    sys_prompt, user_prompt = _prompt.resolve_fast_prompt()
+    assert sys_prompt == _prompt._FAST_JSON_SCHEMA
+    assert user_prompt == _prompt.DEFAULT_FAST_USER
+
+
+def test_resolve_caches(monkeypatch: pytest.MonkeyPatch) -> None:
+    calls = {"n": 0}
+
+    def _load() -> Any:
+        calls["n"] += 1
+        return _fake_tpl("CACHED_PROMPT")
+
+    monkeypatch.setattr(_prompt, "_load_db_template", _load)
+    s1, _ = _prompt.resolve_fast_prompt()
+    s2, _ = _prompt.resolve_fast_prompt()
+    assert s1 == s2 == "CACHED_PROMPT"
+    assert calls["n"] == 1