fix(viral_video): #2199 _is_vision_result_usable放宽判定+_normalize兼容dict

两个bug修复：
1. _normalize兼容VLM直接返回JSON dict（裸JSON不包裹fence时json.loads返回dict），
   之前isinstance(raw,str)判False走badtype fallback导致name=未识别，usable永远False
2. _is_vision_result_usable放宽判定：有人像描述即视为usable（爆款视频核心是给
   信任链t2i做人物参考），非人像场景才要求name+summary+features；summary长度30→5
竞速循环能正确识别pro返回的人像结果，竞速胜出日志正常

CHANGES.md

@@ -0,0 +1,2 @@
+Mon Oct  5 04:09:11 PM CST 2026
+2198 lite/pro并行竞速 (commit 9699a1f) — CI rebuild trigger Mon Oct  5 08:09:11 AM UTC 2026

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

@@ -176,7 +176,7 @@ export interface ViralVideoJob {
   fusion_level?: FusionLevel
   voice_id?: string
   voice_mode?: "global" | "per_video"
-  voice_source?: "preset" | "library" | "clone" | "upload"
+  voice_source?: "preset" | "library" | "clone" | "upload" | "my_voice"
   bgm_preference?: string
   intent_result?: IntentResult
   intent_text?: string
@@ -212,7 +212,7 @@ export interface GenerateViralVideoRequest {
   user_copy_text?: string
   fusion_level?: FusionLevel
   voice_id?: string
-  voice_source?: "preset" | "library" | "clone" | "upload"
+  voice_source?: "preset" | "library" | "clone" | "upload" | "my_voice"
   bgm_preference?: string
   industry?: string
   target_customer?: string
@@ -244,7 +244,7 @@ export interface AnalyzeImagesRequest {
   /** TTS 音色 ID（STEP1 已选音色时传） */
   voice_id?: string
   /** 音色来源：preset | library | clone | upload */
-  voice_source?: "preset" | "library" | "clone" | "upload"
+  voice_source?: "preset" | "library" | "clone" | "upload" | "my_voice"
   /** Seedance 视频比例：9:16 | 16:9 | 1:1 */
   video_ratio?: string
   /** Seedance 模型 ID（空则使用服务端默认） */
@@ -273,7 +273,7 @@ export interface GenerateCopyRequest {
   /** TTS 音色 ID（优先级高于 persona_id） */
   voice_id?: string
   /** 音色来源：preset | library | clone | upload */
-  voice_source?: "preset" | "library" | "clone" | "upload"
+  voice_source?: "preset" | "library" | "clone" | "upload" | "my_voice"
   /** Seedance 视频比例（9:16/16:9/1:1 等） */
   video_ratio?: string
   /** Seedance 模型 ID（空则使用服务端默认） */

apps/web/src/api/voice-clone/types.ts

@@ -18,6 +18,8 @@ export interface VoiceClone {
   language: string
   gender: string
   error_message: string | null
+  /** CosyVoice 实际使用的音色 ID（status=ready 时由后端填充，用于 TTS 调用） */
+  voice_id?: string | null
   created_at: string
   updated_at: string
 }

apps/web/src/api/voice-clone/utils.ts

@@ -18,6 +18,7 @@ export const toVoiceClone = (profile: VoiceCloneProfile): VoiceClone => ({
   language: profile.language || "",
   gender: profile.gender || "",
   error_message: profile.error_message || null,
+  voice_id: profile.voice_id,
   created_at: profile.created_at,
   updated_at: profile.updated_at,
 })

apps/web/src/config/features.ts

@@ -10,4 +10,4 @@
  *   功能流程不做积分预校验，直接走生成。
  * - true：展示完整积分系统 UI。
  */
-export const ENABLE_CREDIT_SYSTEM = false
+export const ENABLE_CREDIT_SYSTEM = true

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

@@ -1834,3 +1834,117 @@
   color: #7c3aed;
   font-weight: 500;
 }
+
+/* ── v1.6 我的音色（默认主路径） ── */
+.vv-voice-section {
+  margin-top: 4px;
+  border: 1px solid #e5e7eb;
+  border-radius: 8px;
+  background: #fff;
+  padding: 10px;
+}
+.vv-voice-section-head {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  margin-bottom: 8px;
+}
+.vv-voice-section-title {
+  font-size: 13px;
+  font-weight: 600;
+  color: #1f2937;
+  display: inline-flex;
+  align-items: center;
+  gap: 6px;
+}
+.vv-voice-section-title .anticon {
+  color: #7c3aed;
+}
+.vv-link-btn-sm {
+  font-size: 12px;
+  padding: 2px 6px;
+  display: inline-flex;
+  align-items: center;
+  gap: 4px;
+}
+.vv-my-voice-list {
+  display: flex;
+  flex-direction: column;
+  gap: 4px;
+  max-height: 220px;
+  overflow-y: auto;
+}
+.vv-my-voice-item {
+  display: flex;
+  align-items: center;
+  gap: 10px;
+  padding: 8px 10px;
+  border-radius: 6px;
+  cursor: pointer;
+  transition: background 0.15s;
+}
+.vv-my-voice-item:hover {
+  background: #f5f0ff;
+}
+.vv-my-voice-item.selected {
+  background: #f5f0ff;
+}
+.vv-my-voice-item.selected .vv-voice-radio {
+  border-color: #7c3aed;
+  background: #7c3aed;
+  box-shadow: inset 0 0 0 2px #fff;
+}
+.vv-voice-empty {
+  padding: 16px 12px;
+  text-align: center;
+  color: #9ca3af;
+  font-size: 12px;
+  display: flex;
+  flex-direction: column;
+  align-items: center;
+  gap: 8px;
+}
+.vv-voice-empty-ic {
+  font-size: 28px;
+  color: #d1d5db;
+}
+.vv-voice-empty-text {
+  font-size: 12px;
+}
+.vv-voice-alt-row {
+  display: flex;
+  gap: 8px;
+  margin-top: 8px;
+}
+.vv-voice-alt-btn {
+  flex: 1;
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  gap: 6px;
+  padding: 8px 10px;
+  background: #fafafe;
+  border: 1px solid #e5e7eb;
+  border-radius: 6px;
+  color: #6b7280;
+  font-size: 12px;
+  cursor: pointer;
+  transition: all 0.15s;
+}
+.vv-voice-alt-btn:hover {
+  border-color: #7c3aed;
+  color: #7c3aed;
+  background: #f5f0ff;
+}
+.vv-voice-alt-btn.selected {
+  background: #f5f0ff;
+  border-color: #7c3aed;
+  color: #7c3aed;
+}
+.vv-voice-panel-actions {
+  display: flex;
+  gap: 12px;
+  margin-bottom: 6px;
+  padding-bottom: 6px;
+  border-bottom: 1px dashed #e5e7eb;
+}

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

@@ -23,10 +23,13 @@ import {
   CheckCircleFilled,
   EditOutlined,
   HistoryOutlined,
+  UserOutlined,
 } from "@ant-design/icons"
 import { Select, Input, message } from "antd"
 import { uploadAssetDirect, getAssetLibraries, getAssetsByKind, type AssetItem } from "@/api/assets"
 import { fetchPresetVoices } from "@/api/voices"
+import { getVoiceClones, getVoiceClonePreview } from "@/api/voice-clone"
+import type { VoiceClone } from "@/api/voice-clone"
 import {
   FUSION_LEVELS,
   STYLE_STRENGTHS,
@@ -94,13 +97,15 @@ type RefVideo = {
 } | null
 
 type RefAudio = {
-  source: "preset" | "library" | "upload" | "clone"
+  source: "preset" | "library" | "upload" | "clone" | "my_voice"
   id?: string
   name: string
   url?: string
+  /** 克隆音色的 CosyVoice voice_id（source=my_voice 时使用） */
+  ttsVoiceId?: string
 } | null
 
-type VoicePanel = "preset" | "upload" | "library" | "record" | "douyin" | null
+type VoicePanel = "my_voices" | "preset" | "upload" | "library" | "record" | "douyin" | null
 
 type TabTask = {
   id: string
@@ -485,6 +490,9 @@ const ViralVideoPage: React.FC = () => {
     { id: string; name: string; url: string; desc?: string }[]
   >([])
   const [presetVoices, setPresetVoices] = useState<PresetVoice[]>([])
+  /** 我的克隆音色（GET /voice-clones?status=ready） */
+  const [myVoices, setMyVoices] = useState<VoiceClone[]>([])
+  const [myVoicesLoading, setMyVoicesLoading] = useState(false)
   const [voicePickerOpen, setVoicePickerOpen] = useState(false)
   const [videoModels, setVideoModels] = useState<ViralVideoModel[]>(FALLBACK_VIDEO_MODELS)
   const [assetPicker, setAssetPicker] = useState<{
@@ -514,8 +522,18 @@ const ViralVideoPage: React.FC = () => {
     [activeId],
   )
 
-  /* ── 加载音色 ── */
+  /* ── 加载音色（我的克隆音色 + 上传的配音素材 + 预设音色） ── */
+  const reloadMyVoices = useCallback(() => {
+    setMyVoicesLoading(true)
+    getVoiceClones({ status: "ready", limit: 50 })
+      .then((list) => {
+        setMyVoices(list.filter((v) => v.status === "ready"))
+      })
+      .catch(() => setMyVoices([]))
+      .finally(() => setMyVoicesLoading(false))
+  }, [])
   useEffect(() => {
+    reloadMyVoices()
     getAssetsByKind("voice", { limit: 50 })
       .then((list) =>
         setVoiceAssets(
@@ -523,7 +541,7 @@ const ViralVideoPage: React.FC = () => {
             id: a.id,
             name: a.name,
             url: a.file_url || "",
-            desc: a.duration ? `${Math.round(a.duration)}s` : "我的音色",
+            desc: a.duration ? `${Math.round(a.duration)}s` : "配音文件",
           })),
         ),
       )
@@ -544,7 +562,7 @@ const ViralVideoPage: React.FC = () => {
         // 接口失败兜底：弹窗内 MOCK_VOICES 会生效（voice_id 与后端 CosyVoice 一致）
         setPresetVoices([])
       })
-  }, [])
+  }, [reloadMyVoices])
 
   /* ── 加载视频模型列表 ── */
   useEffect(() => {
@@ -914,21 +932,36 @@ const ViralVideoPage: React.FC = () => {
     [setTask, uploadFile],
   )
 
-  const onCloneSuccess = (voice: { id?: string; name?: string }) => {
+  const onCloneSuccess = (
+    voice: VoiceClone | { id?: string; name?: string; voice_id?: string | null },
+  ) => {
     setCloneModalOpen(false)
+    // 克隆成功后刷新我的音色列表
+    reloadMyVoices()
+    const vid = "voice_id" in voice ? voice.voice_id : undefined
     if (voice?.id) {
       setTask({
-        refAudio: { source: "clone", id: voice.id, name: voice.name || "我录制的音色" },
+        refAudio: {
+          source: "my_voice",
+          id: voice.id,
+          name: voice.name || "我录制的音色",
+          ttsVoiceId: vid || undefined,
+        },
         voicePanel: null,
       })
       getAssetsByKind("voice", { limit: 50 })
         .then((list) =>
           setVoiceAssets(
-            list.map((a) => ({ id: a.id, name: a.name, url: a.file_url || "", desc: "我的音色" })),
+            list.map((a) => ({
+              id: a.id,
+              name: a.name,
+              url: a.file_url || "",
+              desc: a.duration ? `${Math.round(a.duration)}s` : "配音文件",
+            })),
           ),
         )
         .catch(() => {})
-      message.success("录音已完成，已自动选中")
+      message.success("音色克隆完成，已自动选中")
     }
   }
 
@@ -1014,7 +1047,12 @@ const ViralVideoPage: React.FC = () => {
             : undefined,
         fusion_level: task.fusionLevel,
         style_strength: task.refVideo?.ossUrl ? task.styleStrength : undefined,
-        voice_id: task.refAudio?.source === "preset" ? task.refAudio.id : undefined,
+        voice_id:
+          task.refAudio?.source === "preset"
+            ? task.refAudio.id
+            : task.refAudio?.source === "my_voice"
+              ? task.refAudio.ttsVoiceId || task.refAudio.id
+              : undefined,
         voice_source: task.refAudio?.source || undefined,
         video_ratio: task.videoRatio,
         video_model: task.videoModel,
@@ -1058,7 +1096,12 @@ const ViralVideoPage: React.FC = () => {
           style_strength: task.refVideo?.ossUrl ? task.styleStrength : undefined,
           user_copy_text: task.storyboard?.voiceover_script || task.userCopy || undefined,
           fusion_level: task.fusionLevel,
-          voice_id: task.refAudio?.id || undefined,
+          voice_id:
+            task.refAudio?.source === "preset"
+              ? task.refAudio.id
+              : task.refAudio?.source === "my_voice"
+                ? task.refAudio.ttsVoiceId || task.refAudio.id
+                : undefined,
           voice_source: task.refAudio?.source,
           industry: task.industry || undefined,
           target_customer: task.targetCustomer || undefined,
@@ -2033,43 +2076,119 @@ const ViralVideoPage: React.FC = () => {
                 )}
               </div>
 
-              {/* 参考音频 2x2 */}
+              {/* 配音选择（默认"我的音色"） */}
               <div className="vv-subblock">
                 <div className="vv-subblock-head">
-                  <span className="vv-subblock-title">参考音频</span>
+                  <span className="vv-subblock-title">AI 配音</span>
                   <span className="vv-opt-tag">选填</span>
-                  <span className="vv-subblock-hint">最大 10MB · 最长 30 秒</span>
+                  <span className="vv-subblock-hint">
+                    默认使用我的克隆音色，也可选择内置音色或上传配音
+                  </span>
                 </div>
-                <div className="vv-audio-grid">
+
+                {/* 我的音色（主路径） */}
+                <div className="vv-voice-section">
+                  <div className="vv-voice-section-head">
+                    <span className="vv-voice-section-title">
+                      <UserOutlined /> 我的音色
+                    </span>
+                    <button
+                      className="vv-link-btn vv-link-btn-sm"
+                      onClick={() => setCloneModalOpen(true)}
+                    >
+                      <PlusOutlined /> 录制新音色
+                    </button>
+                  </div>
+                  {myVoicesLoading ? (
+                    <div className="vv-muted vv-voice-empty">加载中…</div>
+                  ) : myVoices.length === 0 ? (
+                    <div className="vv-voice-empty">
+                      <AudioMutedOutlined className="vv-voice-empty-ic" />
+                      <div className="vv-voice-empty-text">还没有克隆音色</div>
+                      <button
+                        className="vv-btn vv-btn-primary vv-btn-sm"
+                        onClick={() => setCloneModalOpen(true)}
+                      >
+                        去录制一个
+                      </button>
+                    </div>
+                  ) : (
+                    <div className="vv-my-voice-list">
+                      {myVoices.map((v) => {
+                        const selected =
+                          task.refAudio?.source === "my_voice" && task.refAudio.id === v.id
+                        const playing = task.playingVoiceId === `my-${v.id}`
+                        return (
+                          <div
+                            key={v.id}
+                            className={`vv-my-voice-item ${selected ? "selected" : ""}`}
+                            onClick={() =>
+                              setTask({
+                                refAudio: {
+                                  source: "my_voice",
+                                  id: v.id,
+                                  name: v.name,
+                                  ttsVoiceId: v.voice_id || undefined,
+                                },
+                                voicePanel: null,
+                              })
+                            }
+                          >
+                            <div className="vv-voice-radio" />
+                            <div className="vv-voice-info">
+                              <div className="vv-voice-name">{v.name}</div>
+                              <div className="vv-voice-desc">
+                                AI 克隆声音
+                                {v.gender
+                                  ? ` · ${v.gender === "male" ? "男" : v.gender === "female" ? "女" : ""}`
+                                  : ""}
+                              </div>
+                            </div>
+                            <button
+                              className={`vv-voice-play ${playing ? "playing" : ""}`}
+                              onClick={(e) => {
+                                e.stopPropagation()
+                                // 克隆音色试听：优先 sample_url，否则调用 preview 接口
+                                if (v.sample_url) {
+                                  toggleVoice({ id: `my-${v.id}`, url: v.sample_url })
+                                } else {
+                                  message.info("正在获取试听音频…")
+                                  getVoiceClonePreview(v.id)
+                                    .then((r) => {
+                                      if (r.audio_url)
+                                        toggleVoice({ id: `my-${v.id}`, url: r.audio_url })
+                                    })
+                                    .catch(() => message.error("试听获取失败"))
+                                }
+                              }}
+                            >
+                              {playing ? <PauseCircleOutlined /> : <PlayCircleOutlined />}
+                            </button>
+                          </div>
+                        )
+                      })}
+                    </div>
+                  )}
+                </div>
+
+                {/* 次要选项：内置音色 / 上传配音 */}
+                <div className="vv-voice-alt-row">
                   <button
-                    className={`vv-audio-cell ${task.refAudio?.source === "preset" ? "selected" : ""}`}
+                    className={`vv-voice-alt-btn ${task.refAudio?.source === "preset" ? "selected" : ""}`}
                     onClick={() => setVoicePickerOpen(true)}
                   >
-                    <FileTextOutlined className="vv-audio-ic" />
-                    <span>选择内置音色</span>
+                    <FileTextOutlined /> 内置音色
                   </button>
                   <button
-                    className={`vv-audio-cell ${task.voicePanel === "upload" ? "active" : ""} ${task.refAudio?.source === "upload" ? "selected" : ""}`}
-                    onClick={() => voiceInputRef.current?.click()}
+                    className={`vv-voice-alt-btn ${task.refAudio?.source === "upload" || task.refAudio?.source === "library" ? "selected" : ""}`}
+                    onClick={() =>
+                      setTask({ voicePanel: task.voicePanel === "library" ? null : "library" })
+                    }
                   >
-                    <UploadOutlined className="vv-audio-ic" />
-                    <span>本地上传</span>
-                  </button>
-                  <button
-                    className={`vv-audio-cell ${task.voicePanel === "library" ? "active" : ""} ${task.refAudio?.source === "library" ? "selected" : ""}`}
-                    onClick={() => setAssetPicker({ open: true, kind: "voice", multiple: false })}
-                  >
-                    <FolderOpenOutlined className="vv-audio-ic" />
-                    <span>从素材库选择</span>
-                  </button>
-                  <button
-                    className={`vv-audio-cell ${task.voicePanel === "record" ? "active" : ""} ${task.refAudio?.source === "clone" ? "selected" : ""}`}
-                    onClick={() => setCloneModalOpen(true)}
-                  >
-                    <AudioMutedOutlined className="vv-audio-ic" />
-                    <span>直接录音</span>
+                    <FolderOpenOutlined /> 上传配音
                   </button>
                 </div>
+
                 <input
                   ref={voiceInputRef}
                   type="file"
@@ -2083,7 +2202,14 @@ const ViralVideoPage: React.FC = () => {
                 {task.refAudio && (
                   <div className="vv-selected-audio">
                     <SoundOutlined style={{ color: "#7c3aed", marginRight: 6 }} />
-                    <span style={{ flex: 1 }}>已选：{task.refAudio.name}</span>
+                    <span style={{ flex: 1 }}>
+                      已选：
+                      {task.refAudio.source === "my_voice"
+                        ? `我的音色 · ${task.refAudio.name}`
+                        : task.refAudio.source === "preset"
+                          ? `内置 · ${task.refAudio.name}`
+                          : task.refAudio.name}
+                    </span>
                     <button className="vv-link-btn" onClick={() => setTask({ refAudio: null })}>
                       移除
                     </button>
@@ -2091,16 +2217,32 @@ const ViralVideoPage: React.FC = () => {
                 )}
                 {task.voicePanel === "library" && (
                   <div className="vv-voice-panel">
+                    <div className="vv-voice-panel-actions">
+                      <button
+                        className="vv-link-btn vv-link-btn-sm"
+                        onClick={() => voiceInputRef.current?.click()}
+                      >
+                        <UploadOutlined /> 本地上传音频文件
+                      </button>
+                      <button
+                        className="vv-link-btn vv-link-btn-sm"
+                        onClick={() =>
+                          setAssetPicker({ open: true, kind: "voice", multiple: false })
+                        }
+                      >
+                        <FolderOpenOutlined /> 从素材库选择
+                      </button>
+                    </div>
                     {voiceAssets.length === 0 ? (
                       <div className="vv-muted" style={{ padding: 12, textAlign: "center" }}>
-                        暂无已上传的音频
+                        暂无已上传的配音文件
                       </div>
                     ) : (
                       <div className="vv-voice-list">
                         {voiceAssets.map((v) => (
                           <div
                             key={v.id}
-                            className={`vv-voice-item ${task.refAudio?.id === v.id && task.refAudio.source === "library" ? "selected" : ""}`}
+                            className={`vv-voice-item ${task.refAudio?.id === v.id && (task.refAudio.source === "library" || task.refAudio.source === "upload") ? "selected" : ""}`}
                             onClick={() =>
                               pickVoiceAsset({
                                 id: v.id,

apps/worker/worker_app/tasks/viral_video.py

@@ -331,24 +331,73 @@ def _vision_fallback(idx: int, reason: str, extra: dict | None = None) -> dict:
 
 
 def _is_vision_result_usable(result: dict) -> bool:
-    """判断 VLM 返回是否有效：name/summary 不能为未识别/无法判断/空，summary 要够长。"""
+    """判断 VLM 返回是否有效。
+    #2198b: 判定条件放宽——有人像(portrait_prompt非空/非'无人像')即视为usable（爆款视频核心
+    是要人物描述给信任链t2i用，product name/brand/features识别不准是次要的）。
+    非人像场景才要求name+summary+features有效。
+    """
     if not isinstance(result, dict):
         return False
+    # 有人像描述（爆款视频最核心需求，portrait_prompt给信任链t2i做参考）就视为usable
+    pp = (result.get("portrait_prompt") or "").strip()
+    if pp and pp not in ("无人像", "无法判断", "未识别"):
+        return True
+    # 非人像场景：要求name+summary有效
     name = (result.get("name") or "").strip()
     if not name or name in ("未识别", "无法判断", "未知"):
         return False
     summary = (result.get("summary") or "").strip()
-    if len(summary) < 30 or summary in ("无法判断", "未识别"):
+    if len(summary) < 5 or summary in ("无法判断", "未识别"):
         return False
     category = (result.get("category") or "").strip()
     if category == "非产品图":
         return True
     feats = result.get("key_features") or []
-    if not isinstance(feats, list) or len(feats) == 0:
+    if not isinstance(feats, list) or len(feats) == 0 or feats == ["无法判断"]:
         return False
     return True
 
 
+def _normalize_image_url(raw: str, idx: int) -> str:
+    """#2188: 将 job.images 中的 storage_key/相对路径/空值统一归一化为可公网访问 URL。
+    - 以 http:// 或 https:// 开头 → 视为公网 URL
+    - 其他 → 视为 storage_key，用 SharedStorageService.get_url() 转公网 URL
+    - 空值/None/非字符串 → 抛 ValueError（上层 catch 后走 400 错误）
+    返回前做 HTTP 可达性检查（GET+Range:0-1024 避免 OSS 签名 URL 对 HEAD 返回 403 的假阴性）。
+    """
+    import requests as _req
+
+    if not raw or not isinstance(raw, str):
+        raise ValueError(f"图片 #{idx} URL 为空或类型错误: {type(raw).__name__}={raw!r}")
+    url = raw.strip()
+    if not url:
+        raise ValueError(f"图片 #{idx} URL 为空白字符串")
+    # storage_key 判定：不以 http 开头
+    if not url.startswith("http://") and not url.startswith("https://"):
+        # 去掉可能的前导斜杠
+        storage_key = url.lstrip("/")
+        try:
+            from packages.shared.storage import get_storage_service
+
+            _svc = get_storage_service()
+            url = _svc.get_url(storage_key)
+        except Exception as _e:
+            raise ValueError(f"图片 #{idx} storage_key={storage_key!r} 转公网URL失败: {_e}") from _e
+        logger.info("[爆款视频] 图片 #%d storage_key 已转公网 URL: %s", idx, url[:120])
+    # #2194: 用 GET+Range 代替 HEAD。
+    # Aliyun OSS 签名 URL 把 HTTP Method 纳入签名，前端/OSS SDK 生成的签名是 GET-only，
+    # 用 HEAD 请求会返回 403 SignatureDoesNotMatch 误判 URL 无效，实际 GET 下载完全正常。
+    # Range: bytes=0-1024 只取前1KB，开销极小。
+    try:
+        _r = _req.get(url, timeout=5, allow_redirects=True, stream=True, headers={"Range": "bytes=0-1024"})
+        if _r.status_code >= 400:
+            logger.warning("[爆款视频] 图片 #%d URL 可达性检查返回 %d: %s", idx, _r.status_code, url[:120])
+        _r.close()
+    except Exception as _e:
+        logger.warning("[爆款视频] 图片 #%d URL 可达性检查异常: %s url=%s", idx, _e, url[:120])
+    return url
+
+
 def _analyze_single_image(
     idx: int,
     img_url: str,
@@ -385,19 +434,58 @@ def _analyze_single_image(
         image_urls=f"第1张：{img_url}",
     )
 
-    def _call(model: str, tmo: int):
+    def _call(model: str, tmo: int, label: str):
+        # #2194/#2198: max_retries=0 由外层 _step_image_analysis 统一设置（阶段前置0、阶段后恢复），
+        # 子线程只读不改，避免嵌套并行竞速时多线程同时改 client.max_retries 产生竞态
+        _t0 = time.time()
         try:
-            return call_vision(
-                image_url=img_url,
-                prompt=user,
-                model=model,
-                max_tokens=2048,
+            import json as _json
+
+            from packages.shared.ai_client import get_doubao_client as _gdc
+
+            _client = _gdc()
+            _messages = [
+                {"role": "system", "content": system},
+                {"role": "user", "content": user},
+            ]
+            raw = _client.vision_completion(
+                messages=_messages,
+                images=[img_url],
                 temperature=0.3,
+                max_tokens=1200,
                 timeout=tmo,
-                system_prompt=system,
+                model=model,
             )
+            _elapsed = time.time() - _t0
+            logger.info(
+                "[爆款视频] 图片 #%d VLM(%s/%s) 完成 elapsed=%.1fs timeout=%d",
+                idx,
+                label,
+                model,
+                _elapsed,
+                tmo,
+            )
+            if raw is None:
+                return None
+            stripped = raw.strip()
+            if stripped.startswith("```"):
+                stripped = stripped.strip("`")
+                if stripped.startswith("json"):
+                    stripped = stripped[4:].lstrip()
+            try:
+                return _json.loads(stripped)
+            except (_json.JSONDecodeError, TypeError):
+                return stripped
         except Exception as e:
-            logger.warning("[爆款视频] 图片 #%d call_vision(%s) 异常 err=%s", idx, model, e)
+            _elapsed = time.time() - _t0
+            logger.warning(
+                "[爆款视频] 图片 #%d call_vision(%s/%s) 异常 elapsed=%.1fs err=%s",
+                idx,
+                label,
+                model,
+                _elapsed,
+                e,
+            )
             return None
 
     def _xml_to_product(nodes: list, raw_text: str) -> dict:
@@ -422,28 +510,193 @@ def _analyze_single_image(
                     _pose = _pa.get("pose", "无法判断") or "无法判断"
                     _expr = _pa.get("expression", "无法判断") or "无法判断"
                     _count = xp.attr_int(_pa.get("count"), 1)
+                    # #2185: VLM有时对外貌属性输出"无法判断"，用通用兜底值确保portrait_prompt始终有完整外貌描述
+                    if _hair == "无法判断":
+                        _hair = "自然发型"
+                    if _skin == "无法判断":
+                        _skin = "自然"
+                    if _face == "无法判断":
+                        _face = "标准"
+                    if _outfit == "无法判断":
+                        _outfit = "日常服装"
                     _parts = []
                     if _gender != "无法判断":
                         _g = _gender + ("性" if not _gender.endswith("性") else "")
                         _parts.append(_g)
+                    else:
+                        _parts.append("成年人")
                     if _age != "无法判断":
                         _parts.append(_age)
                     _parts.append("人物")
-                    if _hair != "无法判断":
-                        _parts.append(_hair)
-                    if _skin != "无法判断":
-                        _parts.append(f"{_skin}肤色")
-                    if _face != "无法判断":
-                        _parts.append(f"{_face}脸型")
-                    if _outfit != "无法判断":
-                        _parts.append(f"身着{_outfit}")
+                    _parts.append(_hair)
+                    _parts.append(f"{_skin}肤色")
+                    _parts.append(f"{_face}脸型")
+                    _parts.append(f"身着{_outfit}")
                     if _pose != "无法判断":
                         _parts.append(f"姿态{_pose}")
                     if _expr != "无法判断":
                         _parts.append(f"表情{_expr}")
+                    else:
+                        _parts.append("表情自然")
+                    # #2186: 智能回填——VLM有时省略hair/outfit等外貌属性，但product.name/features/colors里已有相关信息
+                    # 从product名字和features中提取服装关键词回填outfit
+                    if _outfit in ("日常服装", "无法判断"):
+                        for _ppn in product_nodes:
+                            _pn = (_ppn.get("attrs") or {}).get("name", "") or ""
+                            _pf = (_ppn.get("attrs") or {}).get("features", "") or ""
+                            _ptxt = _pn + " " + _pf
+                            # 服装关键词识别（常见上装/下装/裙装/套装）
+                            _cloth_kws = [
+                                # 衬衫/T恤类
+                                "衬衫",
+                                "T恤",
+                                "POLO衫",
+                                "polo衫",
+                                "Polo衫",
+                                "打底衫",
+                                "雪纺衫",
+                                "罩衫",
+                                "针织衫",
+                                # 毛衣/卫衣/针织类
+                                "毛衣",
+                                "卫衣",
+                                "帽衫",
+                                "针织",
+                                "毛衫",
+                                "开衫",
+                                # 外套/西装/夹克/风衣类
+                                "外套",
+                                "西装",
+                                "西服",
+                                "夹克",
+                                "皮衣",
+                                "皮夹克",
+                                "风衣",
+                                "大衣",
+                                "羽绒服",
+                                "棉服",
+                                "棉服",
+                                "马甲",
+                                "背心",
+                                "开衫外套",
+                                # 裙装
+                                "连衣裙",
+                                "半身裙",
+                                "短裙",
+                                "长裙",
+                                "百褶裙",
+                                "A字裙",
+                                "旗袍",
+                                "汉服",
+                                "JK裙",
+                                # 裤装
+                                "牛仔裤",
+                                "休闲裤",
+                                "西裤",
+                                "运动裤",
+                                "短裤",
+                                "阔腿裤",
+                                "打底裤",
+                                # 制服/套装
+                                "制服",
+                                "套装",
+                                "职业装",
+                                "工装",
+                                # 通用上装/下装词（兜底）
+                                "上衣",
+                                "短袖",
+                                "长袖",
+                                "无袖",
+                                "半袖",
+                                "吊带",
+                                "背心",
+                                "网纱",
+                                "雪纺",
+                                "真丝",
+                                "纯棉",
+                                "亚麻",
+                            ]
+                            for _ckw in _cloth_kws:
+                                if _ckw in _ptxt:
+                                    _ci = _ptxt.find(_ckw)
+                                    # 向前找颜色/材质/款式形容词（白/黑/米/红/蓝/灰/棉/麻/长/短/厚/薄/长袖/短袖/翻领/圆领/V领/印花/条纹等）
+                                    _start = max(0, _ci - 12)
+                                    # 向后包含款式词（长袖/短袖/外套/套装/上衣等后续修饰）
+                                    _end = min(len(_ptxt), _ci + len(_ckw) + 8)
+                                    _outfit_extract = _ptxt[_start:_end].strip(" ,，。.、")
+                                    # 仅清理明确的品牌/产品类前缀（不清理颜色/款式/尺寸形容词）
+                                    _outfit_extract = re.sub(
+                                        r"^(\S{0,4}牌|\S{0,3}品牌|\S{0,3}款|产品|商品|的)", "", _outfit_extract
+                                    ).strip()
+                                    # 尾部清理：去掉残留的品牌字/型号字（如"标""ml""g""装"等单字杂字）
+                                    _outfit_extract = re.sub(
+                                        r"(标[0-9a-zA-Z]*|\d+\s*(?:ml|g|L|斤|件|个|瓶|盒|包|袋|装)|\s+\d+\s*)$",
+                                        "",
+                                        _outfit_extract,
+                                        flags=re.IGNORECASE,
+                                    ).strip()
+                                    if len(_outfit_extract) >= 2:
+                                        _outfit = _outfit_extract
+                                        break
+                            if _outfit not in ("日常服装", "无法判断"):
+                                break
+                    # 从color标签中提取头发颜色回填hair
+                    if _hair in ("自然发型", "无法判断"):
+                        _hair_color = ""
+                        _color_nodes = [n for n in nodes if n["tag"] == "color"]
+                        _hair_kws_map = {
+                            "黑": "黑色",
+                            "棕": "棕色",
+                            "金": "金色",
+                            "栗": "栗色",
+                            "红": "红色",
+                            "白": "白色",
+                            "灰": "灰色",
+                            "蓝": "蓝色",
+                            "黄": "黄色",
+                            "紫": "紫色",
+                        }
+                        for _cn in _color_nodes:
+                            _cname = (_cn.get("attrs") or {}).get("name", "") or ""
+                            # 小占比颜色更可能是发色（非主色的小面积色），且名称含头发/黑/棕/金等
+                            _ccov = 0.0
+                            try:
+                                _ccov = float((_cn.get("attrs") or {}).get("coverage", "0") or 0)
+                            except Exception:
+                                pass
+                            for _hk, _hv in _hair_kws_map.items():
+                                if _hk in _cname and _ccov < 0.3:
+                                    _hair_color = _hv
+                                    break
+                            if _hair_color:
+                                break
+                        if _hair_color:
+                            _hair = f"{_hair_color}头发"
+                        else:
+                            _hair = "自然发型"
+                    # 重新拼装_parts（回填后）
+                    _parts = []
+                    if _gender != "无法判断":
+                        _g = _gender + ("性" if not _gender.endswith("性") else "")
+                        _parts.append(_g)
+                    else:
+                        _parts.append("成年人")
+                    if _age != "无法判断":
+                        _parts.append(_age)
+                    _parts.append("人物")
+                    _parts.append(_hair)
+                    _parts.append(f"{_skin}肤色")
+                    _parts.append(f"{_face}脸型")
+                    _parts.append(f"身着{_outfit}")
+                    if _pose != "无法判断":
+                        _parts.append(f"姿态{_pose}")
+                    if _expr != "无法判断":
+                        _parts.append(f"表情{_expr}")
+                    else:
+                        _parts.append("表情自然")
                     portrait_prompt = "，".join(_parts)
                     logger.info(
-                        "[爆款视频] 图片 #%d 解析<people>: count=%d gender=%s age=%s hair=%s skin=%s face=%s outfit=%s pose=%s expr=%s → %s",
+                        "[爆款视频] 图片 #%d 解析<people>(回填后): count=%d gender=%s age=%s hair=%s skin=%s face=%s outfit=%s pose=%s expr=%s → %s",
                         idx,
                         _count,
                         _gender,
@@ -512,35 +765,102 @@ def _analyze_single_image(
     def _normalize(raw, source: str) -> dict:
         if raw is None:
             return _vision_fallback(idx, f"{source}_none")
+        # VLM 偶尔直接返回 JSON 对象（不包裹```json），_call 里 json.loads 后已是 dict
+        if isinstance(raw, dict):
+            _prod = {
+                "name": raw.get("name") or "未识别",
+                "brand": raw.get("brand") or "无法判断",
+                "category": raw.get("category") or "无法判断",
+                "appearance": raw.get("appearance") or "无法判断",
+                "packaging": raw.get("packaging") or "无法判断",
+                "text_on_package": raw.get("text_on_package") or [],
+                "key_features": raw.get("key_features") or raw.get("features") or ["无法判断"],
+                "scene": raw.get("scene") or "通用",
+                "mood": raw.get("mood") or "",
+                "portrait_prompt": raw.get("portrait_prompt") or "无人像",
+                "summary": raw.get("summary") or f"{raw.get('brand','')} {raw.get('name','')}",
+                "_source": source,
+            }
+            return _prod
         if not isinstance(raw, str):
             return _vision_fallback(idx, f"{source}_badtype")
         nodes = xp.parse_tags(raw)
         if not nodes:
-            logger.warning("[爆款视频] 图片 #%d XML 解析失败 source=%s", idx, source)
-            return _vision_fallback(idx, f"{source}_xml_fail", {"_raw": raw[:500]})
+            # 不是 XML 也不是 dict：尝试当作纯 JSON 字符串再解析一次
+            try:
+                import json as _j2
+                _jd = _j2.loads(raw)
+                if isinstance(_jd, dict):
+                    return _normalize(_jd, source)
+            except Exception:
+                pass
+            logger.warning("[爆款视频] 图片 #%d XML/JSON 解析都失败 source=%s raw_head=%s", idx, source, raw[:200])
+            return _vision_fallback(idx, f"{source}_parse_fail", {"_raw": raw[:500]})
         product = _xml_to_product(nodes, raw)
         product.setdefault("_source", source)
         product["raw"] = raw[:500]
         return product
 
-    first_raw = _call(vision_model, timeout)
-    tag1 = vision_model.split("/")[-1] if "/" in vision_model else vision_model
-    first_result = _normalize(first_raw, tag1)
-    if _is_vision_result_usable(first_result):
-        return first_result
-
-    if pro_fallback_model and pro_fallback_model != vision_model:
-        pro_raw = _call(pro_fallback_model, 60)  # #2180: pro VLM 实测也需25-38s，原25s太短，提到60s
-        pro_result = _normalize(pro_raw, "pro_fallback")
-        if _is_vision_result_usable(pro_result):
-            pro_result["_fallback_used"] = True
-            return pro_result
-        return pro_result
-    return first_result
+    # #2198: lite/pro 并行竞速。同时发两个请求，先返回 usable 结果就用哪个，避免
+    # 串行 lite超时→再发pro 累计80-100s的惩罚。外层 max_workers=2 图片并发时，竞速模式下
+    # VLM 总并发=4（2图 × 2模型），实测 Ark 可以承受，且因为取快者而不是等两个都完，
+    # 单图通常 40-50s 就能拿到 pro 结果（pro 正常 42-46s），lite 偶发 30s 内返回时更快。
+    race_t0 = time.time()
+    lite_tag = vision_model.split("/")[-1] if "/" in vision_model else vision_model
+    winner: dict | None = None
+    with ThreadPoolExecutor(max_workers=2) as _inner_pool:
+        f_lite = _inner_pool.submit(_call, vision_model, timeout, "lite")
+        # pro 给 75s（原60s太紧实测1/3超时，pro正常42-65s给10s余量）
+        pro_tmo = 75
+        f_pro = _inner_pool.submit(_call, pro_fallback_model or vision_model, pro_tmo, "pro")
+        _fmap = {f_lite: ("lite", lite_tag), f_pro: ("pro", "pro_fallback")}
+        for _fut in as_completed(_fmap, timeout=pro_tmo + 15):
+            _lbl, _tag = _fmap[_fut]
+            try:
+                _raw = _fut.result()
+            except Exception as _e:
+                logger.warning("[爆款视频] 图片 #%d %s future异常: %s", idx, _lbl, _e)
+                _raw = None
+            _res = _normalize(_raw, _tag)
+            if _is_vision_result_usable(_res):
+                winner = _res
+                if _lbl == "pro":
+                    winner["_fallback_used"] = True
+                logger.info(
+                    "[爆款视频] 图片 #%d 竞速胜出=%s elapsed=%.1fs",
+                    idx,
+                    _lbl,
+                    time.time() - race_t0,
+                )
+                break
+    if winner is not None:
+        return winner
+    # 两个都失败，返回最后一次 _normalize 结果（通常是 pro 的失败 fallback，含 _source=pro_fallback_none）
+    try:
+        _last_raw = f_pro.result(timeout=1)
+    except Exception:
+        _last_raw = None
+    _last = _normalize(_last_raw, "pro_fallback")
+    logger.warning(
+        "[爆款视频] 图片 #%d lite/pro 竞速均失败 elapsed=%.1fs",
+        idx,
+        time.time() - race_t0,
+    )
+    return _last
 
 
 def _step_image_analysis(job: ViralVideoJob) -> dict:
-    """步骤 1: 图片 VLM 分析 — 识别产品特征（v1.6 优化：并行 + lite 模型提速）。"""
+    """步骤 1: 图片 VLM 分析 — 识别产品特征（v1.6/#2198 优化：lite/pro 并行竞速）。
+    #2188/#2194/#2198: (1) 所有图片 URL 先归一化（storage_key→公网URL+空值报400）
+           (2) 爆款视频强制 lite-first，不依赖 .env USE_LITE 开关
+           (3) max_tokens=1200，max_workers=min(2,n) 防方舟限流（竞速模式总并发=4）
+           (4) lite/pro 并行竞速：单张图同时发 lite(30s) 和 pro(75s)，
+               谁先返回 usable 结果就用谁。单图最坏 75s（pro慢），典型 40-50s，
+               3图2并发最坏约75s，比原串行 lite→pro 240s 改善70%+
+           (5) 整个阶段统一关闭底层 httpx 重试（外层 max_retries=0，finally 恢复），
+               子线程只读不改 client 属性避免竞态
+           (6) 每张图 VLM 调用结束打印 elapsed 耗时日志便于排查
+    """
     try:
         from packages.shared.ai_service import call_vision  # noqa: F401
     except ImportError:
@@ -551,46 +871,64 @@ def _step_image_analysis(job: ViralVideoJob) -> dict:
         logger.warning("[爆款视频] 任务无 images，跳过图片分析")
         return {"products": []}
 
-    # 选择视觉模型：lite 速度优先（默认），pro 作为降级备用
+    # #2188 BUG1: URL 归一化 — storage_key→公网URL + 空值报400
+    normalized_urls: list[str] = []
+    for idx, raw in enumerate(job.images):
+        try:
+            normalized_urls.append(_normalize_image_url(raw, idx))
+        except ValueError as _ve:
+            # 空/非法URL：直接让任务失败，不默默走 fallback
+            logger.error("[爆款视频] 图片 #%d URL 归一化失败: %s", idx, _ve)
+            raise  # 上层 celery 捕获后标记任务失败，避免"未识别·无法判断"误导
+
+    # #2188/#2198 BUG2: 爆款视频强制 lite-first（不依赖 .env 开关），lite/pro 并行竞速
     try:
         _s = get_shared_settings()
-        if _s.doubao_vision_use_lite:
-            vision_model = _s.doubao_vision_lite_model
-            pro_model = _s.doubao_vision_model
-            vision_timeout = 45  # #2180: 方舟 VLM 实测服务端处理24-38s，原15s必超时3次重试全挂，提到45s
-        else:
-            vision_model = _s.doubao_vision_model
-            pro_model = None  # 已经是 pro，不再降级
-            vision_timeout = 60
+        lite_model = _s.doubao_vision_lite_model
+        pro_model = _s.doubao_vision_model
     except Exception:
-        vision_model = "doubao-1-5-vision-lite-250315"
-        pro_model = "doubao-1-5-vision-pro-250328"
-        vision_timeout = 45
+        lite_model = "doubao-seed-2-1-lite-260915"
+        pro_model = "doubao-seed-2-1-pro-260915"
+    vision_model = lite_model
+    # #2198: lite 单次 30s 封顶（竞速快速路径，30s 还没出就等 pro），pro 75s（在 _analyze_single_image
+    # 的内部竞速池里设置），外层不感知。单图最坏 75s（仅 pro 成功），典型 40-50s（pro 正常返回）。
+    vision_timeout = 30
 
-    results: list[dict] = [None] * len(job.images)  # type: ignore
-    max_workers = min(4, max(1, len(job.images)))
+    # #2194/#2198: 整个并行图片分析阶段统一把共享 client 的 max_retries 置 0，
+    # 阶段结束 finally 恢复。子线程 _call 只读不改，避免竞态。
+    from packages.shared.ai_client import get_doubao_client as _gdc_step
+
+    _step_client = _gdc_step()
+    _step_orig_retries = _step_client.max_retries
+    _step_client.max_retries = 0
+
+    results: list[dict] = [None] * len(normalized_urls)  # type: ignore
+    max_workers = min(2, max(1, len(normalized_urls)))  # 并发≤2 防方舟限流（竞速模式下总并发=4）
     logger.info(
-        "[爆款视频] 开始并行图片分析 n=%d model=%s pro_fallback=%s timeout=%d workers=%d",
-        len(job.images),
+        "[爆款视频] 开始并行竞速图片分析 n=%d lite=%s(%ds) pro=%s(75s) img_workers=%d",
+        len(normalized_urls),
         vision_model,
-        pro_model,
         vision_timeout,
+        pro_model,
         max_workers,
     )
-    with ThreadPoolExecutor(max_workers=max_workers) as pool:
-        future_to_idx = {
-            pool.submit(
-                _analyze_single_image, idx, url, vision_model, vision_timeout, pro_fallback_model=pro_model
-            ): idx
-            for idx, url in enumerate(job.images)
-        }
-        for fut in as_completed(future_to_idx):
-            idx = future_to_idx[fut]
-            try:
-                results[idx] = fut.result()
-            except Exception as e:
-                logger.warning("[爆款视频] 图片 #%d future 异常 err=%s", idx, e, exc_info=True)
-                results[idx] = _vision_fallback(idx, "future_exception", {"_error": str(e)[:200]})
+    try:
+        with ThreadPoolExecutor(max_workers=max_workers) as pool:
+            future_to_idx = {
+                pool.submit(
+                    _analyze_single_image, idx, url, vision_model, vision_timeout, pro_fallback_model=pro_model
+                ): idx
+                for idx, url in enumerate(normalized_urls)
+            }
+            for fut in as_completed(future_to_idx):
+                idx = future_to_idx[fut]
+                try:
+                    results[idx] = fut.result()
+                except Exception as e:
+                    logger.warning("[爆款视频] 图片 #%d future 异常 err=%s", idx, e, exc_info=True)
+                    results[idx] = _vision_fallback(idx, "future_exception", {"_error": str(e)[:200]})
+    finally:
+        _step_client.max_retries = _step_orig_retries
 
     return {"products": results}
 
@@ -1221,15 +1559,79 @@ def _step_review(job: ViralVideoJob, copy_result: dict) -> dict:
         return {"passed": True, "score": 75, "details": {}, "issues": []}
 
 
+def _resolve_tts_voice_id(job: ViralVideoJob) -> str:
+    """#2188: 根据 voice_source + voice_id 解析真正传给 CosyVoice 的 voice_id。
+
+    - voice_source 在 ("my_voice", "clone")：voice_id 是 VoiceCloneProfile.id，
+      需要从 DB 查 profile.voice_id（CosyVoice 返回的音色 ID）。
+    - 其他/空：voice_id 直接视为 CosyVoice preset 音色名（longxiaochun_v3 等）。
+    - 任何解析失败都回退默认 longxiaochun_v3，保证任务不崩。
+    """
+    default_voice = "longxiaochun_v3"
+    raw_voice_id = (getattr(job, "voice_id", "") or "").strip()
+    voice_source = (getattr(job, "voice_source", "") or "").strip().lower()
+
+    if not raw_voice_id:
+        return default_voice
+
+    # 克隆音色：前端传 profile.id，需查 DB 取 cosyvoice_voice_id
+    if voice_source in ("my_voice", "clone"):
+        session = None
+        try:
+            from packages.adapters.sqlalchemy_impl.voice_clone_profile_repository import (
+                SQLAlchemyVoiceCloneProfileRepository,
+            )
+
+            session = SessionLocal()
+            _repo = SQLAlchemyVoiceCloneProfileRepository(session)
+            _profile = _repo.get(raw_voice_id)
+            if _profile and _profile.status.value == "ready" and (_profile.voice_id or "").strip():
+                logger.info(
+                    "[爆款视频] 克隆音色解析: profile_id=%s → cosyvoice_voice_id=%s",
+                    raw_voice_id,
+                    _profile.voice_id,
+                )
+                return _profile.voice_id.strip()
+            # 找不到/未就绪/无voice_id
+            if _profile is None:
+                logger.warning("[爆款视频] 克隆音色 profile_id=%s 不存在，回退默认音色", raw_voice_id)
+            elif _profile.status.value != "ready":
+                logger.warning(
+                    "[爆款视频] 克隆音色 profile_id=%s 状态=%s 未就绪，回退默认音色",
+                    raw_voice_id,
+                    _profile.status.value,
+                )
+            else:
+                logger.warning("[爆款视频] 克隆音色 profile_id=%s 就绪但 voice_id 为空，回退默认音色", raw_voice_id)
+            return default_voice
+        except Exception as _e:
+            logger.warning("[爆款视频] 克隆音色解析异常 profile_id=%s err=%s，回退默认音色", raw_voice_id, _e)
+            return default_voice
+        finally:
+            if session is not None:
+                try:
+                    session.close()
+                except Exception:
+                    pass
+
+    # preset/my-voice 直传/空 source：voice_id 就是 CosyVoice 音色名
+    return raw_voice_id or default_voice
+
+
 def _step_tts(job: ViralVideoJob, voiceover_script: str):
-    """步骤 5: CosyVoice 整段配音 → 返回本地 MP3 Path；失败返回 None。"""
+    """步骤 5: CosyVoice 整段配音 → 返回本地 MP3 Path；失败返回 None。
+
+    #2188: 支持 voice_source='my_voice'/'clone'，先把前端传的 profile.id
+    解析成 CosyVoice 真正的克隆音色 voice_id，再走同一条 synthesize 链路。
+    """
     try:
         from pathlib import Path as _Path
 
         from apps.worker.services.tts_service_factory import get_tts_service
 
         tts_service = get_tts_service()
-        voice_id = (getattr(job, "voice_id", "") or "").strip()
+        # #2188: 根据 voice_source 解析实际传给 CosyVoice 的 voice_id
+        voice_id = _resolve_tts_voice_id(job)
         text = (voiceover_script or "").strip()
         if not text:
             logger.warning("[爆款视频] voiceover_script 为空，跳过 TTS")
@@ -1237,21 +1639,19 @@ def _step_tts(job: ViralVideoJob, voiceover_script: str):
         try:
             result = tts_service.synthesize(
                 text=text,
-                voice_id=voice_id or "longxiaochun_v3",
+                voice_id=voice_id,
                 format="mp3",
             )
         except TypeError:
             try:
-                result = tts_service.synthesize(text=text, voice_id=voice_id or "longxiaochun_v3")
+                result = tts_service.synthesize(text=text, voice_id=voice_id)
             except TypeError:
                 result = tts_service.synthesize(text=text)
         if result is None:
             return None
         p = _Path(result) if not isinstance(result, _Path) else result
         if p.exists() and p.stat().st_size > 0:
-            logger.info(
-                "[爆款视频] TTS 合成完成: voice=%s path=%s size=%d", voice_id or "longxiaochun_v3", p, p.stat().st_size
-            )
+            logger.info("[爆款视频] TTS 合成完成: voice=%s path=%s size=%d", voice_id, p, p.stat().st_size)
             return p
         logger.warning("[爆款视频] TTS 返回路径不存在或空文件: %s", p)
         return None

packages/application/viral_video/prompts.py

@@ -57,7 +57,18 @@ _IMAGE_ANALYSIS_SYSTEM = f"""你是电商商品视觉分析师，负责从商品
 <quality> 用一个标签，属性 resolution、lighting、composition、blur 描述画质。
 <key_selling_points> 下面每个卖点用一个 <point> 标签。
 
-看不到或无法判断的内容，属性值填“无法判断”，布尔值填 false，不要留空标签。"""
+【人物属性硬性要求（has_person=true时必须遵守）】
+hair/skin_tone/face_shape/outfit四项绝对禁止填“无法判断”，必须基于图片可见特征给出具体中文描述：
+- hair：必须描述发型+发色，如“黑色齐肩直发”“棕色微卷中长发”“深棕色短发”
+- skin_tone：必须描述肤色，如“暖调自然肤色”“白皙肤色”“小麦色”
+- face_shape：必须描述脸型，如“鹅蛋脸”“圆脸”“瓜子脸”“方脸”
+- outfit：必须描述可见穿着，如“米色翻领衬衫”“白色T恤”“黑色连衣裙”
+即使局部被遮挡也要根据可见部分合理推断；确实看不清时按最接近的直观印象描述。
+
+其他非人物属性看不到或无法判断时填“无法判断”，布尔值填false，不要留空标签。
+
+【有人物场景输出参考（女性手持商品示例，必须写全10个属性，禁止省略）】
+<people has_person="true" count="1" gender="女" age_range="青年" hair="黑色齐肩直发" skin_tone="暖调自然肤色" face_shape="鹅蛋脸" outfit="米色翻领衬衫" pose="正面半身，手持商品" expression="面带微笑"/>"""
 
 _IMAGE_ANALYSIS_USER = """请分析以下商品图片，共 {image_count} 张。
 所属行业：{industry}

packages/config/base.py

@@ -103,7 +103,7 @@ class SharedSettings(BaseSettings):
     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 = False  # #2181: lite视觉模型100%超时，默认关闭走pro（25-38s稳定返回）
+    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）