auto-save 2026-05-14 11:58 (~4)

.memory/worklog.json

@@ -1,26 +1,5 @@
 {
   "entries": [
-    {
-      "files_changed": 1,
-      "hash": "9fcc418",
-      "message": "auto-save 2026-05-13 04:17 (~1)",
-      "ts": "2026-05-13T04:17:55+08:00",
-      "type": "commit"
-    },
-    {
-      "files_changed": 1,
-      "hash": "f3a41e9",
-      "message": "auto-save 2026-05-13 04:23 (~1)",
-      "ts": "2026-05-13T04:23:48+08:00",
-      "type": "commit"
-    },
-    {
-      "files_changed": 1,
-      "hash": "0cfa960",
-      "message": "auto-save 2026-05-13 04:29 (~1)",
-      "ts": "2026-05-13T04:29:41+08:00",
-      "type": "commit"
-    },
     {
       "files_changed": 1,
       "hash": "c7af450",
@@ -3297,6 +3276,25 @@
       "type": "session-heartbeat",
       "message": "Codex 会话活跃 · 最近命令：codex · 2 项未提交变更 · 最近提交：auto-save 2026-05-14 11:47 (~7)",
       "files_changed": 2
+    },
+    {
+      "ts": "2026-05-14T11:53:14+08:00",
+      "type": "commit",
+      "message": "auto-save 2026-05-14 11:53 (~4)",
+      "hash": "801b194",
+      "files_changed": 4
+    },
+    {
+      "ts": "2026-05-14T03:56:10Z",
+      "type": "session-heartbeat",
+      "message": "Codex 会话活跃 · 最近命令：codex · 3 项未提交变更 · 最近提交：auto-save 2026-05-14 11:53 (~4)",
+      "files_changed": 3
+    },
+    {
+      "ts": "2026-05-14T03:58:39Z",
+      "type": "session-heartbeat",
+      "message": "Codex 会话活跃 · 最近命令：codex · 4 项未提交变更 · 最近提交：auto-save 2026-05-14 11:53 (~4)",
+      "files_changed": 4
     }
   ]
 }

api/main.py

@@ -1477,13 +1477,16 @@ def _translate_sync(segments: list[dict]) -> list[str]:
         '[{"i": 0, "zh": "..."}, ...]\n\n输入：\n'
         + json.dumps(payload, ensure_ascii=False)
     )
-    resp = llm().chat.completions.create(
-        model=TRANSLATE_MODEL,
-        messages=[{"role": "user", "content": prompt}],
-        response_format={"type": "json_object"},
-        temperature=0.2,
-    )
-    content = resp.choices[0].message.content or "[]"
+    try:
+        resp = llm().chat.completions.create(
+            model=TRANSLATE_MODEL,
+            messages=[{"role": "user", "content": prompt}],
+            response_format={"type": "json_object"},
+            temperature=0.2,
+        )
+        content = resp.choices[0].message.content or "[]"
+    except Exception:
+        return ["" for _ in segments]
     try:
         data = json.loads(content)
         if isinstance(data, dict):

web/app/page.tsx

@@ -19,7 +19,7 @@ import { ThemeToggle } from "@/components/theme-toggle"
 import { AudioStrip } from "@/components/audio-strip"
 import {
   addManualFrame, analyzeJob, createJob, getJob, listJobs, uploadJob, deleteJob, deleteFrame, deleteGeneratedImage,
-  deleteGeneratedVideo, deleteCutout, generateStoryboardVideo, createProductFusionGuide, triggerTranscribe,
+  deleteGeneratedVideo, deleteCutout, generateStoryboardVideo, triggerTranscribe,
   type Job, type ImageRef, type ProductFusionShot, type StoryboardScene, type FrameExtractMode, type FrameExtractQuality, type FrameExtractTarget,
 } from "@/lib/api"
 import { TRANSPARENT_HUMAN_NEGATIVE_PROMPT, TRANSPARENT_HUMAN_VIDEO_PROMPT } from "@/lib/workflow-target"
@@ -520,46 +520,43 @@ export default function Home() {
     if (!job) return
     const frame = job.frames.find((f) => f.index === frameIdx)
     if (!frame) return
-    if (!shot.product_image || !shot.person_image || !shot.scene_image || !shot.product_region || !shot.action_text?.trim()) {
-      toast.error("产品融合镜头缺少产品图、人物图、区域、场景图或描述词")
+    const productRefs = (shot.product_images ?? []).filter(Boolean).slice(0, 3) as ImageRef[]
+    if (!shot.first_image || !shot.last_image || productRefs.length < 3 || !shot.action_text?.trim()) {
+      toast.error("产品融合镜头缺少首帧、尾帧、三张产品角度图或描述词")
       return
     }
     const duration = shot.duration && shot.duration > 0 ? shot.duration : 5
     const labelOf = (ref?: ImageRef | null, fallback = "未提供") => ref?.label || fallback
     try {
-      toast.info(`生成融合引导图 · GPT Image 2 位置约束 · 镜头 ${shot.id || ""}`)
-      const guideRef = await createProductFusionGuide(job.id, {
-        ...shot,
-        image_model: "gpt-image-2",
-        video_model: "seedance",
-      })
-      const region = shot.product_region
       const prompt = [
-        `竖屏 9:16，${duration.toFixed(1)} 秒，Seedance 产品融合视频。`,
-        "图片模型固定为 GPT Image 2：已根据白底人物图和手动画框生成产品融合引导图；引导图是产品尺寸、位置、贴合关系和起始构图的最高优先级参考。",
-        "视频模型固定为 Seedance：生成单镜头连续视频，不跳切，不换主体，不改变产品身份。",
-        `产品区域：x=${region.x.toFixed(3)}, y=${region.y.toFixed(3)}, w=${region.w.toFixed(3)}, h=${region.h.toFixed(3)}。产品只能在这个框对应的人物/身体/手部区域内融合，不能漂移到其他位置，也不能明显超出框架。`,
-        `产品图：${labelOf(shot.product_image, "SKG 白底产品图")}。严格保持 SKG 产品外观、颜色、材质、U 形结构、按摩触点、按键和比例。`,
-        `白底人物图：${labelOf(shot.person_image, "人物姿态参考")}。人物姿态、手部接触点和产品佩戴关系以这张图为准。`,
-        `场景图：${labelOf(shot.scene_image, "场景参考")}。背景、空间、光线和气氛以这张图为准，但不要改变产品框内位置。`,
+        `竖屏 9:16，${duration.toFixed(1)} 秒，Seedance 图生视频。`,
+        "图片模型固定为 GPT Image 2：首帧和尾帧已经由文字生图生成，用来锁定透明骨架人角色、场景构图和动作起止状态。",
+        "视频模型固定为 Seedance：使用首帧作为起始画面、尾帧作为结束画面，并用三张同一 SKG 产品不同角度白底图作为垫图/产品身份参考。",
+        `首帧：${labelOf(shot.first_image, "透明骨架人首帧")}。起始人物形象、姿态、构图和场景氛围以这张图为准。`,
+        `尾帧：${labelOf(shot.last_image, "透明骨架人尾帧")}。结束人物状态、画面落点和场景延续以这张图为准。`,
+        `产品角度图 1：${labelOf(productRefs[0], "SKG 产品正面/主视角")}。`,
+        `产品角度图 2：${labelOf(productRefs[1], "SKG 产品侧面/斜侧视角")}。`,
+        `产品角度图 3：${labelOf(productRefs[2], "SKG 产品背面/细节视角")}。`,
         `动作描述：${shot.action_text.trim()}`,
         TRANSPARENT_HUMAN_VIDEO_PROMPT,
-        "融合要求：产品必须按引导图位置自然贴合人物或手部，尺寸可信，透视一致，边缘清晰，不能悬浮、穿帮、融化、扭曲或变成其他物体。",
-        "场景要求：把白底人物姿态自然放入场景图的环境中，光线方向和阴影要统一，背景不要出现水印、平台 UI、字幕或竞品包装。",
+        "融合要求：产品必须自然出现在透明骨架人动作中，尺寸可信，透视一致，贴合身体/手部/使用区域，不能悬浮、漂移、融化、扭曲或变成其他物体。",
+        "首尾连续性：镜头从首帧自然运动到尾帧，中间不要跳切，不换角色，不换产品，不突然改变场景。",
+        "产品一致性：严格保持 SKG 产品外观、颜色、材质、U 形结构、按摩触点、按键和比例；三张产品角度图是产品身份真源。",
+        "场景要求：背景、空间、光线和阴影要自然统一，不要出现水印、平台 UI、字幕或竞品包装。",
         "商业质感：真实拍摄感、干净高级、产品清楚可辨、人物动作自然、镜头稳定。",
-        "禁止：文字、水印、随机品牌、非 SKG 产品、医学治疗承诺、夸张病症、恐怖元素、产品位置漂移、产品超过指定融合区域。",
+        "禁止：文字、水印、随机品牌、非 SKG 产品、医学治疗承诺、夸张病症、恐怖元素、产品位置漂移、透明衣服但非透明身体。",
         TRANSPARENT_HUMAN_NEGATIVE_PROMPT,
       ].join("\n")
       const updated = await generateStoryboardVideo(job.id, frameIdx, {
         prompt,
         duration,
-        first_image: guideRef,
-        last_image: null,
-        product_images: [shot.product_image, shot.person_image, shot.scene_image].filter(Boolean) as ImageRef[],
-        subject_image: shot.person_image,
-        scene_image: shot.scene_image,
-        product_image: shot.product_image,
-        action_image: guideRef,
+        first_image: shot.first_image,
+        last_image: shot.last_image,
+        product_images: productRefs,
+        subject_image: shot.first_image,
+        scene_image: shot.last_image,
+        product_image: productRefs[0] ?? null,
+        action_image: null,
         source_ref: null,
         model: "seedance",
         size: "720x1280",

web/components/lightbox.tsx

@@ -6,7 +6,7 @@ import {
   frameUrl, cleanedFrameUrl, apiAssetUrl,
   describeFrame, cleanupFrame, applyCleanedFrame, discardCleanedFrame, addElement, updateElement, deleteElement,
   generateSceneAsset, generateSubjectAssets, generateProductFusionDescriptions, resolveImageRefUrl, uploadStoryboardAsset, updateStoryboard,
-  type AssetBackground, type AssetSize, type KeyFrame, type Job, type ImageRef, type ProductFusionShot, type SceneMode, type SceneStyle, type SubjectKind,
+  type AssetBackground, type AssetSize, type KeyFrame, type Job, type ImageRef, type ProductFusionShot, type SceneAssetRole, type SceneMode, type SceneStyle, type SubjectKind,
 } from "@/lib/api"
 import { ProductLibraryPicker } from "@/components/product-library-picker"
 import { TRANSPARENT_HUMAN_FRAME_STANDARD, TRANSPARENT_HUMAN_UI_SUMMARY } from "@/lib/workflow-target"
@@ -72,7 +72,7 @@ type LightboxTab = "clean" | "scene" | "subject" | "product" | "review"
 const LIGHTBOX_TABS: Array<{ key: LightboxTab; label: string }> = [
   { key: "clean", label: "原图/清洗" },
   { key: "subject", label: "主体资产" },
-  { key: "scene", label: "场景图" },
+  { key: "scene", label: "首尾帧" },
   { key: "product", label: "产品融合" },
   { key: "review", label: "审核" },
 ]
@@ -112,10 +112,20 @@ const SCENE_REFERENCE_OPTIONS = [
 
 const FUSION_SHOT_COUNT = 6
 const FUSION_DURATIONS = [4, 5, 6, 8, 10, 12, 15]
+type FusionUploadTarget = {
+  shotIndex: number
+  slot: "first_image" | "last_image" | "product_images"
+  productIndex?: number
+}
+type FusionFrameRole = "first_image" | "last_image"
+const PRODUCT_ANGLE_LABELS = ["产品角度 1", "产品角度 2", "产品角度 3"]
 
 const createFusionShots = (): ProductFusionShot[] =>
   Array.from({ length: FUSION_SHOT_COUNT }, (_, i) => ({
     id: `shot-${i + 1}`,
+    first_image: null,
+    last_image: null,
+    product_images: [],
     product_image: null,
     person_image: null,
     product_region: null,
@@ -130,7 +140,15 @@ const createFusionShots = (): ProductFusionShot[] =>
 const normalizeFusionShots = (shots?: ProductFusionShot[] | null): ProductFusionShot[] => {
   const base = createFusionShots()
   if (!shots?.length) return base
-  return base.map((item, i) => ({ ...item, ...(shots[i] ?? {}), id: shots[i]?.id || item.id }))
+  return base.map((item, i) => {
+    const shot = shots[i] ?? {}
+    const productImages = shot.product_images?.length
+      ? shot.product_images.slice(0, 3)
+      : shot.product_image
+        ? [shot.product_image]
+        : []
+    return { ...item, ...shot, product_images: productImages, id: shot.id || item.id }
+  })
 }
 
 export function FrameLightbox({ jobId, frames, activeIndex, selected, onClose, onChange, onToggleSelect, onJobUpdate, onSwitchPanel, clipboard, onCopyImage, onGenerateProductFusionVideo, embedded = false }: Props) {
@@ -141,7 +159,7 @@ export function FrameLightbox({ jobId, frames, activeIndex, selected, onClose, o
   const [batchApplying, setBatchApplying] = useState(false)
   const [batchApplyProgress, setBatchApplyProgress] = useState<{ done: number; total: number; failed: number } | null>(null)
   const [applying, setApplying] = useState(false)
-  const [sceneGenerating, setSceneGenerating] = useState(false)
+  const [sceneGenerating, setSceneGenerating] = useState<SceneAssetRole | null>(null)
   const [subjectGenerating, setSubjectGenerating] = useState<string | null>(null)
   const [assetSize, setAssetSize] = useState<AssetSize>("source")
   const [sceneMode, setSceneMode] = useState<SceneMode>("remove_subject")
@@ -156,7 +174,7 @@ export function FrameLightbox({ jobId, frames, activeIndex, selected, onClose, o
   const [activeTab, setActiveTab] = useState<LightboxTab>("clean")
   const [fusionShots, setFusionShots] = useState<ProductFusionShot[]>(() => createFusionShots())
   const [activeFusionShot, setActiveFusionShot] = useState(0)
-  const [fusionUploadTarget, setFusionUploadTarget] = useState<"product_image" | "person_image" | "scene_image" | null>(null)
+  const [fusionUploadTarget, setFusionUploadTarget] = useState<FusionUploadTarget | null>(null)
   const [fusionGenerating, setFusionGenerating] = useState<number | "all" | null>(null)
   const [fusionSaving, setFusionSaving] = useState(false)
   const [fusionDraftRegion, setFusionDraftRegion] = useState<{ x: number; y: number; w: number; h: number } | null>(null)
@@ -286,24 +304,29 @@ export function FrameLightbox({ jobId, frames, activeIndex, selected, onClose, o
   const unifiedSubjectName = subjectElementRefs[0]?.element.name_zh || "统一主体"
   const sceneLocationLabel = SCENE_LOCATION_OPTIONS.find(([value]) => value === sceneLocation)?.[1] ?? sceneLocation
   const sceneStyleLabel = SCENE_STYLE_OPTIONS.find(([value]) => value === sceneStyle)?.[1] ?? sceneStyle
-  const sceneModeLabel = SCENE_MODE_OPTIONS.find(([value]) => value === sceneMode)?.[1] ?? sceneMode
   const sceneReferenceLabels = sceneReferenceKeys
     .map((key) => SCENE_REFERENCE_OPTIONS.find(([value]) => value === key)?.[1] ?? key)
   const scenePromptDraft = [
-    `主体：移除 ${unifiedSubjectName} 后生成空场景。`,
+    `目标：为透明骨架人视频生成首帧或尾帧，不再生成空背景板。`,
+    `人物：保持 ${unifiedSubjectName} 的透明/半透明外壳、干净白色骨架、非恐怖广告角色气质。`,
     `地点：${sceneLocationLabel}。`,
-    `生成方式：${sceneModeLabel}。`,
     `风格：${sceneStyleLabel}。`,
     `参考帧：${sceneReferenceFrames.map((frame) => `分镜${frame.index + 1}`).join("、") || `分镜${f.index + 1}`}。`,
     sceneReferenceLabels.length > 0 ? `保留参考：${sceneReferenceLabels.join("、")}。` : "",
     sceneExtraKeywords.trim() ? `额外关键词：${sceneExtraKeywords.trim()}。` : "",
-    "要求：无主体、无人物动物产品、无文字水印，保持可用于后续视频生成的干净背景板。",
+    "要求：单一透明骨架人清晰可见，人物占画面主体，首尾帧可连续生成视频；无文字、水印、平台 UI、恐怖解剖感。",
   ].filter(Boolean).join("\n")
   const currentFusionShot = fusionShots[activeFusionShot] ?? fusionShots[0]
-  const currentFusionProductUrl = currentFusionShot?.product_image ? resolveImageRefUrl(jobId, currentFusionShot.product_image) : ""
+  const currentFusionProducts = currentFusionShot?.product_images ?? []
+  const currentFusionProductCount = currentFusionProducts.filter(Boolean).length
+  const currentFusionFirstUrl = currentFusionShot?.first_image ? resolveImageRefUrl(jobId, currentFusionShot.first_image) : ""
+  const currentFusionLastUrl = currentFusionShot?.last_image ? resolveImageRefUrl(jobId, currentFusionShot.last_image) : ""
+  const currentFusionProductUrl = currentFusionProducts[0] ? resolveImageRefUrl(jobId, currentFusionProducts[0]) : ""
   const currentFusionPersonUrl = currentFusionShot?.person_image ? resolveImageRefUrl(jobId, currentFusionShot.person_image) : ""
   const currentFusionSceneUrl = currentFusionShot?.scene_image ? resolveImageRefUrl(jobId, currentFusionShot.scene_image) : ""
-  const fusionReadyCount = fusionShots.filter((shot) => shot.product_image && shot.person_image && shot.product_region && shot.scene_image && shot.action_text?.trim()).length
+  const fusionReadyCount = fusionShots.filter((shot) =>
+    shot.first_image && shot.last_image && (shot.product_images ?? []).filter(Boolean).length >= 3 && shot.action_text?.trim()
+  ).length
 
   const persistFusionShots = async (nextShots: ProductFusionShot[]) => {
     setFusionSaving(true)
@@ -326,8 +349,19 @@ export function FrameLightbox({ jobId, frames, activeIndex, selected, onClose, o
     if (persist) void persistFusionShots(next)
   }
 
-  const assignFusionImage = (slot: "product_image" | "person_image" | "scene_image", ref: ImageRef, index = activeFusionShot) => {
-    updateFusionShot(index, { [slot]: ref, guide_image: null }, true)
+  const assignFusionImage = (target: FusionUploadTarget, ref: ImageRef) => {
+    const index = target.shotIndex
+    const current = fusionShots[index]
+    if (!current) return
+    if (target.slot === "product_images") {
+      const productImages = [...(current.product_images ?? [])].slice(0, 3)
+      const productIndex = Math.max(0, Math.min(2, target.productIndex ?? productImages.findIndex((item) => !item)))
+      const safeIndex = productIndex >= 0 ? productIndex : 0
+      productImages[safeIndex] = ref
+      updateFusionShot(index, { product_images: productImages, product_image: productImages[0] ?? null, guide_image: null }, true)
+      return
+    }
+    updateFusionShot(index, { [target.slot]: ref, guide_image: null }, true)
   }
 
   const uploadFusionFiles = async (files: FileList | File[]) => {
@@ -348,9 +382,9 @@ export function FrameLightbox({ jobId, frames, activeIndex, selected, onClose, o
     }
   }
 
-  const openFusionUpload = (slot: "product_image" | "person_image" | "scene_image", index = activeFusionShot) => {
-    setActiveFusionShot(index)
-    setFusionUploadTarget(slot)
+  const openFusionUpload = (target: FusionUploadTarget) => {
+    setActiveFusionShot(target.shotIndex)
+    setFusionUploadTarget(target)
     requestAnimationFrame(() => fusionFileInputRef.current?.click())
   }
 
@@ -419,8 +453,8 @@ export function FrameLightbox({ jobId, frames, activeIndex, selected, onClose, o
 
   const runFusionVideo = async (index: number) => {
     const shot = fusionShots[index]
-    if (!shot?.product_image || !shot.person_image || !shot.scene_image || !shot.product_region || !shot.action_text?.trim()) {
-      toast.error(`镜头 ${index + 1} 还缺产品图、人物图、区域、场景图或描述词`)
+    if (!shot?.first_image || !shot.last_image || (shot.product_images ?? []).filter(Boolean).length < 3 || !shot.action_text?.trim()) {
+      toast.error(`镜头 ${index + 1} 还缺首帧、尾帧、三张产品角度图或描述词`)
       return
     }
     setFusionGenerating(index)
@@ -434,7 +468,7 @@ export function FrameLightbox({ jobId, frames, activeIndex, selected, onClose, o
   const runAllFusionVideos = async () => {
     const indexes = fusionShots
       .map((shot, i) => ({ shot, i }))
-      .filter(({ shot }) => shot.product_image && shot.person_image && shot.scene_image && shot.product_region && shot.action_text?.trim())
+      .filter(({ shot }) => shot.first_image && shot.last_image && (shot.product_images ?? []).filter(Boolean).length >= 3 && shot.action_text?.trim())
       .map(({ i }) => i)
     if (indexes.length === 0) {
       toast.error("还没有完整的融合镜头")
@@ -548,27 +582,47 @@ export function FrameLightbox({ jobId, frames, activeIndex, selected, onClose, o
     }
   }
 
-  const handleGenerateSceneAsset = async () => {
+  const handleGenerateSceneAsset = async (role: Exclude<SceneAssetRole, "scene">) => {
+    const roleLabel = role === "first_frame" ? "首帧" : "尾帧"
+    const targetSlot: FusionFrameRole = role === "first_frame" ? "first_image" : "last_image"
     if (!hasSubjectAssets) {
-      setActiveTab("subject")
-      toast.message("先生成主体资产，再生成去主体场景图")
-      return
+      toast.message("还没有主体资产，也会按当前参考帧理解人物；一致性可能弱一些")
     }
-    setSceneGenerating(true)
+    setSceneGenerating(role)
     try {
       const updated = await generateSceneAsset(jobId, f.index, {
         size: assetSize,
-        scene_mode: sceneMode,
+        scene_mode: "similar",
         scene_style: sceneStyle,
-        prompt: scenePrompt.trim() || scenePromptDraft,
+        asset_role: role,
+        prompt: [
+          role === "first_frame"
+            ? "生成这个产品融合镜头的首帧：人物处于动作开始状态，构图稳定，适合作为视频第一帧。"
+            : "生成这个产品融合镜头的尾帧：人物处于动作完成状态，与首帧连续但画面不要完全相同。",
+          scenePrompt.trim() || scenePromptDraft,
+        ].join("\n"),
         source_frame_indices: sceneReferenceFrameIndices,
       })
       onJobUpdate?.(updated)
-      toast.success(`分镜 ${f.index + 1} 场景图已生成`)
+      const updatedFrame = updated.frames.find((frame) => frame.index === f.index)
+      const asset = [...(updatedFrame?.scene_assets ?? [])].reverse().find((item) => item.asset_role === role)
+      if (asset) {
+        assignFusionImage({
+          shotIndex: activeFusionShot,
+          slot: targetSlot,
+        }, {
+          kind: "asset",
+          frame_idx: f.index,
+          element_id: asset.id,
+          cutout_id: asset.id,
+          label: asset.label,
+        })
+      }
+      toast.success(`分镜 ${f.index + 1} ${roleLabel}已生成，并填入镜头 ${activeFusionShot + 1}`)
     } catch (e) {
-      toast.error("场景图生成失败：" + (e instanceof Error ? e.message : String(e)))
+      toast.error(`${roleLabel}生成失败：` + (e instanceof Error ? e.message : String(e)))
     } finally {
-      setSceneGenerating(false)
+      setSceneGenerating(null)
     }
   }
 
@@ -875,7 +929,7 @@ export function FrameLightbox({ jobId, frames, activeIndex, selected, onClose, o
           ) : isSceneTab ? (
             <section className="rounded-lg border border-emerald-300/15 bg-emerald-500/[0.06] p-2.5">
               <div className="mb-2 flex items-center justify-between gap-2">
-                <div className="text-[12px] font-semibold text-white">场景参考图</div>
+                <div className="text-[12px] font-semibold text-white">首尾帧参考图</div>
                 <span className="text-[9.5px] font-mono text-white/38">
                   {selectedFrameIndices.length > 0 ? `${selectedFrameIndices.length} 已选参考` : "默认当前帧"}
                 </span>
@@ -939,7 +993,7 @@ export function FrameLightbox({ jobId, frames, activeIndex, selected, onClose, o
                 })}
               </div>
               <div className="mt-2 text-[10px] leading-relaxed text-white/38">
-                左侧显示全部关键帧；点图片设为生成目标，点“选”加入场景参考。未选择时默认只参考当前目标帧。
+                左侧显示全部关键帧；点图片设为生成目标，点“选”加入人物/机位参考。未选择时默认只参考当前目标帧。
               </div>
             </section>
           ) : isProductTab ? (
@@ -967,75 +1021,90 @@ export function FrameLightbox({ jobId, frames, activeIndex, selected, onClose, o
                 </span>
               </div>
               <div className="mb-2 rounded-md border border-white/10 bg-black/25 px-2 py-1.5 text-[10px] leading-relaxed text-white/50">
-                6 条视频镜头从上到下排列；每条都能直接看到产品图、白底人物、融合区域、场景、描述和秒数。
+                6 条视频镜头从上到下排列；每条使用文字描述 + 首帧 + 尾帧 + 同一产品 3 个角度图，作为 Seedance 垫图生成视频。
               </div>
               <div className="space-y-2">
                 {fusionShots.map((shot, i) => {
                   const active = i === activeFusionShot
-                  const productUrl = shot.product_image ? resolveImageRefUrl(jobId, shot.product_image) : ""
-                  const personUrl = shot.person_image ? resolveImageRefUrl(jobId, shot.person_image) : ""
-                  const sceneUrl = shot.scene_image ? resolveImageRefUrl(jobId, shot.scene_image) : ""
-                  const ready = !!(shot.product_image && shot.person_image && shot.product_region && shot.scene_image && shot.action_text?.trim())
+                  const productImages = shot.product_images ?? []
+                  const firstUrl = shot.first_image ? resolveImageRefUrl(jobId, shot.first_image) : ""
+                  const lastUrl = shot.last_image ? resolveImageRefUrl(jobId, shot.last_image) : ""
+                  const productUrls = [0, 1, 2].map((productIndex) =>
+                    productImages[productIndex] ? resolveImageRefUrl(jobId, productImages[productIndex]) : ""
+                  )
+                  const ready = !!(shot.first_image && shot.last_image && productImages.filter(Boolean).length >= 3 && shot.action_text?.trim())
                   const busy = fusionGenerating === i || fusionGenerating === "all"
-                  const pasteIntoSlot = (slot: "product_image" | "person_image" | "scene_image", label: string) => {
+                  const pasteIntoSlot = (target: FusionUploadTarget, label: string) => {
                     setActiveFusionShot(i)
                     if (clipboard) {
-                      assignFusionImage(slot, clipboard, i)
+                      assignFusionImage(target, clipboard)
                       toast.success(`已粘贴到镜头 ${i + 1}「${label}」：${clipboard.label || "剪贴板图片"}`)
                       return
                     }
-                    setFusionUploadTarget(slot)
+                    setFusionUploadTarget(target)
                     toast.message(`镜头 ${i + 1} 已选中「${label}」槽位，现在可 Cmd+V 粘贴系统图片`)
                   }
-                  const imageSlot = (slot: "product_image" | "scene_image", label: string, url: string) => {
-                    const ref = shot[slot]
-                    return (
-                      <div className="overflow-hidden rounded-md border border-white/10 bg-black/24">
-                        <div className="relative aspect-square bg-white">
-                          {url ? (
-                            <button
-                              type="button"
-                              onClick={() => setActiveFusionShot(i)}
-                              className="absolute inset-0 cursor-pointer"
-                              title={`选中镜头 ${i + 1}`}
-                            >
-                              <img src={url} alt={label} className="h-full w-full object-contain" draggable={false} />
-                            </button>
-                          ) : (
-                            <button
-                              type="button"
-                              onClick={() => openFusionUpload(slot, i)}
-                              className="absolute inset-0 flex flex-col items-center justify-center gap-1 text-[9.5px] text-black/35 hover:text-black/65"
-                            >
-                              <Upload className="h-3.5 w-3.5" />
-                              上传
-                            </button>
-                          )}
-                        </div>
-                        <div className="border-t border-white/10 px-1 py-1">
-                          <div className="mb-1 truncate text-[8.5px] text-white/42">{ref?.label || label}</div>
-                          <div className="grid grid-cols-2 gap-1">
-                            <button
-                              type="button"
-                              onClick={() => pasteIntoSlot(slot, label)}
-                              className={`rounded px-1 py-0.5 text-[8.5px] transition ${
-                                clipboard ? "bg-violet-500/60 text-white hover:bg-violet-400/70" : "bg-white/10 text-white/58 hover:bg-white/18 hover:text-white"
-                              }`}
-                            >
-                              粘贴
-                            </button>
-                            <button
-                              type="button"
-                              onClick={() => openFusionUpload(slot, i)}
-                              className="rounded bg-white/10 px-1 py-0.5 text-[8.5px] text-white/65 transition hover:bg-white/18 hover:text-white"
-                            >
-                              上传
-                            </button>
-                          </div>
+                  const imageSlot = (target: FusionUploadTarget, label: string, url: string, ref?: ImageRef | null, white = false) => (
+                    <div className="overflow-hidden rounded-md border border-white/10 bg-black/24">
+                      <div className={`relative aspect-[4/5] ${white ? "bg-white" : "bg-black"}`}>
+                        {url ? (
+                          <button
+                            type="button"
+                            onClick={() => setActiveFusionShot(i)}
+                            className="absolute inset-0 cursor-pointer"
+                            title={`选中镜头 ${i + 1}`}
+                          >
+                            <img src={url} alt={label} className="h-full w-full object-contain" draggable={false} />
+                          </button>
+                        ) : (
+                          <button
+                            type="button"
+                            onClick={() => openFusionUpload(target)}
+                            className={`absolute inset-0 flex flex-col items-center justify-center gap-1 text-[9.5px] ${white ? "text-black/35 hover:text-black/65" : "text-white/35 hover:text-white/65"}`}
+                          >
+                            <Upload className="h-3.5 w-3.5" />
+                            {label}
+                          </button>
+                        )}
+                      </div>
+                      <div className="border-t border-white/10 px-1 py-1">
+                        <div className="mb-1 truncate text-[8.5px] text-white/42">{ref?.label || label}</div>
+                        <div className="grid grid-cols-2 gap-1">
+                          <button
+                            type="button"
+                            onClick={() => pasteIntoSlot(target, label)}
+                            className={`rounded px-1 py-0.5 text-[8.5px] transition ${
+                              clipboard ? "bg-violet-500/60 text-white hover:bg-violet-400/70" : "bg-white/10 text-white/58 hover:bg-white/18 hover:text-white"
+                            }`}
+                          >
+                            粘贴
+                          </button>
+                          <button
+                            type="button"
+                            onClick={() => openFusionUpload(target)}
+                            className="rounded bg-white/10 px-1 py-0.5 text-[8.5px] text-white/65 transition hover:bg-white/18 hover:text-white"
+                          >
+                            上传
+                          </button>
                         </div>
                       </div>
-                    )
-                  }
+                    </div>
+                  )
+                  const productAngleSlots = (
+                    <div className="grid grid-cols-3 gap-1.5">
+                      {[0, 1, 2].map((productIndex) => (
+                        <div key={productIndex}>
+                          {imageSlot(
+                            { shotIndex: i, slot: "product_images", productIndex },
+                            PRODUCT_ANGLE_LABELS[productIndex],
+                            productUrls[productIndex],
+                            productImages[productIndex],
+                            true,
+                          )}
+                        </div>
+                      ))}
+                    </div>
+                  )
                   return (
                     <div
                       key={shot.id}
@@ -1045,7 +1114,7 @@ export function FrameLightbox({ jobId, frames, activeIndex, selected, onClose, o
                           : "border-white/10 bg-black/20 hover:border-amber-300/35"
                       }`}
                     >
-                      <div className="grid grid-cols-[34px_82px_112px_82px_minmax(150px,1fr)_78px] items-start gap-2">
+                      <div className="grid grid-cols-[34px_86px_86px_240px_minmax(150px,1fr)_78px] items-start gap-2">
                         <div className="flex flex-col items-center gap-1 pt-1">
                           <button
                             type="button"
@@ -1064,82 +1133,20 @@ export function FrameLightbox({ jobId, frames, activeIndex, selected, onClose, o
                           </span>
                         </div>
 
-                        {imageSlot("product_image", "产品图", productUrl)}
+                        {imageSlot({ shotIndex: i, slot: "first_image" }, "首帧", firstUrl, shot.first_image)}
 
-                        <div className="overflow-hidden rounded-md border border-white/10 bg-black/24">
-                          <div
-                            ref={active ? fusionPersonWrapRef : undefined}
-                            onMouseDown={personUrl ? (ev) => {
-                              setActiveFusionShot(i)
-                              if (active) onFusionRegionDown(ev)
-                            } : undefined}
-                            onMouseMove={active ? onFusionRegionMove : undefined}
-                            onMouseUp={active ? onFusionRegionUp : undefined}
-                            onMouseLeave={active ? onFusionRegionUp : undefined}
-                            className={`relative aspect-[4/5] bg-white ${
-                              personUrl ? (active ? "cursor-crosshair" : "cursor-pointer") : ""
-                            }`}
-                            title={personUrl ? (active ? "拖动画出产品融合区域" : "点击后编辑此镜头区域") : "上传白底人物图"}
-                          >
-                            {personUrl ? (
-                              <>
-                                <img src={personUrl} alt="白底人物图" className="h-full w-full select-none object-contain" draggable={false} />
-                                {!active && (
-                                  <div className="absolute inset-x-1 bottom-1 rounded bg-black/55 px-1 py-0.5 text-center text-[8px] text-white/75">
-                                    点此编辑区域
-                                  </div>
-                                )}
-                              </>
-                            ) : (
-                              <button
-                                type="button"
-                                onClick={() => openFusionUpload("person_image", i)}
-                                className="absolute inset-0 flex flex-col items-center justify-center gap-1 text-[9.5px] text-black/35 hover:text-black/65"
-                              >
-                                <Upload className="h-3.5 w-3.5" />
-                                白底人物
-                              </button>
-                            )}
-                            {[shot.product_region, active ? fusionDraftRegion : null].filter(Boolean).map((region, regionIdx) => region && (
-                              <div
-                                key={regionIdx}
-                                className={`absolute pointer-events-none border-2 ${regionIdx === 0 ? "border-amber-300 bg-amber-300/10" : "border-dashed border-cyan-300"}`}
-                                style={{
-                                  left: `${region.x * 100}%`,
-                                  top: `${region.y * 100}%`,
-                                  width: `${region.w * 100}%`,
-                                  height: `${region.h * 100}%`,
-                                }}
-                              />
-                            ))}
-                          </div>
-                          <div className="border-t border-white/10 px-1 py-1">
-                            <div className="mb-1 truncate text-[8.5px] text-white/42">
-                              {shot.person_image?.label || (shot.product_region ? "人物图 · 已画区域" : "白底人物图")}
-                            </div>
-                            <div className="grid grid-cols-2 gap-1">
-                              <button
-                                type="button"
-                                onClick={() => pasteIntoSlot("person_image", "白底人物图")}
-                                className={`rounded px-1 py-0.5 text-[8.5px] transition ${
-                                  clipboard ? "bg-violet-500/60 text-white hover:bg-violet-400/70" : "bg-white/10 text-white/58 hover:bg-white/18 hover:text-white"
-                                }`}
-                              >
-                                粘贴
-                              </button>
-                              <button
-                                type="button"
-                                onClick={() => openFusionUpload("person_image", i)}
-                                className="rounded bg-white/10 px-1 py-0.5 text-[8.5px] text-white/65 transition hover:bg-white/18 hover:text-white"
-                              >
-                                上传
-                              </button>
-                            </div>
+                        {imageSlot({ shotIndex: i, slot: "last_image" }, "尾帧", lastUrl, shot.last_image)}
+
+                        <div>
+                          <div className="mb-1 flex items-center justify-between gap-2">
+                            <span className="text-[9px] text-white/38">同一产品 · 三个角度</span>
+                            <span className={`text-[8.5px] ${productImages.filter(Boolean).length >= 3 ? "text-emerald-200/70" : "text-white/30"}`}>
+                              {productImages.filter(Boolean).length}/3
+                            </span>
                           </div>
+                          {productAngleSlots}
                         </div>
 
-                        {imageSlot("scene_image", "场景图", sceneUrl)}
-
                         <label className="block">
                           <div className="mb-1 flex items-center justify-between gap-2">
                             <span className="text-[9px] text-white/38">描述词 · 人在干什么</span>
@@ -1153,7 +1160,7 @@ export function FrameLightbox({ jobId, frames, activeIndex, selected, onClose, o
                               const next = fusionShots.map((item, idx) => (idx === i ? { ...item, action_text: e.currentTarget.value } : item))
                               void persistFusionShots(next)
                             }}
-                            placeholder="描述这个镜头里人物和产品的动作、位置、节奏。"
+                            placeholder="描述这个镜头里透明骨架人、SKG 产品和动作起止状态。"
                             className="h-[92px] w-full resize-none rounded-md border border-white/10 bg-black/35 px-2 py-1.5 text-[10px] leading-relaxed text-white/75 outline-none placeholder:text-white/25 focus:border-amber-300/45"
                           />
                         </label>