feat: make subject conversion dialog-driven

.memory/assistant-handoff.md

@@ -1,6 +1,6 @@
 # 项目接力
 
-- 生成时间：May 19, 2026 at 08:44
+- 生成时间：May 20, 2026 at 13:43
 - 项目：SKG Marketing Studio / SKG 营销内容工作台
 - 路径：/Users/kangwan/Projects/business/20260512-20260512-skg-tk-二创验证
 - 状态：active
@@ -9,8 +9,8 @@
 ## 最近助手会话概览
 
 - Claude：be53a031-9311-4ee8-b822-d4cfb4f5e78b · 时间未知
-- Codex：019e3dae-6045-7203-bf4e-8fbeae65cadf · 时间未知
+- Codex：019e3db1-012e-7163-bc78-acf7cde326e7 · 时间未知
-- Cursor：3e4af548-3b65-45a5-8698-6e75510f24b5 · May 19, 2026 at 08:43
+- Cursor：未找到匹配当前项目的最近会话
 
 ## Claude 最近会话
 
@@ -44,64 +44,37 @@
 
 ## Codex 最近会话
 
-- Session ID：019e3dae-6045-7203-bf4e-8fbeae65cadf
+- Session ID：019e3db1-012e-7163-bc78-acf7cde326e7
-- Transcript：/Users/kangwan/.codex/sessions/2026/05/19/rollout-2026-05-19T08-41-38-019e3dae-6045-7203-bf4e-8fbeae65cadf.jsonl
+- Transcript：/Users/kangwan/.codex/sessions/2026/05/19/rollout-2026-05-19T08-44-30-019e3db1-012e-7163-bc78-acf7cde326e7.jsonl
 - 工作目录：/Users/kangwan/Projects/business/20260512-20260512-skg-tk-二创验证
 - 分支：main
 - 敏感字段：已对 token / key / password / secret 做脱敏
 
 ### 最近用户要求
 
-- # AGENTS.md instructions for /Users/kangwan/Projects/business/20260512-20260512-skg-tk-二创验证  <INSTRUCTIONS> # SKG AI 素材管线 - TK 二创验证 Agent Rules  ## Must Read First  - `.project.json` 是机器真源：公网链接、快捷登录、凭证引用都以它为准 - `RULES.md` 是人工规则和部署事实：启动命令、平台、域名、注意事项都写这里 - 工作看板全局规则是本项目的开发收口默认规则；本地规则缺失时按 `/Users/kangwan/Projects/code/20260317-rules-dashboard/RULES.md`、`SCHEMA.md`、`rules/03-Git约定.md`、`rules/04-版本发布规则.md` 补齐 - 不允许编造不存在的域名、账号、密码；未知就保持空白并明确标记待补充  ## Development Ledger Contract  - 开发任务结束前必须执行并汇报 `git status -sb` - 功能、修复、规则或部署元数据变更完成后，必须创建人工语义 commit；`auto-save` 只算安全快照 - Gitea 是主远端，`origin` 必须指向 Gitea；…
+- 可以
-- 你现在是在 Codex CLI 里的项目承接会话。 这不是对话立项台；项目目录已经存在，当前目标是接住上下文并等待用户下一步。  项目路径：/Users/kangwan/Projects/business/20260512-20260512-skg-tk-二创验证  承接步骤： - 先读取 `.memory/assistant-handoff.md` - 再阅读 `.project.json`、`RULES.md`、`AGENTS.md` - 结合当前仓库状态建立上下文  要求： - 不要要求用户重新粘贴 Claude / Codex / Cursor 的总结 - 先理解最近进度、未完成项和下一步候选 - 不要立即修改代码、继续上一个任务或自行执行命令 - 等待用户给出下一条明确指令后再开始动手 - 如果后续需要 Cursor，提醒用户从看板打开 Cursor GUI 到当前项目目录
+- 继续
+- <image name=[Image #1]> </image> [Image #1] 那会这个是不是就不用了 ， 转换层 这里要重构样式 ，直接对话式 那个数量直接就在对话中完成
 
 ### Codex 最近回复
 
-> 本地状态和 handoff 里的快照有一处漂移：现在除了 `.memory/*`，`.project.json` 也处于未提交修改状态。我再补一眼远端、最新提交和领先/落后计数，只做状态确认。
+> 我准备动两个关键点：后端让 `/subject-agent/message` 从中文对话里识别“方向 + 数量”，前端删除四个模式卡和数量按钮，只展示当前识别出的方向/张数作为摘要。这样你后面说“卡通，生成 8 张”或“形象锁定，6 张”就直接改变状态。
 
-> 已完成承接，只读，没有改代码、没有启动服务、没有继续上个任务。
+> 前端显式卡片已经先移掉了，同时我顺手避免了“卡通默认 3D 风格”硬覆盖对话里的风格要求。接下来补文档和运行验证。
-> 
-> 当前理解：
-> - 项目是 `SKG Marketing Studio / SKG 营销内容工作台`，生产入口 `https://marketing.skg.com`，后端走 `/api`，主远端是 Gitea。
-> - 当前产品方向：先做信息流广告快速复刻第一步，源视频下载优先；下载后并行做音频文案分析和视觉参考帧准备；直接提交视频模型目前是暂停状态。
-> - 最近 Codex 进度：资源库 / 提示词库线上 API 链路已测通，登录、健康检查、创建、搜索、预览、删除、recent 都验证过；最新提交是 `3b7a062 chore: persist resource libraries in production`。
-> - 最近候选问题：音频处理卡片过大且空展示不美观、需要说明用途和调用模型；缩略图下方横向滚动条需要更明显、更好拖。
-> - 后续媒体 UI 改动要复用 `web/components/media-asset-tile.tsx`；影响产品理解、节点职责、接口、数据模型或操作路径的改动，要同步更新 `docs/source-analysis.html`。
-> 
-> 当前仓库状态：
-> ```text
-> ## main...origin/main
->  M .memory/assistant-handoff.md
->  M .memory/worklog.json
->  M .project.json
-> ```
-> 
-> `main` 对 `origin/main` 领先/落后是 `0/0`。`.project.json` 当前未提交 diff 看起来主要是格式化变化，但也观察到 `urls` 里原来的 `source-analysis` 文档项被替换/移除了，这个我不会处理，等你下一条明确指令。
-> 
-> 如果后续需要 Cursor，请从看板打开 Cursor GUI 到这个精确目录：`/Users/kangwan/Projects/business/20260512-20260512-skg-tk-二创验证`。
-> 
-> 我现在暂停，等你下一步指令。
-> 
-> <…
 
 ## Cursor 最近会话
 
-- Session ID：3e4af548-3b65-45a5-8698-6e75510f24b5
+- 未在本机找到匹配这个项目的 Cursor transcript。
-- Transcript：/Users/kangwan/.cursor/projects/Users-kangwan-Projects-business-20260512-20260512-skg-tk/agent-transcripts/3e4af548-3b65-45a5-8698-6e75510f24b5/3e4af548-3b65-45a5-8698-6e75510f24b5.jsonl
-- 工作目录：/Users/kangwan/Projects/business/20260512-20260512-skg-tk-二创验证
-- 最后活动：May 19, 2026 at 08:43
-- 敏感字段：已对 token / key / password / secret 做脱敏
-- 内容隔离：公司项目默认不写入 Cursor transcript 的用户要求和助手回复；这里只记录 Cursor 曾在该项目下工作。
 
 ## 当前仓库状态
 
 - 当前分支：main
-- 未提交变更：2 项
+- 未提交变更：8 项
-- 最近提交：chore: persist resource libraries in production
+- 最近提交：docs: record image timeout deployment
 - 变更文件：
--    M .memory/assistant-handoff.md
 -    M .memory/worklog.json
--    M .project.json
+-    M api/main.py
+-    M web/components/ad-recreation-board.tsx
 
 ## 统一接力要求
 

RULES.md

@@ -15,7 +15,7 @@
 
 ## 部署事实
 - 平台：VPS `76.13.31.179`（Ubuntu 24.04 / Docker Compose / Coolify Traefik）
-- 发布状态：已部署并验证（2026-05-19，主体元素改为按套图文件夹分组展示，主体生成接口提交后立即返回 queued 占位并后台逐视角生成、逐张回填；转换层为真人重构 / 卡通重构 / 元素重构 / 自主描述四个入口，每个入口最多 3 张参考帧；拖入只加入参考队列，点击生成后固定生成全新 6 视图；胶片双击/拖拽加入参考帧池 + 胶片缓存复用 + 音频解析失败可重试，右侧三栏主体管线：竖向参考帧池 + 转换层 + 主体元素，参考帧缩略图保持小尺寸 9:16 比例 + hover 左侧紧凑预览 + 转换层多参考滚动，旧主体模板区移出主路径 + 逐句时间轴移到原版视频下方并支持双行文案 + 波形同框时间对齐画面胶片 + 胶片密度按钮上移波形顶部 + 去分隔线 + 胶片上下错落 + body 顶层原位大放大 + 隐藏源视频工作区音频解析摘要卡 + 隐藏工作区顶部状态提示条 + 三字段候选生成工作流 + 折叠紧凑候选区）；`https://marketing.skg.com` 已启用应用内登录页，未登录 API 返回 401，认证后首页 200；容器内 `/health` 返回 `ok:true`
+- 发布状态：已部署并验证（2026-05-20，主体元素按套图文件夹分组展示，主体生成接口提交后立即返回 queued 占位并后台逐视角生成、逐张回填；右侧三栏主体管线为竖向参考帧池 + 对话式转换层 + 主体元素；转换层只保留一个参考区和生图对话，不再显示方向卡片、卡通风格下拉或单独数量按钮，方向、数量、风格、服装统一、人物占比和保留/删除元素都由对话识别后写入 `Job.subject_agent`；胶片双击/拖拽加入参考帧池 + 胶片缓存复用 + 音频解析失败可重试，参考帧缩略图保持小尺寸 9:16 比例 + hover 左侧紧凑预览 + 转换层多参考滚动，旧主体模板区移出主路径 + 逐句时间轴移到原版视频下方并支持双行文案 + 波形同框时间对齐画面胶片 + 胶片密度按钮上移波形顶部 + 去分隔线 + 胶片上下错落 + body 顶层原位大放大 + 隐藏源视频工作区音频解析摘要卡 + 隐藏工作区顶部状态提示条 + 三字段候选生成工作流 + 折叠紧凑候选区）；`https://marketing.skg.com` 已启用应用内登录页，未登录 API 返回 401，认证后首页 200；容器内 `/health` 返回 `ok:true`
 - 最近部署验证（2026-05-19）：`fd794e3` 已推送并部署到 `/opt/skg-marketing-studio`；生产 `/health` 显示 `image=gpt-image-2`、`subject_image=gpt-image-2`、`image_request_timeout_seconds=60`、`image_base_url=https://ai.skg.com/ezlink/v1`。容器内最小文字生图探针在 20 秒限制下返回 `ReadTimeout`，说明当前阻塞点是 `https://ai.skg.com/ezlink/v1` 的 `gpt-image-2` 上游通道超时，服务端不会更换图片模型。
 - 最近部署验证（2026-05-19）：`3756259` 已推送并部署到 `/opt/skg-marketing-studio`；生产 `/health` 显示 `image=gpt-image-2`、`image_fallbacks=['gemini-3-pro-image-preview']`、`subject_image_fallbacks=['gpt-image-2','gemini-3-pro-image-preview']`、短时熔断阈值 2 次 / 600 秒。线上真实探针确认 `gpt-image-2` 读超时后同次调用可自动兜底到 `gemini-3-pro-image-preview` 并返回图片；模拟探针确认连续 2 次主模型失败后第三次直接走 Gemini。
 - 最近部署验证（2026-05-20）：`c245bff` 已推送并部署到 `/opt/skg-marketing-studio`；本地 `python3 -m py_compile api/main.py` 和 `web/npm run build` 通过，生产 Docker 重建后 `./scripts/verify-prod-docker.sh` 通过（web 容器 Up、API 容器 Up、`/login/` 200、缺失 `_next` 资源 404、未登录 `/api/health` 401、容器内 `api:health ok`）。容器内模型偏好探针确认转换层 `image_model_preference` 路由：`auto -> ['gpt-image-2','gemini-3-pro-image-preview']`、`gpt-image-2 -> ['gpt-image-2']`、`gemini-3-pro-image-preview -> ['gemini-3-pro-image-preview']`。

api/main.py

@@ -107,6 +107,7 @@ IMAGE_MODEL = GPT_IMAGE_MODEL
 PRODUCT_VIEW_MODEL = GPT_IMAGE_MODEL
 SUBJECT_ASSET_IMAGE_MODEL = GPT_IMAGE_MODEL
 SubjectModelBundle = Literal["gpt", "gemini"]
+SubjectAgentMode = Literal["realistic", "cartoon", "elements", "custom"]
 SUBJECT_AGENT_GPT_MODEL = gpt_model_env("SUBJECT_AGENT_GPT_MODEL", VISION_MODEL)
 SUBJECT_AGENT_GEMINI_MODEL = os.getenv("SUBJECT_AGENT_GEMINI_MODEL", "gemini-2.5-flash").strip() or "gemini-2.5-flash"
 SUBJECT_ASSET_IMAGE_MODELS = [GPT_IMAGE_MODEL] + (
@@ -766,7 +767,7 @@ class SubjectAgentState(BaseModel):
     source_frame_indices: list[int] = Field(default_factory=list)
     analysis: SubjectAgentAnalysis | None = None
     messages: list[SubjectAgentMessage] = Field(default_factory=list)
-    selected_mode: Literal["realistic", "cartoon", "elements", "custom"] = "custom"
+    selected_mode: SubjectAgentMode = "custom"
     selected_traits: list[str] = Field(default_factory=list)
     requirements_zh: str = ""
     generation_prompt_en: str = ""
@@ -4103,27 +4104,85 @@ def _subject_agent_analysis(job_id: str, source_indices: list[int], bundle: Subj
     )
 
 
-def _subject_agent_message_update(state: SubjectAgentState, user_message: str) -> tuple[str, str, str, int, list[str]]:
+_SUBJECT_AGENT_MODES: set[str] = {"realistic", "cartoon", "elements", "custom"}
+
+
+def _subject_agent_quantity_from_text(text: str, fallback: int) -> int:
+    quantity = max(1, min(10, int(fallback or 6)))
+    text = text or ""
+    if re.fullmatch(r"\s*\d{1,2}\s*", text):
+        return max(1, min(10, int(text.strip())))
+    digit_match = re.search(r"(\d{1,2})\s*(?:张|个|视图|张图|图|views?)", text, flags=re.I)
+    if digit_match:
+        return max(1, min(10, int(digit_match.group(1))))
+    cn_numbers = {
+        "一": 1,
+        "二": 2,
+        "两": 2,
+        "三": 3,
+        "四": 4,
+        "五": 5,
+        "六": 6,
+        "七": 7,
+        "八": 8,
+        "九": 9,
+        "十": 10,
+    }
+    cn_match = re.search(r"([一二两三四五六七八九十])\s*(?:张|个|视图|张图|图)", text)
+    if cn_match:
+        return max(1, min(10, cn_numbers.get(cn_match.group(1), quantity)))
+    return quantity
+
+
+def _subject_agent_mode_from_text(text: str, fallback: SubjectAgentMode = "custom") -> SubjectAgentMode:
+    compact = re.sub(r"\s+", "", text or "").lower()
+    if re.search(r"卡通|动画|插画|公仔|潮玩|二次元|cartoon|anime|illustration|toy|stylized", compact):
+        return "cartoon"
+    if re.search(r"创意复刻|创意模式|元素|参考创新|不像|换人|全新主体|全新人物|不同人|newperson|newactor|concept|element", compact):
+        return "elements"
+    if re.search(r"形象锁定|复刻这个人|复刻形象|同一主体|同一个人|保持这个人|保持原主体|完全复刻|source locked|same subject|sameperson", compact):
+        return "realistic"
+    if re.search(r"自主描述|只按文字|不依赖|不用参考|按描述|fromdescription|custom", compact):
+        return "custom"
+    return fallback
+
+
+def _subject_agent_mode_from_value(value: object, fallback: SubjectAgentMode) -> SubjectAgentMode:
+    text = str(value or "").strip()
+    return text if text in _SUBJECT_AGENT_MODES else fallback
+
+
+def _subject_agent_message_update(state: SubjectAgentState, user_message: str) -> tuple[str, str, str, int, list[str], SubjectAgentMode]:
     current_req = state.requirements_zh.strip()
     selected_traits = state.selected_traits[:20]
-    quantity = max(1, min(10, int(state.quantity or 6)))
+    quantity = _subject_agent_quantity_from_text(user_message, int(state.quantity or 6))
-    qty_match = re.search(r"(\d{1,2})\s*张", user_message)
+    selected_mode = _subject_agent_mode_from_text(user_message, state.selected_mode)
-    if qty_match:
-        quantity = max(1, min(10, int(qty_match.group(1))))
     fallback_req = "；".join(part for part in [current_req, user_message.strip()] if part).strip("；")
+    mode_label = {
+        "realistic": "source-locked same visible subject reconstruction",
+        "cartoon": "cartoon or stylized reconstruction",
+        "elements": "creative element reconstruction with a different new subject",
+        "custom": "custom description driven subject generation",
+    }.get(selected_mode, "custom description driven subject generation")
     fallback_prompt = _ensure_english(
         "Subject image generation requirements: "
         + (fallback_req or "create a consistent SKG ad subject pack")
+        + f". Direction mode: {mode_label}."
+        + f" Generate exactly {quantity} separate views."
         + ". Keep one identity and one outfit bible across all generated views. "
         + (f"Selected traits: {', '.join(selected_traits)}." if selected_traits else "")
     )
     if not LLM_API_KEY:
-        return "已记录这条生图要求。继续补充要保留/删除的元素，确认后我会按当前要求生成。", fallback_req, fallback_prompt, quantity, selected_traits
+        return "已记录这条生图要求。继续补充要保留/删除的元素，确认后我会按当前要求生成。", fallback_req, fallback_prompt, quantity, selected_traits, selected_mode
     system = (
         "You are an SKG subject image-generation requirements agent. Your scope is only image generation for a subject view pack. "
         "Do not answer unrelated video, audio, download, coding, copywriting, or general chat requests; redirect to subject image requirements. "
         "Normalize the user's fuzzy Chinese request into precise generation constraints. "
-        "Return strict JSON with keys: assistant_message_zh, updated_requirements_zh, generation_prompt_en, quantity, selected_traits. "
+        "Infer selected_mode from the conversation. Allowed selected_mode values are realistic, cartoon, elements, custom. "
+        "Use realistic when the user wants to lock or replicate the visible reference subject; cartoon for stylized/cartoon/toy/illustration; "
+        "elements when the user wants the creative logic but a different new subject; custom when the user wants free text generation without relying on references. "
+        "Infer quantity from Chinese or English requests such as 4张, 六视图, generate 8 views. "
+        "Return strict JSON with keys: assistant_message_zh, updated_requirements_zh, generation_prompt_en, quantity, selected_traits, selected_mode. "
         "generation_prompt_en must be English and must enforce: one consistent identity, one consistent outfit bible, neck/shoulder readability, no text/watermarks/UI, and legal-safe reconstruction."
     )
     user_payload = {
@@ -4153,12 +4212,13 @@ def _subject_agent_message_update(state: SubjectAgentState, user_message: str) -
         assistant = str(data.get("assistant_message_zh") or "已记录这条生图要求。").strip()[:1200]
         updated_req = str(data.get("updated_requirements_zh") or fallback_req).strip()[:2200]
         prompt_en = _ensure_english(str(data.get("generation_prompt_en") or fallback_prompt).strip())[:2600]
-        out_quantity = max(1, min(10, int(data.get("quantity") or quantity)))
+        out_quantity = _subject_agent_quantity_from_text(str(data.get("quantity") or ""), quantity)
         out_traits = _list_of_strings(data.get("selected_traits"), 24) or selected_traits
-        return assistant, updated_req, prompt_en, out_quantity, out_traits
+        out_mode = _subject_agent_mode_from_value(data.get("selected_mode"), selected_mode)
+        return assistant, updated_req, prompt_en, out_quantity, out_traits, out_mode
     except Exception as e:
         print(f"[subject agent message failed] bundle={state.model_bundle} error={e}", flush=True)
-        return "已先按本地规则记录这条要求；模型回复失败时仍可直接生成。", fallback_req, fallback_prompt, quantity, selected_traits
+        return "已先按本地规则记录这条要求；模型回复失败时仍可直接生成。", fallback_req, fallback_prompt, quantity, selected_traits, selected_mode
 
 
 # ---------- API 路由 ----------
@@ -4179,7 +4239,7 @@ class SubjectAgentMessageReq(BaseModel):
 
     model_bundle: SubjectModelBundle = "gpt"
     source_frame_indices: list[int] = Field(default_factory=list)
-    selected_mode: Literal["realistic", "cartoon", "elements", "custom"] = "custom"
+    selected_mode: SubjectAgentMode = "custom"
     selected_traits: list[str] = Field(default_factory=list)
     requirements_zh: str = ""
     message: str = ""
@@ -4666,7 +4726,7 @@ def analyze_subject_agent(job_id: str, req: SubjectAgentAnalyzeReq) -> Job:
     state = job.subject_agent.model_copy(deep=True)
     assistant_text = (
         f"我已用 {req.model_bundle.upper()} 套件分析这些参考帧。"
-        "你可以选择形象锁定、创意复刻、元素混合或自主描述，也可以继续告诉我要改数量、风格、服装、人物大小。"
+        "接下来直接告诉我要复刻形象、卡通化、参考创意换新人，还是只按文字生成；数量、风格、服装和人物大小也都写在对话里。"
     )
     messages = (state.messages + [SubjectAgentMessage(role="assistant", content=assistant_text, created_at=time.time())])[-30:]
     state = state.model_copy(update={
@@ -4689,10 +4749,11 @@ def message_subject_agent(job_id: str, req: SubjectAgentMessageReq) -> Job:
         raise HTTPException(404, "job not found")
     state = job.subject_agent.model_copy(deep=True)
     source_indices = [idx for idx in req.source_frame_indices if any(frame.index == idx for frame in job.frames)][:8]
+    fallback_mode = req.selected_mode or state.selected_mode
     state = state.model_copy(update={
         "model_bundle": req.model_bundle,
         "source_frame_indices": source_indices or state.source_frame_indices,
-        "selected_mode": req.selected_mode,
+        "selected_mode": fallback_mode,
         "selected_traits": [str(item).strip()[:80] for item in req.selected_traits if str(item).strip()][:24],
         "requirements_zh": req.requirements_zh.strip()[:2200] or state.requirements_zh,
         "quantity": max(1, min(10, int(req.quantity or state.quantity or 6))),
@@ -4700,7 +4761,7 @@ def message_subject_agent(job_id: str, req: SubjectAgentMessageReq) -> Job:
     user_message = req.message.strip()
     if not user_message:
         user_message = state.requirements_zh or "按当前设置准备主体套图生成要求"
-    assistant_text, requirements_zh, prompt_en, quantity, selected_traits = _subject_agent_message_update(state, user_message)
+    assistant_text, requirements_zh, prompt_en, quantity, selected_traits, selected_mode = _subject_agent_message_update(state, user_message)
     messages = (
         state.messages
         + [SubjectAgentMessage(role="user", content=user_message, created_at=time.time())]
@@ -4709,6 +4770,7 @@ def message_subject_agent(job_id: str, req: SubjectAgentMessageReq) -> Job:
     state = state.model_copy(update={
         "requirements_zh": requirements_zh,
         "generation_prompt_en": prompt_en,
+        "selected_mode": selected_mode,
         "quantity": quantity,
         "selected_traits": selected_traits,
         "messages": messages,

web/components/ad-recreation-board.tsx

@@ -1147,7 +1147,9 @@ function buildReconstructionDirection(
   } else if (mode === "cartoon") {
     common.push(
       "Direction mode: cartoon reconstruction.",
-      `Cartoon style: ${style.label}; ${style.prompt}.`,
+      trimmed
+        ? `Cartoon style: follow the user's requested style from the direction text; if no explicit cartoon style is specified, use ${style.label}; ${style.prompt}.`
+        : `Cartoon style: ${style.label}; ${style.prompt}.`,
       "Transform broad pose, emotion, body-readability, and ad energy into a fully original stylized character, not a realistic human and not a traced version of the source.",
     )
   } else if (mode === "elements") {
@@ -3316,8 +3318,7 @@ function SourceSubjectPipeline({
   const [agentInput, setAgentInput] = useState("")
   const [subjectAgentBusy, setSubjectAgentBusy] = useState<"analyze" | "message" | null>(null)
   const [promptMemoryByMode, setPromptMemoryByMode] = useState<Record<SubjectReconstructionMode, string[]>>(() => loadSubjectPromptMemory(job.id))
-  const [cartoonStyle, setCartoonStyle] = useState<CartoonReconstructionStyle>("3d_animation")
+  const [cartoonStyle] = useState<CartoonReconstructionStyle>("3d_animation")
-  const [cartoonStyleOpen, setCartoonStyleOpen] = useState(false)
   const [subjectBusyFor, setSubjectBusyFor] = useState<{ jobId: string; jobLabel: string; mode: SubjectReconstructionMode; viewCount: number; sourceCount: number; profileLabel: string; modelLabel: string } | null>(null)
   const [subjectAssetBusy, setSubjectAssetBusy] = useState<string | null>(null)
   const [expandedSubjectPackKey, setExpandedSubjectPackKey] = useState<string | null>(null)
@@ -3417,7 +3418,6 @@ function SourceSubjectPipeline({
     setLastSubjectProfile(null)
     setSubjectBusyFor(null)
     setSubjectAssetBusy(null)
-    setCartoonStyleOpen(false)
     setExpandedSubjectPackKey(null)
   }, [job.id])
 
@@ -3917,72 +3917,20 @@ function SourceSubjectPipeline({
             </button>
           </div>
 
-          <div className="mt-2 grid grid-cols-2 gap-1">
+          <div className="mt-2 rounded-md border border-[#d6b36a]/25 bg-[#d6b36a]/[0.065] px-2.5 py-2">
-            {RECONSTRUCTION_MODES.map((modeConfig) => (
+            <div className="flex flex-wrap items-center gap-1.5">
-              <button
+              <span className="rounded-full border border-[#d6b36a]/35 bg-black/28 px-2 py-0.5 text-[9.5px] font-semibold text-[#f4dd98]">
-                key={modeConfig.value}
+                方向：{reconstructionModeConfig(agentMode).label}
-                type="button"
+              </span>
-                onClick={() => setAgentMode(modeConfig.value)}
+              <span className="rounded-full border border-white/10 bg-black/28 px-2 py-0.5 text-[9.5px] font-semibold text-white/58">
-                className={`rounded-md border px-2 py-1.5 text-left transition ${
+                数量：{selectedSubjectViews.length} 张
-                  agentMode === modeConfig.value
+              </span>
-                    ? "border-[#d6b36a]/70 bg-[#d6b36a]/14 text-white"
+              <span className="rounded-full border border-white/10 bg-black/28 px-2 py-0.5 text-[9.5px] text-white/42">
-                    : "border-white/10 bg-black/24 text-white/48 hover:border-white/22 hover:text-white"
+                对话自动识别
-                }`}
+              </span>
-                title={modeConfig.subtitle}
-              >
-                <span className="block text-[10.5px] font-semibold">{modeConfig.label}</span>
-                <span className="mt-0.5 block truncate text-[8.5px] text-white/34">{modeConfig.subtitle}</span>
-              </button>
-            ))}
-          </div>
-
-          {agentMode === "cartoon" ? (
-            <div className="mt-2 rounded-md border border-white/10 bg-black/24 p-1.5">
-              <button
-                type="button"
-                onClick={() => setCartoonStyleOpen((open) => !open)}
-                className="inline-flex h-7 w-full items-center justify-between rounded-md border border-white/10 bg-black/28 px-2 text-[10px] font-semibold text-white/58 transition hover:border-white/20 hover:text-white"
-              >
-                <span>卡通风格：{cartoonStyleConfig(cartoonStyle).label}</span>
-                <ChevronDown className={`h-3.5 w-3.5 transition ${cartoonStyleOpen ? "rotate-180" : ""}`} />
-              </button>
-              {cartoonStyleOpen ? (
-                <div className="mt-1 grid grid-cols-2 gap-1">
-                  {CARTOON_RECONSTRUCTION_STYLES.map((style) => (
-                    <button
-                      key={style.value}
-                      type="button"
-                      onClick={() => {
-                        setCartoonStyle(style.value)
-                        setCartoonStyleOpen(false)
-                      }}
-                      className={`h-7 rounded-md border px-1.5 text-[9.5px] font-semibold transition ${
-                        cartoonStyle === style.value ? "border-white bg-white text-black" : "border-white/10 bg-black/24 text-white/45 hover:text-white"
-                      }`}
-                    >
-                      {style.label}
-                    </button>
-                  ))}
-                </div>
-              ) : null}
             </div>
-          ) : null}
+            <div className="mt-1.5 text-[9.5px] leading-snug text-white/42">
-
+              在下方直接说“形象锁定 / 卡通重构 / 创意复刻 / 自主描述”和“生成几张”，系统会更新这里，不再单独点卡片或数量。
-          <div className="mt-2 flex items-center justify-between gap-2 rounded-md border border-white/10 bg-black/24 px-2 py-1.5">
-            <span className="text-[10px] font-semibold text-white/58">数量</span>
-            <div className="flex gap-1">
-              {[4, 6, 8, 10].map((count) => (
-                <button
-                  key={count}
-                  type="button"
-                  onClick={() => setAgentQuantity(count)}
-                  className={`h-6 rounded-full border px-2 text-[9.5px] font-semibold transition ${
-                    agentQuantity === count ? "border-white bg-white text-black" : "border-white/10 bg-black/24 text-white/45 hover:text-white"
-                  }`}
-                >
-                  {count}
-                </button>
-              ))}
             </div>
           </div>