feat: 接入形态图库与 Seedream 生图操作

展示已有主图、候选图和失败记录，支持单图及批量生成、主图选择。
补充正式 ID 校验、费用确认、图片加载占位和回归测试。
依赖后端 eac8dc3 的项目形态图库只读接口。

README.md

@@ -34,8 +34,9 @@ API_PROXY_TARGET=http://localhost:3412
 | 分镜检查 | 按剧集、Beat、正式 Shot ID 浏览；导演设计、环境与主体即时状态；单集覆盖率与 JSON 导出 |
 | 分镜生成 | Direction / VisualState 单集生成、项目批量补齐、覆盖重生成；并发和自动修复次数；部分失败诊断 |
 | 镜头工具 | 主体参考图与缺失原因、确定性 GenerationSpec 查询与导出、单镜头／批量视频提示词 |
+| 形态图片 | 正式形态图库、主图／候选图预览、图片历史与失败记录、Seedream 单图／批量生图、切换主参考图 |
 
-尚未实现：手工编辑剧本或分镜设计（后端暂无对应写接口）、主体图生成、实际视频生成、用户登录。模型输出按纯文本显示，避免执行不可信 HTML。参考图只接受 HTTP(S) 或后端 `/storage/` 路径。
+尚未实现：手工编辑剧本或分镜设计（后端暂无对应写接口）、实际视频生成、用户登录。模型输出按纯文本显示，避免执行不可信 HTML。参考图只接受 HTTP(S) 或后端 `/storage/` 路径。
 
 ## 技术与规范
 
@@ -65,6 +66,7 @@ Oxfmt 不负责代码质量，Oxlint 不负责 Vue 的完整类型推导；Vue S
 | `src/features/create-drama` | 剧本创作 graph 页面 |
 | `src/features/breakdown` | 拆解 graph 的 API、类型、配置页面、主体与分镜组件 |
 | `src/features/storyboard` | 分镜 API、类型、生成操作、依赖判断、设计与状态、镜头资源、生成回执 |
+| `src/features/subject-images` | 正式形态图库、图片生成参数、历史图片、主图选择与批量回执 |
 | `src/features/workflows` | checkpoint 选择、执行记录、长请求互斥与操作确认 |
 | `src/components/ui` | 跨业务使用的 Reka UI 封装与公共展示组件 |
 | `src/composables` | 具备取消和竞态保护的轮询 |
@@ -76,7 +78,7 @@ Oxfmt 不负责代码质量，Oxlint 不负责 Vue 的完整类型推导；Vue S
 ## 接口基线
 
 对齐后端 `qianlanse/short-drama-agent` 的 `dev`：
-`6ce284b9d93e8c9d0b7ca57d265148e354642095`。
+`eac8dc3b1b2ce1ff2bdfeb01d7281cd838f07285`，包含本次新增的只读 `GET /projects/:projectId/subject-forms` 接口。请同步更新前后端 `dev` 后重启后端。
 
 | HTTP | 路径（以 /api 为前缀） | 用途 |
 | --- | --- | --- |
@@ -100,6 +102,11 @@ Oxfmt 不负责代码质量，Oxlint 不负责 Vue 的完整类型推导；Vue S
 | GET | `/storyboard-shots/:shotId/generation-spec` | 后端确定性编译结果，不调用模型 |
 | POST | `/storyboard-shots/:shotId/video-prompt` | 复用已有 Prompt 或调用模型生成；`force` 控制覆盖 |
 | POST | `/projects/:id/video-prompts/generate` | 项目批量 Prompt；`concurrency`、`force` |
+| GET | `/projects/:id/subject-forms` | 正式 SubjectForm ID、主体名称／引用、主图与图片元数据；本次新增只读接口 |
+| GET | `/subject-forms/:formId/images` | 此形态的全部图片历史及实际 Prompt |
+| POST | `/subject-forms/:formId/images` | 单图生成；显式 `provider: seedream`，可选 Prompt 和成对尺寸，`setPrimary` |
+| PUT | `/subject-forms/:formId/images/:imageId/primary` | 选择已完成图片作为主参考图，不调用模型 |
+| POST | `/projects/:id/subject-images/generate` | 全项目批量生图；`provider`、`concurrency`、`force` |
 
 API 层也提供 `state` 与 `breakdown/latest` 方法。页面通过共享 checkpoint 查询取得阶段状态，避免重复拉取同样内容。
 
@@ -127,6 +134,18 @@ API 层也提供 `state` 与 `breakdown/latest` 方法。页面通过共享 chec
 9. 参考图和 GenerationSpec 按需 GET，切换镜头或设计变化时取消旧查询。GenerationSpec 需要有效的 Direction、VisualState 和形态关联，不生成图片或视频。
 10. 后端当前的 Video Prompt 仍使用镜头描述和参考图，**尚未消费 GenerationSpec**。没有独立 GET Prompt 接口，“读取／生成提示词”仍是 POST；不存在 Prompt 时会调用模型，因此必须确认。`force: false` 复用已有正文；批量响应只有计数和失败列表，不能凭计数构造 Prompt 正文。`generated` 是后端处理成功数，可能包含已存在但状态不为 `prompt_ready` 的复用项。
 
+### 形态图片约定
+
+1. 入口为 `/projects/:projectId/subject-images`；项目导航、拆解主体列表和分镜参考图缺失提示均可进入。图库自动读取后端已生成图片，不需要重新生图才能显示。
+2. 生图使用正式数据库 `SubjectForm.id`。Checkpoint 的领域 `formId` 不是数据库 ID，不能直接调用图片接口。新增查询从项目关联读取正式形态，返回轻量图片元数据，不包含图片的 `rawJson` 或长 Prompt；打开历史弹窗后再读取详细记录。
+3. 单图明确发送 `provider: 'seedream'`，避免后端单图入口默认使用 mock。模型和凭证仅配置在后端，前端不保存 API Key。默认使用后端的 generationPrompt，其次 appearancePrompt；可为本次输入自定义 Prompt，详情显示后端最终实际使用的文本。
+4. 单图宽高同时留空使用后端默认 2K，自定义时必须同时提供正整数，具体尺寸限制仍由模型校验。当前 Seedream Provider 没有使用 negativePrompt，所以页面不提供无效的负向提示词输入。
+5. 单图生成新增历史记录，不删除旧图。无主图时默认勾选“生成成功后设为主图”；已有主图时默认不替换。候选图可以在历史弹窗二次确认后通过 PUT 设为主图，只允许已完成图片。
+6. 批量默认跳过已有主图的形态；勾选“已有主图也新增候选图”发送 `force: true`。后端批量 force 只生成新图，**不会替换已有主图**。批量作用于整个项目，不受类型／文字／缺图筛选影响，也不接收单图的尺寸与自定义 Prompt。
+7. 图库每 6 秒读取一次真实记录，弹窗只在打开时查询该形态图片。数据库存在 pending / generating 时暂停再次生图；网络错误不自动重试有费用的请求。HTTP 200 中的批量部分失败单独展示。
+8. 图片使用持久化后的 `/storage/...` 地址，开发代理已配置。生产环境也必须代理 `/storage/`，不能让 SPA 回退返回 index.html。无效 URL、加载失败、无图和后端生成失败有各自提示；不会自动使用可能过期的 Provider 临时地址。
+9. 设为主图后，回到分镜点击“刷新参考图”读取最新绑定。已有视频提示词不会自动重生成，需要按需覆盖。本页面不会调用 `/production/start` 或实际视频生成。
+
 ### 后端限制
 
 目前 `checkpoints` 接口返回完整历史与 state，没有分页或轻量状态接口。长篇剧本会增加轮询流量。下一步建议后端增加：
@@ -182,5 +201,7 @@ server {
 6. 进入分镜设计，确认正式镜头 ID 与 Beat 对应；先生成单集 Direction，再生成 VisualState，检查覆盖率和主体状态。
 7. 用小型测试项目验证批量补齐、跳过完整集、失败诊断和明确覆盖；确认关闭覆盖后不会重生成完整设计。
 8. 给主体形态准备参考图，读取镜头参考图和 GenerationSpec；确认 Prompt 无正文时经确认生成、有正文且不覆盖时复用。不提交实际视频任务。
+9. 打开“形态图片”，检查已有主图自动显示；单个形态新增候选图，查看历史、失败原因和实际 Prompt，再设为主图。
+10. 在小型测试项目中验证批量补齐与“已有主图也新增候选图”；回到分镜刷新参考图，确认使用新选择的主图。
 
-自动测试使用模拟 API，覆盖两条既有流程及分镜 API 参数、依赖判断、正式 ID 合并、校验与部分失败、切换剧集的竞态、查询错误恢复、Prompt 确认和纯文本展示。不等同于真实模型、MySQL 或浏览器视觉联调。请勿用正式项目随意制造失败以测试恢复。
+自动测试使用模拟 API，覆盖既有流程、分镜接口与竞态，以及形态图片自动读取、正式 ID 生图、主图 PUT、费用确认、尺寸校验、候选图选择、批量部分失败和图片加载错误。不等同于真实模型、MySQL 或浏览器视觉联调。请勿用正式项目随意制造失败以测试恢复。

src/App.vue

@@ -4,6 +4,7 @@ import { useRoute } from 'vue-router'
 import {
     Clapperboard,
     Camera,
+    Images,
     FolderOpen,
     FileText,
     Layers,
@@ -55,13 +56,20 @@ const apiBase = import.meta.env.VITE_API_BASE_URL || '/api'
                         title="剧本拆解"
                         ><Layers :size="18" /><span class="sidebar-label">剧本拆解</span
                         ><span class="sidebar-label ml-auto text-[10px] text-stone-500">02</span></RouterLink
+                    ><RouterLink
+                        :to="`/projects/${projectId}/subject-images`"
+                        class="side-link"
+                        active-class="selected"
+                        title="形态图片"
+                        ><Images :size="18" /><span class="sidebar-label">形态图片</span
+                        ><span class="sidebar-label ml-auto text-[10px] text-stone-500">03</span></RouterLink
                     ><RouterLink
                         :to="`/projects/${projectId}/storyboard`"
                         class="side-link"
                         active-class="selected"
                         title="分镜设计"
                         ><Camera :size="18" /><span class="sidebar-label">分镜设计</span
-                        ><span class="sidebar-label ml-auto text-[10px] text-stone-500">03</span></RouterLink
+                        ><span class="sidebar-label ml-auto text-[10px] text-stone-500">04</span></RouterLink
                     >
                 </nav></template
             >

src/components/ui/AppDialog.vue

@@ -11,7 +11,7 @@ import {
 import { X } from '@lucide/vue'
 
 /** 统一弹窗容器，Reka UI 负责焦点锁定、Escape 和无障碍语义。 */
-defineProps<{ title: string; description: string; busy?: boolean }>()
+defineProps<{ title: string; description: string; busy?: boolean; wide?: boolean }>()
 const open = defineModel<boolean>('open', { default: false })
 </script>
 
@@ -22,6 +22,7 @@ const open = defineModel<boolean>('open', { default: false })
             <DialogOverlay class="dialog-overlay" />
             <DialogContent
                 class="dialog-content"
+                :class="{ 'dialog-wide': wide }"
                 @interact-outside="busy && $event.preventDefault()"
                 @escape-key-down="busy && $event.preventDefault()"
             >

src/components/ui/AssetImage.vue

@@ -0,0 +1,40 @@
+<script setup lang="ts">
+import { computed, ref, watch } from 'vue'
+import { ImageOff } from '@lucide/vue'
+import { referenceImageUrl } from '../../lib/assets'
+
+/** 安全图片展示，资源失效或地址不合法时提供明确占位，不自动改用临时 Provider URL。 */
+const props = withDefaults(
+    defineProps<{ src?: string | null; alt: string; emptyText?: string; retryable?: boolean }>(),
+    { retryable: true }
+)
+const failed = ref(false)
+const url = computed(() => (props.src ? referenceImageUrl(props.src) : null))
+watch(
+    () => props.src,
+    () => {
+        failed.value = false
+    }
+)
+</script>
+
+<template>
+    <div class="asset-image">
+        <img
+            v-if="url && !failed"
+            :src="url"
+            :alt="alt"
+            loading="lazy"
+            referrerpolicy="no-referrer"
+            @error="failed = true"
+        />
+        <span v-else class="asset-image-empty"
+            ><ImageOff :size="22" :stroke-width="1.3" /><span>{{
+                src ? '图片无法加载' : emptyText || '尚未生成图片'
+            }}</span
+            ><button v-if="failed && retryable" type="button" class="text-button" @click.stop="failed = false">
+                重试加载
+            </button></span
+        >
+    </div>
+</template>

src/components/ui/index.ts

@@ -2,3 +2,4 @@
 export { default as AppDialog } from './AppDialog.vue'
 export { default as EmptyState } from './EmptyState.vue'
 export { default as StatusBadge } from './StatusBadge.vue'
+export { default as AssetImage } from './AssetImage.vue'

src/features/breakdown/BreakdownPage.vue

@@ -341,6 +341,7 @@ function exportResult() {
                     <main class="min-w-0">
                         <TabsContent v-for="option in moduleOptions" :key="option.value" :value="option.value"
                             ><SubjectList
+                                :project-id="project?.id"
                                 :subjects="subjects.filter(item => item.module === option.value)"
                                 :forms="forms" /></TabsContent
                         ><TabsContent value="storyboard"

src/features/breakdown/components/SubjectList.vue

@@ -5,7 +5,7 @@ import { EmptyState } from '../../../components/ui'
 import type { SubjectCandidate, SubjectForm } from '../types'
 
 /** 主体列表按稳定 ref 展示；形态以 profileId 关联。 */
-const props = defineProps<{ subjects: SubjectCandidate[]; forms: SubjectForm[] }>()
+const props = defineProps<{ subjects: SubjectCandidate[]; forms: SubjectForm[]; projectId?: string }>()
 const search = ref('')
 const filtered = computed(() =>
     props.subjects.filter(item =>
@@ -34,6 +34,12 @@ const filtered = computed(() =>
                     >
                 </div>
                 <p class="text-sm leading-7">{{ subject.description }}</p>
+                <RouterLink
+                    v-if="projectId"
+                    :to="{ path: `/projects/${projectId}/subject-images`, query: { subjectRef: subject.ref } }"
+                    class="text-button mt-3"
+                    >查看形态图片／生图</RouterLink
+                >
                 <details class="subject-details mt-4">
                     <summary>
                         <ChevronDown :size="14" />外观描述与形态

src/features/projects/ProjectLayout.vue

@@ -1,7 +1,7 @@
 <script setup lang="ts">
 import { computed, provide } from 'vue'
 import { useRoute } from 'vue-router'
-import { ArrowLeft, RefreshCw, FileText, Layers, Camera, LoaderCircle } from '@lucide/vue'
+import { ArrowLeft, RefreshCw, FileText, Layers, Camera, Images, LoaderCircle } from '@lucide/vue'
 import { StatusBadge } from '../../components/ui'
 import { projectContextKey, useProjectData } from './context'
 import { getOperation } from '../workflows/operations'
@@ -45,6 +45,9 @@ const operation = computed(() => getOperation(id.value))
             <RouterLink :to="`/projects/${id}/breakdown`"
                 ><Layers :size="17" />剧本拆解<span class="nav-code">breakdown</span></RouterLink
             >
+            <RouterLink :to="`/projects/${id}/subject-images`"
+                ><Images :size="17" />形态图片<span class="nav-code">subject-image</span></RouterLink
+            >
             <RouterLink :to="`/projects/${id}/storyboard`"
                 ><Camera :size="17" />分镜设计<span class="nav-code">storyboard</span></RouterLink
             >

src/features/storyboard/components/ShotTools.vue

@@ -155,7 +155,10 @@ function exportResult(kind: 'spec' | 'prompt') {
                     后端未返回参考图或缺失记录。
                 </p>
                 <p v-if="references.missing.length" class="mt-3 text-xs text-muted">
-                    请先在后端补齐对应形态的图片，再刷新。本页暂不包含主体图生成。
+                    请在形态图库补齐图片或设置主图，再刷新本镜头参考图。
+                    <RouterLink :to="`/projects/${projectId}/subject-images`" class="text-button ml-2"
+                        >前往形态图片</RouterLink
+                    >
                 </p>
             </template>
         </section>

src/features/storyboard/model.ts

@@ -110,19 +110,5 @@ export function directionLabel(value: string, field?: string): string {
     return labels[value] ?? value
 }
 
-/** 只开放 http(s) 与同源 /storage 路径，拦截模型/数据中的 javascript 或协议相对 URL。 */
-export function referenceImageUrl(value: string): string | null {
-    const raw = value.trim()
-    try {
-        if (raw.startsWith('/storage/') && !raw.includes('\\')) {
-            const api = import.meta.env.VITE_API_BASE_URL || '/api'
-            const origin = /^https?:\/\//i.test(api) ? new URL(api).origin : window.location.origin
-            const url = new URL(raw, origin)
-            return url.pathname.startsWith('/storage/') ? url.href : null
-        }
-        const url = new URL(raw)
-        return ['http:', 'https:'].includes(url.protocol) ? url.href : null
-    } catch {
-        return null
-    }
-}
+/** 保留已有导出路径；图库与分镜共用同一套图片地址校验。 */
+export { referenceImageUrl } from '../../lib/assets'

src/features/subject-images/SubjectImagesPage.vue

@@ -0,0 +1,260 @@
+<script setup lang="ts">
+import { computed, ref, watch } from 'vue'
+import { useRoute } from 'vue-router'
+import { Search, RefreshCw, Download, ImagePlus, ArrowRight } from '@lucide/vue'
+import { AssetImage, EmptyState, StatusBadge } from '../../components/ui'
+import { downloadText } from '../../lib/format'
+import ConfirmAction from '../workflows/ConfirmAction.vue'
+import { useSubjectImages } from './useSubjectImages'
+import { coverImage, hasRunningImages, primaryImage } from './model'
+import type { GenerateFormImageInput, SubjectFormAsset } from './types'
+import GenerateImageDialog from './components/GenerateImageDialog.vue'
+import ImageGalleryDialog from './components/ImageGalleryDialog.vue'
+
+/** 形态图库以正式数据库为准，自动显示已在后端生成的图片。 */
+const route = useRoute()
+const {
+    id,
+    forms,
+    query,
+    operation,
+    session,
+    concurrency,
+    force,
+    blocked,
+    running,
+    batchValid,
+    generate,
+    generateProject
+} = useSubjectImages()
+const search = ref('')
+const module = ref('all')
+const onlyMissing = ref(false)
+const generateOpen = ref(false)
+const galleryOpen = ref(false)
+const generateId = ref('')
+const galleryId = ref('')
+const generateForm = computed(() => forms.value.find(form => form.id === generateId.value) ?? null)
+const galleryForm = computed(() => forms.value.find(form => form.id === galleryId.value) ?? null)
+const completeCount = computed(() => forms.value.filter(form => primaryImage(form.images)).length)
+const filtered = computed(() =>
+    forms.value.filter(
+        form =>
+            (module.value === 'all' || form.subject.module === module.value) &&
+            (!onlyMissing.value || !primaryImage(form.images)) &&
+            `${form.subject.name} ${form.subject.ref} ${form.name}`
+                .toLowerCase()
+                .includes(search.value.trim().toLowerCase())
+    )
+)
+watch(
+    () => route.query.subjectRef,
+    value => {
+        search.value = typeof value === 'string' ? value : ''
+    },
+    { immediate: true }
+)
+
+/** 打开指定正式形态的生图表单。 */
+function openGenerate(form: SubjectFormAsset) {
+    generateId.value = form.id
+    generateOpen.value = true
+}
+/** 查看形态全部候选图和失败记录。 */
+function openGallery(form: SubjectFormAsset) {
+    galleryId.value = form.id
+    galleryOpen.value = true
+}
+/** 本次提交的 ID 和配置由弹窗固定，关闭弹窗不取消后台任务。 */
+function submitGenerate(formId: string, input: GenerateFormImageInput) {
+    void generate(formId, input)
+}
+/** 导出真实批量回执供排障，不构造成功图片地址。 */
+function exportReceipt() {
+    downloadText('subject-images-receipt.json', JSON.stringify(session.value.receipt, null, 2), 'application/json')
+}
+/** 搜索为空时可一次恢复完整图库。 */
+function resetFilters() {
+    search.value = ''
+    module.value = 'all'
+    onlyMissing.value = false
+}
+</script>
+
+<template>
+    <section class="mt-7">
+        <div class="flex flex-wrap items-end justify-between gap-4">
+            <div>
+                <h2 class="text-lg font-semibold">形态图片</h2>
+                <p class="mt-2 text-sm text-muted">为人物、场景和道具准备参考图。已有图片直接从数据库读取。</p>
+            </div>
+            <RouterLink :to="`/projects/${id}/storyboard`" class="text-button"
+                >去检查分镜参考图<ArrowRight :size="14"
+            /></RouterLink>
+        </div>
+        <div class="panel mt-5 p-5">
+            <div class="flex flex-wrap items-center justify-between gap-4">
+                <p class="text-sm">
+                    <strong>{{ completeCount }}</strong> / {{ forms.length }} 个形态已有主图<span
+                        class="ml-3 text-xs text-muted"
+                        >Seedream · 后端配置</span
+                    >
+                </p>
+                <button class="text-button" :disabled="query.loading.value" @click="query.refresh">
+                    <RefreshCw :size="13" :class="{ 'animate-spin': query.loading.value }" />刷新图库
+                </button>
+            </div>
+            <p class="mt-2 text-[11px] leading-6 text-muted">
+                每 6 秒更新图片记录；主图供分镜引用，候选图和历史失败记录均保留。覆盖率不是当前批次进度。
+            </p>
+            <details class="mt-4 border-t border-line pt-4">
+                <summary class="cursor-pointer text-xs font-medium">项目批量生图</summary>
+                <div class="mt-4 flex flex-wrap items-end gap-4">
+                    <label class="w-28"
+                        ><span class="field-label">批量并发</span
+                        ><input
+                            id="image-concurrency"
+                            v-model.number="concurrency"
+                            type="number"
+                            min="1"
+                            step="1"
+                            class="input"
+                            :disabled="operation.pending" /></label
+                    ><label class="flex items-center gap-2 pb-2 text-xs"
+                        ><input
+                            v-model="force"
+                            type="checkbox"
+                            class="accent-accent"
+                            :disabled="operation.pending"
+                        />已有主图也新增候选图</label
+                    ><ConfirmAction
+                        :label="force ? '为全项目新增候选图' : '补齐项目主参考图'"
+                        :disabled="blocked || !batchValid || !forms.length"
+                        acknowledgement
+                        :description="
+                            force
+                                ? '对项目全部形态调用 Seedream 新增一张图片。已有主图不会被替换，生成后可在图片记录中手动选择主图。'
+                                : '仅对没有主图的形态调用 Seedream 生图，已有主图的形态跳过。操作面向整个项目，不受列表筛选影响。'
+                        "
+                        @confirm="generateProject"
+                    />
+                </div>
+                <p v-if="!batchValid" class="mt-2 text-xs text-danger">并发必须是正整数。</p>
+                <p class="mt-3 text-xs leading-6 text-muted">
+                    作用于整个项目，不受下方筛选影响。批量使用后端默认尺寸与形态提示词，不会启动视频生成。
+                </p>
+            </details>
+        </div>
+        <p v-if="query.error.value" class="alert alert-error mt-4" role="alert">
+            {{ query.error.value }}<span v-if="forms.length"> 当前展示上次读取的数据。</span>
+        </p>
+        <p v-if="running" class="alert mt-4" role="status">
+            数据库中仍有排队或生成中的图片，已暂停新的生图操作。长时间无变化时，请先核对后台任务；刷新或关闭页面不会取消任务。
+        </p>
+        <section v-if="session.receipt" class="panel mt-4 p-5" aria-label="生图批量回执">
+            <div class="flex items-center justify-between gap-3">
+                <h3 class="text-sm font-medium">{{ session.receipt.title }} · 本次回执</h3>
+                <button class="text-button" @click="exportReceipt"><Download :size="13" />导出诊断</button>
+            </div>
+            <p class="mt-3 text-xs text-muted">
+                共 {{ session.receipt.result.total }} 个形态 · 本次目标 {{ session.receipt.result.targetCount }} · 生成
+                {{ session.receipt.result.generated }} · 跳过 {{ session.receipt.result.skipped }} · 失败
+                {{ session.receipt.result.failed }}
+            </p>
+            <p v-if="session.receipt.result.failed" class="alert alert-error mt-3" role="alert">
+                部分形态生图失败。已生成的图片保留；先查看失败原因，再按形态重新生图，或关闭新增候选图后补齐缺失主图。
+            </p>
+            <ul class="mt-3 space-y-2 text-xs text-danger">
+                <li v-for="failure in session.receipt.result.failures" :key="failure.subjectFormId">
+                    <code>{{ failure.subjectFormId }}</code
+                    >：{{ failure.error }}
+                </li>
+            </ul>
+            <p class="mt-3 text-[11px] text-muted">回执仅保留于当前浏览器会话；图片以刷新后的数据库记录为准。</p>
+        </section>
+        <div class="my-6 flex flex-wrap items-center gap-4">
+            <label class="search-field"
+                ><Search :size="14" /><input
+                    v-model="search"
+                    placeholder="搜索主体、引用或形态"
+                    aria-label="搜索形态图片" /></label
+            ><select v-model="module" class="input w-auto" aria-label="筛选主体类型">
+                <option value="all">全部类型</option>
+                <option value="character">人物</option>
+                <option value="scene">场景</option>
+                <option value="prop">道具</option></select
+            ><label class="flex items-center gap-2 text-xs"
+                ><input v-model="onlyMissing" type="checkbox" class="accent-accent" />仅看缺少主图</label
+            ><span class="ml-auto text-xs text-muted">{{ filtered.length }} 个形态</span>
+        </div>
+        <p v-if="!query.data.value && query.loading.value" class="py-10 text-sm text-muted" role="status">
+            正在读取形态和已有图片…
+        </p>
+        <div v-else-if="filtered.length" class="form-image-grid">
+            <article v-for="form in filtered" :key="form.id" class="panel form-image-card">
+                <AssetImage
+                    :src="coverImage(form)?.imageUrl"
+                    :alt="`${form.subject.name} · ${form.name}`"
+                    :empty-text="hasRunningImages(form.images) ? '后台正在生成图片' : '尚未生成图片'"
+                />
+                <div class="p-4">
+                    <div class="flex flex-wrap items-center gap-2">
+                        <h3 class="text-sm font-semibold">{{ form.subject.name }}</h3>
+                        <code class="subject-ref">{{ form.subject.ref }}</code>
+                    </div>
+                    <p class="mt-2 text-xs">
+                        {{ form.name }}<span v-if="form.isDefault" class="tag ml-2">默认形态</span>
+                    </p>
+                    <p v-if="form.description" class="mt-3 line-clamp-2 text-xs leading-6 text-muted">
+                        {{ form.description }}
+                    </p>
+                    <div class="mt-3 flex flex-wrap items-center gap-2">
+                        <StatusBadge v-if="hasRunningImages(form.images)" status="generating" /><span
+                            v-if="primaryImage(form.images)"
+                            class="tag"
+                            >主参考图</span
+                        ><span v-else-if="coverImage(form)" class="tag">候选图 · 尚未设主图</span
+                        ><span v-else-if="form.images[0]?.status === 'failed'" class="text-xs text-danger"
+                            >最近一次生成失败</span
+                        ><span class="text-[10px] text-muted">{{ form.images.length }} 条记录</span>
+                    </div>
+                    <p v-if="form.images[0]?.status === 'failed'" class="mt-2 line-clamp-2 text-xs text-danger">
+                        {{ form.images[0].error }}
+                    </p>
+                    <div class="mt-4 flex flex-wrap items-center justify-between gap-2">
+                        <button class="text-button" @click="openGallery(form)">查看图片与记录</button
+                        ><button class="button button-secondary" :disabled="blocked" @click="openGenerate(form)">
+                            <ImagePlus :size="14" />{{ coverImage(form) ? '再生成一张' : '生成图片' }}
+                        </button>
+                    </div>
+                </div>
+            </article>
+        </div>
+        <EmptyState
+            v-else-if="query.data.value"
+            :title="forms.length ? '没有匹配的形态' : '还没有正式形态'"
+            :description="
+                forms.length
+                    ? '调整筛选条件后再查看。批量生图不受筛选影响。'
+                    : '先完成剧本拆解和主体形态持久化，再来生成参考图。'
+            "
+            ><button v-if="forms.length" class="button button-secondary" @click="resetFilters">清除筛选</button
+            ><RouterLink v-else :to="`/projects/${id}/breakdown`" class="button button-secondary"
+                >前往剧本拆解</RouterLink
+            ></EmptyState
+        >
+        <GenerateImageDialog
+            v-model:open="generateOpen"
+            :form="generateForm"
+            :disabled="blocked || !generateForm"
+            @generate="submitGenerate"
+        />
+        <ImageGalleryDialog
+            v-model:open="galleryOpen"
+            :project-id="id"
+            :form="galleryForm"
+            :disabled="blocked || !galleryForm"
+            @changed="query.refresh"
+        />
+    </section>
+</template>

src/features/subject-images/api.ts

@@ -0,0 +1,30 @@
+import { request } from '../../lib/http'
+import type {
+    GenerateFormImageInput,
+    GenerateProjectImagesInput,
+    ImageBatchResult,
+    SubjectFormAsset,
+    SubjectImage
+} from './types'
+
+/** 对正式 SubjectForm 数据库 ID 编码，不接受按名称拼出的路径。 */
+function formPath(id: string) {
+    return `/subject-forms/${encodeURIComponent(id)}/images`
+}
+
+/** 形态图库 API：读取与生成分离，生成不自动重试，不设置客户端短超时。 */
+export const subjectImagesApi = {
+    listForms: (projectId: string, signal?: AbortSignal) =>
+        request<SubjectFormAsset[]>(`/projects/${encodeURIComponent(projectId)}/subject-forms`, { signal }),
+    listImages: (formId: string, signal?: AbortSignal) => request<SubjectImage[]>(formPath(formId), { signal }),
+    generate: (formId: string, input: GenerateFormImageInput) =>
+        request<SubjectImage | null>(formPath(formId), { method: 'POST', body: input, timeoutMs: 0 }),
+    generateProject: (projectId: string, input: GenerateProjectImagesInput) =>
+        request<ImageBatchResult>(`/projects/${encodeURIComponent(projectId)}/subject-images/generate`, {
+            method: 'POST',
+            body: input,
+            timeoutMs: 0
+        }),
+    setPrimary: (formId: string, imageId: string) =>
+        request<SubjectImage | null>(`${formPath(formId)}/${encodeURIComponent(imageId)}/primary`, { method: 'PUT' })
+}

src/features/subject-images/components/GenerateImageDialog.vue

@@ -0,0 +1,120 @@
+<script setup lang="ts">
+import { computed, ref, watch } from 'vue'
+import { AppDialog } from '../../../components/ui'
+import { primaryImage, validImageSize } from '../model'
+import type { GenerateFormImageInput, SubjectFormAsset } from '../types'
+
+/** 单形态生成确认：尺寸与自定义 Prompt 只用于本次请求。 */
+const props = defineProps<{ form: SubjectFormAsset | null; disabled: boolean }>()
+const open = defineModel<boolean>('open', { default: false })
+const emit = defineEmits<{ generate: [formId: string, input: GenerateFormImageInput] }>()
+const prompt = ref('')
+const width = ref<number | ''>('')
+const height = ref<number | ''>('')
+const setPrimary = ref(false)
+const acknowledged = ref(false)
+const dimensionsValid = computed(() => validImageSize(width.value, height.value))
+const hasPrompt = computed(
+    () => !!(prompt.value.trim() || (props.form?.generationPrompt ?? props.form?.appearancePrompt ?? '').trim())
+)
+
+/** 每次打开重新建立表单，避免把另一形态的 Prompt 或尺寸带入请求。 */
+watch(
+    () => [open.value, props.form?.id],
+    () => {
+        prompt.value = ''
+        width.value = ''
+        height.value = ''
+        acknowledged.value = false
+        setPrimary.value = props.form ? !primaryImage(props.form.images) : false
+    },
+    { immediate: true }
+)
+
+/** 完成前置确认后只发出一次明确的生图请求，不触发整条生产流程。 */
+function submit() {
+    if (!props.form || props.disabled || !dimensionsValid.value || !hasPrompt.value || !acknowledged.value) return
+    emit('generate', props.form.id, {
+        provider: 'seedream',
+        setPrimary: setPrimary.value,
+        ...(prompt.value.trim() ? { prompt: prompt.value.trim() } : {}),
+        ...(width.value !== '' && height.value !== '' ? { width: width.value, height: height.value } : {})
+    })
+    open.value = false
+}
+</script>
+
+<template>
+    <AppDialog
+        v-model:open="open"
+        title="生成形态图片"
+        :description="`${form?.subject.name || ''} · ${form?.name || ''}`"
+    >
+        <form class="mt-6 space-y-5" @submit.prevent="submit">
+            <p class="alert text-xs">使用后端配置的 Seedream 模型，可能产生费用。每次新增一张图片，不删除历史结果。</p>
+            <label class="block"
+                ><span class="field-label">自定义提示词（可选，仅用于本次）</span
+                ><textarea
+                    id="image-prompt"
+                    v-model="prompt"
+                    class="input min-h-28"
+                    placeholder="留空使用已保存的 generationPrompt，其次使用 appearancePrompt"
+                />
+            </label>
+            <details v-if="form?.generationPrompt || form?.appearancePrompt" class="text-xs">
+                <summary class="cursor-pointer text-muted">查看后端默认提示词</summary>
+                <p class="mt-3 whitespace-pre-wrap leading-6">{{ form.generationPrompt ?? form.appearancePrompt }}</p>
+            </details>
+            <div class="grid grid-cols-2 gap-4">
+                <label
+                    ><span class="field-label">宽度（px）</span
+                    ><input
+                        id="image-width"
+                        v-model.number="width"
+                        class="input"
+                        type="number"
+                        min="1"
+                        step="1"
+                        placeholder="后端默认" /></label
+                ><label
+                    ><span class="field-label">高度（px）</span
+                    ><input
+                        id="image-height"
+                        v-model.number="height"
+                        class="input"
+                        type="number"
+                        min="1"
+                        step="1"
+                        placeholder="后端默认"
+                /></label>
+            </div>
+            <p class="text-xs text-muted">宽高同时留空时使用后端默认 2K；自定义尺寸须满足模型限制。</p>
+            <p v-if="!dimensionsValid" class="text-xs text-danger" role="alert">宽高需要同时填写正整数，或同时留空。</p>
+            <p v-if="!hasPrompt" class="text-xs text-danger" role="alert">当前形态没有可用提示词，请填写本次提示词。</p>
+            <label class="flex gap-2 text-sm"
+                ><input v-model="setPrimary" type="checkbox" class="accent-accent" />生成成功后设为此形态主参考图</label
+            >
+            <p v-if="setPrimary && form && primaryImage(form.images)" class="text-xs text-muted">
+                会替换当前主图标记，旧图片保留。已有视频提示词不会自动更新。
+            </p>
+            <label class="flex items-start gap-2 text-xs leading-6"
+                ><input
+                    id="confirm-image-cost"
+                    v-model="acknowledged"
+                    type="checkbox"
+                    class="mt-1.5 accent-accent"
+                />我已确认后台没有同项目的生成任务，了解本次会调用模型并产生费用。</label
+            >
+            <div class="dialog-footer">
+                <button type="button" class="button button-secondary" @click="open = false">取消</button
+                ><button
+                    type="submit"
+                    class="button button-primary"
+                    :disabled="disabled || !acknowledged || !dimensionsValid || !hasPrompt"
+                >
+                    确认生成图片
+                </button>
+            </div>
+        </form>
+    </AppDialog>
+</template>

src/features/subject-images/components/ImageGalleryDialog.vue

@@ -0,0 +1,180 @@
+<script setup lang="ts">
+import { computed, ref, watch } from 'vue'
+import { ExternalLink, RefreshCw } from '@lucide/vue'
+import { AppDialog, AssetImage, StatusBadge } from '../../../components/ui'
+import { usePolling } from '../../../composables/usePolling'
+import { referenceImageUrl } from '../../../lib/assets'
+import { formatDate } from '../../../lib/format'
+import { getOperation, runOperation } from '../../workflows/operations'
+import { subjectImagesApi } from '../api'
+import { hasRunningImages, primaryImage } from '../model'
+import type { SubjectFormAsset } from '../types'
+
+/** 形态图片历史与主图选择；历史查询只在弹窗打开时进行。 */
+const props = defineProps<{ projectId: string; form: SubjectFormAsset | null; disabled: boolean }>()
+const open = defineModel<boolean>('open', { default: false })
+const emit = defineEmits<{ changed: [] }>()
+const selectedId = ref('')
+const confirming = ref(false)
+const key = computed(() => (open.value ? (props.form?.id ?? '') : ''))
+const query = usePolling(key, async (id, signal) => {
+    if (!id) return null
+    const rows = await subjectImagesApi.listImages(id, signal)
+    if (rows.some(image => image.subjectFormId !== id)) throw new Error('图片记录与当前形态不匹配，请刷新后重试。')
+    return rows
+})
+const images = computed(() => query.data.value ?? [])
+const selected = computed(
+    () => images.value.find(image => image.id === selectedId.value) ?? primaryImage(images.value) ?? images.value[0]
+)
+const operation = computed(() => getOperation(props.projectId))
+const canSetPrimary = computed(
+    () =>
+        !props.disabled &&
+        !operation.value.pending &&
+        !query.error.value &&
+        !hasRunningImages(images.value) &&
+        selected.value?.status === 'completed' &&
+        !!selected.value.imageUrl &&
+        !selected.value.isPrimary
+)
+const imageUrl = computed(() => (selected.value?.imageUrl ? referenceImageUrl(selected.value.imageUrl) : null))
+watch(key, () => {
+    selectedId.value = ''
+    confirming.value = false
+})
+watch(
+    () => selected.value?.id,
+    () => {
+        confirming.value = false
+    }
+)
+
+/** 切换主图只调用 PUT，不调用模型；目标在提交时固定。 */
+async function choosePrimary() {
+    if (!canSetPrimary.value || !confirming.value || !selected.value || !props.form) return
+    const formId = props.form.id
+    const imageId = selected.value.id
+    const projectId = props.projectId
+    confirming.value = false
+    await runOperation(projectId, `设置 ${props.form.subject.name} · ${props.form.name} 主图`, async () => {
+        const result = await subjectImagesApi.setPrimary(formId, imageId)
+        if (
+            !result ||
+            result.id !== imageId ||
+            result.subjectFormId !== formId ||
+            !result.isPrimary ||
+            result.status !== 'completed'
+        )
+            throw new Error('接口未确认主图切换，请先刷新核对后端状态。')
+    })
+    emit('changed')
+    await query.refresh()
+}
+</script>
+
+<template>
+    <AppDialog
+        v-model:open="open"
+        :title="`${form?.subject.name || ''} · ${form?.name || '形态图片'}`"
+        description="主参考图会供后续分镜使用。切换主图不会重新生成已有视频提示词。"
+        wide
+    >
+        <div class="mt-5 flex items-center justify-between">
+            <span class="text-xs text-muted">{{ images.length }} 条图片记录</span
+            ><button class="text-button" :disabled="query.loading.value" @click="query.refresh">
+                <RefreshCw :size="13" :class="{ 'animate-spin': query.loading.value }" />刷新图片
+            </button>
+        </div>
+        <p v-if="query.error.value" class="alert alert-error mt-3" role="alert">
+            {{ query.error.value }} 当前保留上次读取的记录，主图操作已暂停。
+        </p>
+        <p v-if="operation.error" class="alert alert-error mt-3" role="alert">{{ operation.error }}</p>
+        <div v-if="selected" class="mt-4">
+            <AssetImage
+                :key="selected.id"
+                :src="selected.status === 'completed' ? selected.imageUrl : null"
+                :alt="`${form?.subject.name} · ${form?.name}`"
+                :empty-text="selected.status === 'failed' ? '本次生成失败' : '等待后端生成结果'"
+                class="asset-image-preview"
+            />
+            <div class="mt-4 flex flex-wrap items-center justify-between gap-3">
+                <div class="flex items-center gap-2">
+                    <StatusBadge :status="selected.status" /><span v-if="selected.isPrimary" class="tag"
+                        >当前主参考图</span
+                    ><span class="text-xs text-muted">{{ selected.width || '—' }} × {{ selected.height || '—' }}</span>
+                </div>
+                <a
+                    v-if="imageUrl && selected.status === 'completed'"
+                    :href="imageUrl"
+                    target="_blank"
+                    rel="noopener noreferrer"
+                    class="text-button"
+                    >打开原图<ExternalLink :size="12"
+                /></a>
+            </div>
+            <p v-if="selected.error" class="alert alert-error mt-3 whitespace-pre-wrap" role="alert">
+                {{ selected.error }}
+            </p>
+            <div class="mt-4 flex flex-wrap items-center gap-3">
+                <button
+                    v-if="!selected.isPrimary"
+                    class="button button-secondary"
+                    :disabled="!canSetPrimary"
+                    @click="confirming = true"
+                >
+                    设为主参考图</button
+                ><span class="text-[11px] text-muted"
+                    >{{ selected.provider }} · {{ selected.model }} · {{ formatDate(selected.createdAt) }}</span
+                >
+            </div>
+            <div v-if="confirming" class="alert mt-3">
+                <p class="text-xs">确认将这张图用于此形态后续的分镜参考？旧主图仍会保留，此操作不调用模型。</p>
+                <div class="mt-3 flex gap-3">
+                    <button class="button button-primary" :disabled="!canSetPrimary" @click="choosePrimary">
+                        确认切换主图</button
+                    ><button class="text-button" @click="confirming = false">取消</button>
+                </div>
+            </div>
+            <details v-if="selected.prompt" class="mt-4 text-xs">
+                <summary class="cursor-pointer text-muted">查看本次实际提示词</summary>
+                <p class="mt-3 whitespace-pre-wrap leading-6">{{ selected.prompt }}</p>
+                <p v-if="selected.negativePrompt" class="mt-3 whitespace-pre-wrap leading-6">
+                    Negative prompt: {{ selected.negativePrompt }}
+                </p>
+            </details>
+            <p class="mt-3 break-all font-mono text-[10px] text-muted">Image ID · {{ selected.id }}</p>
+            <div class="image-history mt-5" aria-label="图片历史">
+                <button
+                    v-for="image in images"
+                    :key="image.id"
+                    class="image-history-item"
+                    :class="{ selected: image.id === selected.id }"
+                    :aria-label="`查看图片 ${image.id}`"
+                    :aria-pressed="image.id === selected.id"
+                    @click="selectedId = image.id"
+                >
+                    <AssetImage
+                        :src="image.status === 'completed' ? image.imageUrl : null"
+                        :alt="`${form?.name}候选图`"
+                        :retryable="false"
+                        :empty-text="
+                            { pending: '排队中', generating: '生成中', failed: '生成失败', completed: '图片缺失' }[
+                                image.status
+                            ]
+                        "
+                    /><span class="mt-1 block text-[10px]">{{
+                        image.isPrimary
+                            ? '主参考图'
+                            : { completed: '候选图', failed: '失败记录', pending: '排队中', generating: '生成中' }[
+                                  image.status
+                              ]
+                    }}</span>
+                </button>
+            </div>
+        </div>
+        <p v-else class="py-10 text-center text-sm text-muted">
+            {{ query.loading.value ? '正在读取图片记录…' : '此形态尚无图片记录，可以先生成一张。' }}
+        </p>
+    </AppDialog>
+</template>

src/features/subject-images/images.test.ts

@@ -0,0 +1,172 @@
+import { flushPromises, mount, type VueWrapper } from '@vue/test-utils'
+import { afterEach, describe, expect, it, vi } from 'vitest'
+import { AssetImage } from '../../components/ui'
+import { getOperation } from '../workflows/operations'
+import GenerateImageDialog from './components/GenerateImageDialog.vue'
+import ImageGalleryDialog from './components/ImageGalleryDialog.vue'
+import { coverImage, hasRunningImages, primaryImage, validImageSize } from './model'
+import { formFixture, imageFixture } from './testing/fixtures'
+
+let wrapper: VueWrapper | undefined
+/** Reka 将弹窗挂载到 body，需要从实际弹窗找到按钮。 */
+function button(label: string): HTMLButtonElement {
+    const element = [...document.querySelectorAll('button')].find(item => item.textContent?.trim() === label)
+    if (!element) throw new Error('找不到按钮：' + label)
+    return element
+}
+afterEach(() => {
+    wrapper?.unmount()
+    wrapper = undefined
+    document.body.innerHTML = ''
+    vi.unstubAllGlobals()
+    Object.assign(getOperation('gallery-test'), { pending: false, label: '', error: '', notice: '' })
+})
+
+describe('形态图片选择与操作', () => {
+    it('已有主图优先于新候选图和失败记录，无主图才回退到成功候选图', () => {
+        const form = formFixture()
+        form.images.unshift(
+            imageFixture({ id: 'failed', isPrimary: false, status: 'failed', imageUrl: null }),
+            imageFixture({ id: 'candidate', isPrimary: false })
+        )
+        expect(coverImage(form)?.id).toBe('image-db-1')
+        form.images.pop()
+        expect(primaryImage(form.images)).toBeUndefined()
+        expect(coverImage(form)?.id).toBe('candidate')
+        expect(hasRunningImages(form.images)).toBe(false)
+        form.images.push(imageFixture({ status: 'generating', isPrimary: false }))
+        expect(hasRunningImages(form.images)).toBe(true)
+    })
+
+    it('尺寸可同时留空，但不接受单边、零、负数或非整数', () => {
+        expect(validImageSize('', '')).toBe(true)
+        expect(validImageSize(2048, 2048)).toBe(true)
+        for (const [width, height] of [
+            [1024, ''],
+            ['', 1024],
+            [0, 1024],
+            [-1, 1024],
+            [10.5, 1024]
+        ] as const)
+            expect(validImageSize(width, height)).toBe(false)
+    })
+
+    it('生图确认发送 Seedream 与正式形态 ID，默认不替换已有主图', async () => {
+        wrapper = mount(GenerateImageDialog, {
+            attachTo: document.body,
+            props: { open: true, form: formFixture(), disabled: false }
+        })
+        await flushPromises()
+        expect(button('确认生成图片').disabled).toBe(true)
+        document.querySelector<HTMLInputElement>('#confirm-image-cost')!.click()
+        await flushPromises()
+        button('确认生成图片').click()
+        await flushPromises()
+        expect(wrapper.emitted('generate')).toEqual([['form-db-1', { provider: 'seedream', setPrimary: false }]])
+    })
+
+    it('新形态默认设主图，填写一侧尺寸时不能提交，切换形态清空自定义提示词', async () => {
+        const form = formFixture()
+        form.images = []
+        wrapper = mount(GenerateImageDialog, { attachTo: document.body, props: { open: true, form, disabled: false } })
+        await flushPromises()
+        const width = document.querySelector<HTMLInputElement>('#image-width')!
+        width.value = '2048'
+        width.dispatchEvent(new Event('input', { bubbles: true }))
+        document.querySelector<HTMLInputElement>('#confirm-image-cost')!.click()
+        await flushPromises()
+        expect(button('确认生成图片').disabled).toBe(true)
+        await wrapper.setProps({ form: { ...form, id: 'form-db-2' } })
+        await flushPromises()
+        expect(width.value).toBe('')
+        expect(document.querySelector<HTMLInputElement>('#confirm-image-cost')!.checked).toBe(false)
+        document.querySelector<HTMLInputElement>('#confirm-image-cost')!.click()
+        await flushPromises()
+        button('确认生成图片').click()
+        await flushPromises()
+        expect(wrapper.emitted('generate')).toEqual([['form-db-2', { provider: 'seedream', setPrimary: true }]])
+    })
+
+    it('图库不自动生图，主图切换经确认后 PUT，失败图片不能设主图', async () => {
+        let primary = false
+        const fetcher = vi.fn<typeof fetch>().mockImplementation(async (_url, init) => {
+            if (init?.method === 'PUT') {
+                primary = true
+                return new Response(JSON.stringify({ data: imageFixture() }))
+            }
+            return new Response(
+                JSON.stringify({
+                    data: [
+                        imageFixture({ isPrimary: primary }),
+                        imageFixture({
+                            id: 'failed',
+                            isPrimary: false,
+                            status: 'failed',
+                            imageUrl: null,
+                            error: '供应商拒绝请求'
+                        })
+                    ]
+                })
+            )
+        })
+        vi.stubGlobal('fetch', fetcher)
+        wrapper = mount(ImageGalleryDialog, {
+            attachTo: document.body,
+            props: { projectId: 'gallery-test', form: formFixture('gallery-test'), open: true, disabled: false }
+        })
+        await flushPromises()
+        expect(fetcher.mock.calls).toHaveLength(1)
+        expect(document.body.textContent).toContain('<script>模型提示词</script>')
+        expect(document.querySelector('[role="dialog"] script')).toBeNull()
+        button('设为主参考图').click()
+        await flushPromises()
+        expect(fetcher.mock.calls).toHaveLength(1)
+        button('确认切换主图').click()
+        await flushPromises()
+        expect(fetcher.mock.calls.find(([, init]) => init?.method === 'PUT')?.[0]).toBe(
+            '/api/subject-forms/form-db-1/images/image-db-1/primary'
+        )
+        expect(wrapper.emitted('changed')).toHaveLength(1)
+        document.querySelector<HTMLButtonElement>('[aria-label="查看图片 failed"]')!.click()
+        await flushPromises()
+        expect(button('设为主参考图').disabled).toBe(true)
+        expect(document.body.textContent).toContain('供应商拒绝请求')
+    })
+
+    it('关闭图片弹窗取消查询，迟到的旧形态响应不会污染再次打开的形态', async () => {
+        let finish!: (response: Response) => void
+        const fetcher = vi
+            .fn<typeof fetch>()
+            .mockImplementationOnce(
+                () =>
+                    new Promise(resolve => {
+                        finish = resolve
+                    })
+            )
+            .mockImplementation(async () => new Response('{"data":[]}'))
+        vi.stubGlobal('fetch', fetcher)
+        wrapper = mount(ImageGalleryDialog, {
+            attachTo: document.body,
+            props: { projectId: 'gallery-test', form: formFixture('gallery-test'), open: true, disabled: false }
+        })
+        await flushPromises()
+        await wrapper.setProps({ open: false })
+        expect(fetcher.mock.calls[0]?.[1]?.signal?.aborted).toBe(true)
+        await wrapper.setProps({ open: true, form: { ...formFixture('gallery-test'), id: 'form-db-2' } })
+        await flushPromises()
+        finish(new Response(JSON.stringify({ data: [imageFixture()] })))
+        await flushPromises()
+        expect(document.body.textContent).not.toContain('Image ID · image-db-1')
+        expect(document.body.textContent).toContain('此形态尚无图片记录')
+    })
+
+    it('图片加载失败有占位，地址变化可恢复，危险协议不会写入 img', async () => {
+        wrapper = mount(AssetImage, { props: { src: '/storage/a.png', alt: '形态主图' } })
+        await wrapper.get('img').trigger('error')
+        expect(wrapper.text()).toContain('图片无法加载')
+        await wrapper.setProps({ src: '/storage/b.png' })
+        expect(wrapper.get('img').attributes('src')).toContain('/storage/b.png')
+        await wrapper.setProps({ src: 'javascript:alert(1)' })
+        expect(wrapper.find('img').exists()).toBe(false)
+    })
+})

src/features/subject-images/index.ts

@@ -0,0 +1,2 @@
+export { subjectImagesApi } from './api'
+export type { SubjectFormAsset, SubjectImage, GenerateFormImageInput, ImageBatchResult } from './types'

src/features/subject-images/model.ts

@@ -0,0 +1,31 @@
+import { reactive } from 'vue'
+import type { ImageSession, SubjectFormAsset, SubjectImage } from './types'
+
+/** 会话回执不当作后端持久化任务状态。 */
+const sessions = reactive<Record<string, ImageSession>>({})
+
+/** 获取指定项目的生图回执，避免切换路由后丢失长请求结果。 */
+export function getImageSession(projectId: string): ImageSession {
+    return (sessions[projectId] ??= { receipt: null })
+}
+
+/** 主图必须是后端已完成且显式标记 primary 的图片。 */
+export function primaryImage(images: SubjectImage[]): SubjectImage | undefined {
+    return images.find(image => image.isPrimary && image.status === 'completed')
+}
+
+/** 列表优先展示主图，否则展示最新成功候选图；失败记录不会遮住可用图片。 */
+export function coverImage(form: SubjectFormAsset): SubjectImage | undefined {
+    return primaryImage(form.images) ?? form.images.find(image => image.status === 'completed' && image.imageUrl)
+}
+
+/** 查询发现后台仍在运行时也阻止重复提交，不只依赖本地 loading。 */
+export function hasRunningImages(images: SubjectImage[]): boolean {
+    return images.some(image => image.status === 'pending' || image.status === 'generating')
+}
+
+/** 图片尺寸必须成对提供；空值表示由后端使用默认尺寸。 */
+export function validImageSize(width: number | '', height: number | ''): boolean {
+    if (width === '' && height === '') return true
+    return Number.isSafeInteger(width) && Number(width) > 0 && Number.isSafeInteger(height) && Number(height) > 0
+}

src/features/subject-images/testing/fixtures.ts

@@ -0,0 +1,36 @@
+import type { SubjectFormAsset, SubjectImage } from '../types'
+
+/** 仅在测试使用的数据库图片 fixture，模拟已保存的本地资源地址。 */
+export function imageFixture(overrides: Partial<SubjectImage> = {}): SubjectImage {
+    return {
+        id: 'image-db-1',
+        subjectFormId: 'form-db-1',
+        provider: 'seedream',
+        model: 'configured-model',
+        imageUrl: '/storage/subjects/project/form/image.png',
+        width: 2048,
+        height: 2048,
+        status: 'completed',
+        isPrimary: true,
+        error: null,
+        createdAt: '2026-08-28T08:00:00Z',
+        updatedAt: '2026-08-28T08:00:00Z',
+        prompt: '<script>模型提示词</script>',
+        ...overrides
+    }
+}
+
+/** 正式形态 ID 与领域 formId 分开；测试禁止以 checkpoint 编号调用生图。 */
+export function formFixture(projectId = 'page-test-project'): SubjectFormAsset {
+    return {
+        id: 'form-db-1',
+        subjectId: 'subject-db-1',
+        name: '日常造型',
+        description: '深色外套',
+        appearancePrompt: 'a person in a dark coat',
+        generationPrompt: null,
+        isDefault: true,
+        subject: { id: 'subject-db-1', projectId, ref: '@CH0001', name: '林知夏', module: 'character' },
+        images: [imageFixture()]
+    }
+}

src/features/subject-images/types.ts

@@ -0,0 +1,65 @@
+/** 后端图片记录的持久化状态，不以 HTTP 请求完成代替生成成功。 */
+export type SubjectImageStatus = 'pending' | 'generating' | 'completed' | 'failed'
+
+/** 图库列表返回的图片元数据；正式图片 ID 与形态 ID 均来自数据库。 */
+export interface SubjectImage {
+    id: string
+    subjectFormId: string
+    provider: string
+    model: string
+    imageUrl: string | null
+    width: number | null
+    height: number | null
+    status: SubjectImageStatus
+    isPrimary: boolean
+    error: string | null
+    createdAt: string
+    updatedAt: string
+    /** 详情查询才返回实际使用的 Prompt；列表不包含长文本。 */
+    prompt?: string
+    negativePrompt?: string | null
+}
+
+/** 正式 SubjectForm 及其所属主体、图片，不使用 checkpoint 的领域 formId。 */
+export interface SubjectFormAsset {
+    id: string
+    subjectId: string
+    name: string
+    description: string | null
+    appearancePrompt: string | null
+    generationPrompt: string | null
+    isDefault: boolean
+    subject: { id: string; projectId: string; ref: string; name: string; module: string }
+    images: SubjectImage[]
+}
+
+/** 单形态生图配置，明确指定 Seedream，避免后端单图入口默认使用 mock。 */
+export interface GenerateFormImageInput {
+    provider: 'seedream'
+    prompt?: string
+    width?: number
+    height?: number
+    setPrimary: boolean
+}
+
+/** 批量生图只接受这些配置；不能把单张尺寸配置误传给批量接口。 */
+export interface GenerateProjectImagesInput {
+    provider: 'seedream'
+    concurrency: number
+    force: boolean
+}
+
+/** HTTP 200 也可能包含部分失败；此响应不返回成功图片列表。 */
+export interface ImageBatchResult {
+    total: number
+    targetCount: number
+    generated: number
+    skipped: number
+    failed: number
+    failures: { subjectFormId: string; error: string }[]
+}
+
+/** 长请求的回执按项目隔离，在会话内切换页面后仍可查看。 */
+export interface ImageSession {
+    receipt: { title: string; result: ImageBatchResult } | null
+}

src/features/subject-images/useSubjectImages.ts

@@ -0,0 +1,104 @@
+import { computed, ref } from 'vue'
+import { usePolling } from '../../composables/usePolling'
+import { ApiError } from '../../lib/http'
+import { useProjectContext } from '../projects/context'
+import { getOperation, runOperation } from '../workflows/operations'
+import { workflowCheckpoints } from '../workflows/selectors'
+import { subjectImagesApi } from './api'
+import { getImageSession, hasRunningImages } from './model'
+import type { GenerateFormImageInput } from './types'
+
+/** 形态图库统一读取数据库记录并管理单图／批量长请求，不调用生产流程。 */
+export function useSubjectImages() {
+    const context = useProjectContext()
+    const id = computed(() => context.project.value?.id ?? '')
+    const concurrency = ref(2)
+    const force = ref(false)
+    const query = usePolling(id, async (projectId, signal) => {
+        if (!projectId) return []
+        try {
+            const forms = await subjectImagesApi.listForms(projectId, signal)
+            if (
+                forms.some(
+                    form =>
+                        form.subject.projectId !== projectId ||
+                        form.images.some(image => image.subjectFormId !== form.id)
+                )
+            )
+                throw new Error('形态图库返回了不匹配的项目或图片，请刷新后重试。')
+            return forms
+        } catch (error) {
+            if (error instanceof ApiError && error.status === 404)
+                throw new Error(
+                    '后端尚未提供形态图库查询，请更新后端 dev 并重启服务（GET /projects/:id/subject-forms）。',
+                    { cause: error }
+                )
+            throw error
+        }
+    })
+    const forms = computed(() => query.data.value ?? [])
+    const operation = computed(() => getOperation(id.value))
+    const session = computed(() => getImageSession(id.value))
+    const running = computed(() => forms.value.some(form => hasRunningImages(form.images)))
+    const breakdownRunning = computed(
+        () =>
+            workflowCheckpoints(context.checkpoints.value, 'breakdown').at(-1)?.state.workflowExecution?.status ===
+            'running'
+    )
+    const blocked = computed(
+        () =>
+            operation.value.pending ||
+            !!context.error.value ||
+            !!query.error.value ||
+            query.data.value === null ||
+            running.value ||
+            breakdownRunning.value ||
+            context.project.value?.status === 'generating'
+    )
+    const batchValid = computed(() => Number.isSafeInteger(concurrency.value) && concurrency.value > 0)
+
+    /** 捕获正式 Form ID，生成后重新读取图片；刷新失败也不自动重发有费用的请求。 */
+    async function generate(formId: string, input: GenerateFormImageInput) {
+        if (blocked.value) return
+        const form = forms.value.find(item => item.id === formId)
+        if (!form) return
+        const projectId = id.value
+        await runOperation(projectId, `生成 ${form.subject.name} · ${form.name} 图片`, async () => {
+            const image = await subjectImagesApi.generate(formId, input)
+            if (!image || image.subjectFormId !== formId || image.status !== 'completed' || !image.imageUrl)
+                throw new Error(image?.error || '接口未返回已完成的图片，请先刷新图片记录核对后端状态。')
+        })
+        await query.refresh()
+    }
+
+    /** 批量始终面向整个项目，不受页面筛选影响；force 仅新增候选图，不替换已有主图。 */
+    async function generateProject() {
+        if (blocked.value || !batchValid.value || !forms.value.length) return
+        const projectId = id.value
+        const input = { provider: 'seedream' as const, concurrency: concurrency.value, force: force.value }
+        const target = getImageSession(projectId)
+        target.receipt = null
+        await runOperation(projectId, '批量生成项目形态图片', async () => {
+            target.receipt = {
+                title: input.force ? '为全项目新增候选图' : '补齐项目主参考图',
+                result: await subjectImagesApi.generateProject(projectId, input)
+            }
+        })
+        await query.refresh()
+    }
+
+    return {
+        id,
+        forms,
+        query,
+        operation,
+        session,
+        concurrency,
+        force,
+        blocked,
+        running,
+        batchValid,
+        generate,
+        generateProject
+    }
+}

src/features/workflows/pages.test.ts

@@ -19,6 +19,9 @@ import {
 } from '../storyboard/testing/fixtures'
 import { getStoryboardSession } from '../storyboard/model'
 import { getOperation } from './operations'
+import SubjectImagesPage from '../subject-images/SubjectImagesPage.vue'
+import { formFixture, imageFixture } from '../subject-images/testing/fixtures'
+import { getImageSession } from '../subject-images/model'
 
 /** 模拟 API 响应仅用于测试，不会打包进生产页面。 */
 const fixture: ProjectDetail = {
@@ -81,16 +84,133 @@ function mountStoryboard() {
     return provided
 }
 
+/** 形态图库使用真实路由解析来自主体列表的筛选参数。 */
+async function mountSubjectImages() {
+    const router = createRouter({
+        history: createMemoryHistory(),
+        routes: [{ path: '/projects/:projectId/:workspace', component: SubjectImagesPage }]
+    })
+    await router.push(`/projects/${fixture.id}/subject-images`)
+    wrapper = mount(SubjectImagesPage, {
+        attachTo: document.body,
+        global: { plugins: [router], provide: { [projectContextKey as symbol]: context() } }
+    })
+    await flushPromises()
+}
+
 afterEach(() => {
     wrapper?.unmount()
     wrapper = undefined
     document.body.innerHTML = ''
     vi.unstubAllGlobals()
     Object.assign(getStoryboardSession(fixture.id), { receipt: null, prompts: {} })
+    getImageSession(fixture.id).receipt = null
     Object.assign(getOperation(fixture.id), { pending: false, label: '', error: '', notice: '' })
 })
 
 describe('工作台页面交互', () => {
+    it('形态图库自动显示后端已有图片，不需要再次生图；筛选只改变显示', async () => {
+        const fetcher = vi
+            .fn<typeof fetch>()
+            .mockImplementation(async () => new Response(JSON.stringify({ data: [formFixture()] })))
+        vi.stubGlobal('fetch', fetcher)
+        await mountSubjectImages()
+        expect(wrapper!.get('.form-image-card img').attributes('src')).toContain(
+            '/storage/subjects/project/form/image.png'
+        )
+        expect(wrapper!.text()).toContain('林知夏')
+        expect(wrapper!.text()).toContain('主参考图')
+        expect(fetcher.mock.calls.map(([url, init]) => [url, init?.method])).toEqual([
+            ['/api/projects/page-test-project/subject-forms', 'GET']
+        ])
+        await wrapper!.get('[aria-label="搜索形态图片"]').setValue('不存在的形态')
+        expect(wrapper!.text()).toContain('没有匹配的形态')
+        button('清除筛选').click()
+        await flushPromises()
+        expect(wrapper!.findAll('.form-image-card')).toHaveLength(1)
+    })
+
+    it('形态单图确认后发送正式 ID 并刷新候选图，不会调用 production/start', async () => {
+        const form = formFixture()
+        const fetcher = vi.fn<typeof fetch>().mockImplementation(async (_url, init) => {
+            if (init?.method === 'POST') {
+                const candidate = imageFixture({ id: 'new-candidate', isPrimary: false, imageUrl: '/storage/new.png' })
+                form.images.unshift(candidate)
+                return new Response(JSON.stringify({ data: candidate }))
+            }
+            return new Response(JSON.stringify({ data: [form] }))
+        })
+        vi.stubGlobal('fetch', fetcher)
+        await mountSubjectImages()
+        button('再生成一张').click()
+        await flushPromises()
+        document.querySelector<HTMLInputElement>('#confirm-image-cost')!.click()
+        await flushPromises()
+        document.querySelector('form')!.dispatchEvent(new Event('submit', { bubbles: true, cancelable: true }))
+        await flushPromises()
+        const post = fetcher.mock.calls.find(([, init]) => init?.method === 'POST')!
+        expect(post[0]).toBe('/api/subject-forms/form-db-1/images')
+        expect(JSON.parse(post[1]!.body as string)).toEqual({ provider: 'seedream', setPrimary: false })
+        expect(wrapper!.text()).toContain('2 条记录')
+        expect(wrapper!.get('.form-image-card img').attributes('src')).not.toContain('/storage/new.png')
+        expect(fetcher.mock.calls.some(([url]) => String(url).includes('production/start'))).toBe(false)
+    })
+
+    it('批量生图保留部分失败回执，默认跳过有主图项，筛选不会成为批量参数', async () => {
+        const fetcher = vi.fn<typeof fetch>().mockImplementation(
+            async (_url, init) =>
+                new Response(
+                    JSON.stringify({
+                        data:
+                            init?.method === 'POST'
+                                ? {
+                                      total: 2,
+                                      targetCount: 1,
+                                      generated: 0,
+                                      skipped: 1,
+                                      failed: 1,
+                                      failures: [{ subjectFormId: 'form-db-2', error: '服务暂不可用' }]
+                                  }
+                                : [formFixture()]
+                    })
+                )
+        )
+        vi.stubGlobal('fetch', fetcher)
+        await mountSubjectImages()
+        await wrapper!.get('[aria-label="搜索形态图片"]').setValue('@CH0001')
+        await confirmGeneration('补齐项目主参考图')
+        expect(wrapper!.get('[aria-label="生图批量回执"]').text()).toContain('部分形态生图失败')
+        expect(wrapper!.text()).toContain('服务暂不可用')
+        const post = fetcher.mock.calls.find(([, init]) => init?.method === 'POST')!
+        expect(JSON.parse(post[1]!.body as string)).toEqual({ provider: 'seedream', concurrency: 2, force: false })
+        await wrapper!.get('.panel details input[type="checkbox"]').setValue(true)
+        await confirmGeneration('为全项目新增候选图')
+        const lastPost = fetcher.mock.calls.filter(([, init]) => init?.method === 'POST').at(-1)!
+        expect(JSON.parse(lastPost[1]!.body as string).force).toBe(true)
+    })
+
+    it('后端持久化生成中或查询失败时暂停生图，旧图库仍可查看', async () => {
+        let fail = false
+        const form = formFixture()
+        form.images.unshift(imageFixture({ id: 'running', isPrimary: false, status: 'generating', imageUrl: null }))
+        const fetcher = vi
+            .fn<typeof fetch>()
+            .mockImplementation(async () =>
+                fail
+                    ? new Response('{"message":"数据库异常"}', { status: 500 })
+                    : new Response(JSON.stringify({ data: [form] }))
+            )
+        vi.stubGlobal('fetch', fetcher)
+        await mountSubjectImages()
+        expect(button('再生成一张').disabled).toBe(true)
+        expect(wrapper!.text()).toContain('数据库中仍有排队或生成中的图片')
+        fail = true
+        button('刷新图库').click()
+        await flushPromises()
+        expect(wrapper!.text()).toContain('数据库异常')
+        expect(wrapper!.find('.form-image-card img').exists()).toBe(true)
+        expect(button('补齐项目主参考图').disabled).toBe(true)
+    })
     it('参考图拦截危险地址，生成规格只通过 GET 读取正式形态和状态', async () => {
         const shot = designedShot()
         const fetcher = vi

src/lib/assets.ts

@@ -0,0 +1,16 @@
+/** 只允许 HTTP(S) 与后端 /storage 路径；生产环境绝对 API 地址使用其源站。 */
+export function referenceImageUrl(value: string): string | null {
+    const raw = value.trim()
+    try {
+        if (raw.startsWith('/storage/') && !raw.includes('\\')) {
+            const api = import.meta.env.VITE_API_BASE_URL || '/api'
+            const origin = /^https?:\/\//i.test(api) ? new URL(api).origin : window.location.origin
+            const url = new URL(raw, origin)
+            return url.pathname.startsWith('/storage/') ? url.href : null
+        }
+        const url = new URL(raw)
+        return ['http:', 'https:'].includes(url.protocol) ? url.href : null
+    } catch {
+        return null
+    }
+}

src/lib/http.test.ts

@@ -3,6 +3,7 @@ import { ApiError, optionalResource, request } from './http'
 import { projectsApi } from '../features/projects/api'
 import { breakdownApi } from '../features/breakdown/api'
 import { storyboardApi } from '../features/storyboard/api'
+import { subjectImagesApi } from '../features/subject-images/api'
 
 afterEach(() => {
     vi.unstubAllGlobals()
@@ -10,6 +11,41 @@ afterEach(() => {
 })
 
 describe('后端 API 契约', () => {
+    it('图库读取与单图、批量和切换主图使用各自接口，默认显式 Seedream', async () => {
+        const fetcher = vi.fn<typeof fetch>().mockImplementation(async () => new Response('{"data":[]}'))
+        vi.stubGlobal('fetch', fetcher)
+        await subjectImagesApi.listForms('a/b')
+        await subjectImagesApi.listImages('form/id')
+        await subjectImagesApi.generate('form/id', {
+            provider: 'seedream',
+            setPrimary: false,
+            width: 2048,
+            height: 2048,
+            prompt: '本次提示词'
+        })
+        await subjectImagesApi.generateProject('a/b', { provider: 'seedream', concurrency: 2, force: false })
+        await subjectImagesApi.setPrimary('form/id', 'image/id')
+        expect(fetcher.mock.calls.map(([url, init]) => [url, init?.method])).toEqual([
+            ['/api/projects/a%2Fb/subject-forms', 'GET'],
+            ['/api/subject-forms/form%2Fid/images', 'GET'],
+            ['/api/subject-forms/form%2Fid/images', 'POST'],
+            ['/api/projects/a%2Fb/subject-images/generate', 'POST'],
+            ['/api/subject-forms/form%2Fid/images/image%2Fid/primary', 'PUT']
+        ])
+        expect(JSON.parse(fetcher.mock.calls[2]![1]!.body as string)).toEqual({
+            provider: 'seedream',
+            setPrimary: false,
+            width: 2048,
+            height: 2048,
+            prompt: '本次提示词'
+        })
+        expect(JSON.parse(fetcher.mock.calls[3]![1]!.body as string)).toEqual({
+            provider: 'seedream',
+            concurrency: 2,
+            force: false
+        })
+        expect(fetcher.mock.calls[4]![1]!.body).toBeUndefined()
+    })
     it('分镜单集显式持久化，批量保持 force 和零次修复参数', async () => {
         const fetcher = vi.fn<typeof fetch>().mockImplementation(async () => new Response('{"data":{}}'))
         vi.stubGlobal('fetch', fetcher)

src/lib/http.ts

@@ -12,7 +12,7 @@ export class ApiError extends Error {
 
 /** 请求选项；长工作流显式关闭超时，不自动重试任何 POST。 */
 export interface RequestOptions {
-    method?: 'GET' | 'POST'
+    method?: 'GET' | 'POST' | 'PUT'
     body?: unknown
     signal?: AbortSignal
     timeoutMs?: number

src/router/index.ts

@@ -25,6 +25,11 @@ export const router = createRouter({
                     component: () => import('../features/breakdown/BreakdownPage.vue'),
                     meta: { title: '剧本拆解' }
                 },
+                {
+                    path: 'subject-images',
+                    component: () => import('../features/subject-images/SubjectImagesPage.vue'),
+                    meta: { title: '形态图片' }
+                },
                 {
                     path: 'storyboard',
                     component: () => import('../features/storyboard/StoryboardPage.vue'),

src/styles.css

@@ -614,6 +614,72 @@
         background: #8a9b6f;
         box-shadow: 0 0 0 3px #e9edde;
     }
+    .dialog-content.dialog-wide {
+        width: min(820px, calc(100vw - 36px));
+    }
+    .form-image-grid {
+        display: grid;
+        grid-template-columns: repeat(auto-fill, minmax(min(260px, 100%), 1fr));
+        gap: 20px;
+    }
+    .form-image-card {
+        overflow: hidden;
+    }
+    .asset-image {
+        display: grid;
+        place-items: center;
+        width: 100%;
+        aspect-ratio: 4 / 3;
+        overflow: hidden;
+        background: #eff0e9;
+    }
+    .asset-image img {
+        width: 100%;
+        height: 100%;
+        object-fit: contain;
+        min-height: 0;
+    }
+    .asset-image-empty {
+        display: flex;
+        align-items: center;
+        justify-content: center;
+        flex-direction: column;
+        gap: 10px;
+        padding: 20px;
+        color: var(--color-muted);
+        font-size: 11px;
+        text-align: center;
+    }
+    .asset-image-preview {
+        aspect-ratio: auto;
+        height: min(48vh, 480px);
+    }
+    .image-history {
+        display: flex;
+        gap: 12px;
+        overflow-x: auto;
+        padding: 4px;
+    }
+    .image-history-item {
+        width: 110px;
+        flex-shrink: 0;
+        padding: 4px;
+        border: 1px solid var(--color-line);
+        border-radius: 4px;
+    }
+    .image-history-item.selected {
+        border-color: var(--color-accent);
+        background: #faf2ed;
+    }
+    .image-history-item .asset-image-empty {
+        font-size: 9px;
+        padding: 6px;
+        gap: 4px;
+    }
+    .image-history-item .asset-image-empty svg {
+        width: 14px;
+        height: 14px;
+    }
     .storyboard-controls {
         display: grid;
         grid-template-columns: minmax(180px, 1fr) 100px 145px auto;