feat: AI炫图接入gpt-image-2.5云端生图(结构化提示词工程+照片参考合成)

CHANGELOG

@@ -1,5 +1,7 @@
 # CHANGELOG
 
+- [2026-09-25 18:30] FEAT: v2.28 AI 炫图接入 gpt-image-2.5 云端生图+提示词工程。①services/ai_prompt.py 结构化提示词构建器: 从 drawing_data 提取窗型特征(组合单元/宽高mm/格数/扇类型平开推拉内开内倒对开门/开向/纱扇/圆弧/格条)→专业英文提示词(主体先行/材质风格映射/摄影光照/负面约束"保持分格勿增删扇"/质量词)+按宽高比自动选尺寸(1024/1536x1024/1024x1536); ②api/ai_render.py 代理(Key 仅后端 .env HMC_AI_API_KEY, 已 gitignore): POST /ai-render/prompt 提示词预览 + /ai-render/generate(quick→generations, photo→edits 参考图合成, httpx 180s 超时, 错误分级); ③config .env 锚定项目根(修复 cwd 相对路径失效); ④前端 AIRenderView 双模式主按钮接云端(photo 模式参考图压缩≤1536px JPEG), 结果大图+提示词折叠面板+下载, 未配 Key 降级本地 4 方案; ⑤实测: quick 生图 28s 返回三格平开窗产品级渲染(把手/铰链/投影正确), photo 合成 55s 返回窗型融入照片; 端到端 puppeteer 验证通过(shots/verify-aicloud.png); vue-tsc 0 错+构建通过 (Files: pyproject.toml, src/huamenchuang/services/ai_prompt.py, src/huamenchuang/api/ai_render.py, src/huamenchuang/api/router.py, src/huamenchuang/core/config.py, frontend/src/views/AIRenderView.vue, frontend/src/api/index.ts, frontend/scripts/verify-aicloud.mjs, doc/API.md)
+
 - [2026-09-25 16:50] FEAT: v2.27 P0-8 照片式 AI 炫图(7 场景)。①photoCompose.ts 纯前端合成引擎: renderWindowOnly 透明背景纯窗型渲染(分割树重放/玻璃渐变/中梃投影/扇开向虚线三角/装饰格条, 复用 types/drawing 纯函数)+composePhoto 光照匹配(洞口周边亮度→窗户亮度 0.55~1.25)+7 场景后处理(渲图/安装合成投影/门窗擦除内容填充/室内暖色/别墅石基座/去围栏背景虚化/变毛坯去饱和); ②AIRenderView 升级双 Tab(照片合成/风格速览): 上传照片→拖框窗洞→选样板→生成 4 方案画廊+下载; 修复三处: variants 赋值后未等 Vue 挂载渲染空画布(nextFrames)、rect 舞台坐标与照片原始坐标双重缩放(统一原始坐标+CSS↔原始宽换算)、v-if canvas 未挂载 drawStage 静默失败致 photoScale 滞留 1(nextTick+无状态换算); ③端到端验证: 程序化测试照片→拖框→4 方案生成, 窗户准确贴入洞口+投影正确(shots/verify-photorender-*.png); vue-tsc 0 错+构建通过 (Files: frontend/src/components/photoCompose.ts, frontend/src/views/AIRenderView.vue, frontend/scripts/verify-photorender.mjs, doc/API.md)
 
 - [2026-09-25 15:40] FEAT: v2.26 P0-7 玻璃装饰格条设计器。①types/drawing.ts 加 DecBar(decs 挂 FrameShape, ri=重放区域索引)+DEC_PRESETS 8 风格(对齐源站 decorationBar: 现代轻奢/意式简约/奶油风/中式/欧式古典/法式/地中海/田园密格); ②WindoorCanvas: 选格后 HUD"格条"按钮+设计对话框(预设缩略图/行列/条宽/6 色)+2D 渲染(区域 clip 内横竖条+描边, 应用/清除入撤销栈); ③WindoorThreed: 静态格+扇面板玻璃前表面格条体(barX/barY 实心腔); ④后端 takeoff 计 dec_bar 算料(横=区域宽×rows/竖=区域高×cols, 默认 12 元/米); ⑤端到端验证: 无头浏览器订单详情→编辑→选格→欧式古典 3×3 应用→截图渲染正确(shots/verify-decbar-2d.png); 254 套全量校验 0 失败(decs 可选字段无回归); vue-tsc 0 错+构建通过 (Files: frontend/src/types/drawing.ts, frontend/src/components/WindoorCanvas.vue, frontend/src/components/WindoorThreed.vue, src/huamenchuang/services/quote_service.py, frontend/scripts/verify-decbar.mjs, doc/API.md)

doc/API.md

@@ -465,3 +465,17 @@ uv run uvicorn huamenchuang.main:app --reload   # 启动，http://localhost:8000
 - 流程：上传场景照片 → 拖拽框选窗洞（原始照片坐标）→ 选样板窗型 → `renderWindowOnly`（分割树重放渲染透明背景窗型：玻璃渐变/中梃/扇框开向/装饰格条）→ 光照匹配（采样洞口周边亮度调整窗户亮度）→ 边缘贴合 → 场景后处理（安装投影/擦除填充/暖色滤镜/石材基座/背景虚化/去饱和毛坯）
 - 一键生成 4 方案画廊 + PNG 下载；原"风格速览"模式保留为第二 Tab
 - 无新增后端接口（复用 `/api/windoor/templates` 与 `/api/rendering-info`）
+
+### 12.23 云端 AI 生图 `/api/ai-render`（v2.28 新增，gpt-image-2.5）
+
+接入 OpenAI Images 兼容中转站（`geminiapi.ai`，模型 `gpt-image-2.5-1k`）。**Key 仅存后端**（`.env` 的 `HMC_AI_API_KEY`，已 gitignore；Docker 部署用环境变量注入）。
+
+**提示词工程**（`services/ai_prompt.py`，核心资产）：从 `drawing_data` 结构化提取窗型特征（组合单元数/每单元宽高 mm/玻璃格数/扇类型（平开、推拉、内开内倒、对开门）/开向内外/纱扇/圆弧复古/装饰格条）→ 生成专业英文提示词：主体先行（窗型几何量化描述）→ 材质风格（4 风格映射铝型材 finish）→ 摄影光照（3 时间）→ 负面约束（"保持窗型分格不变，勿增删扇"）→ 质量词。生成尺寸按窗型宽高比自动选择（1024x1024 / 1536x1024 / 1024x1536）。
+
+| 接口名称 | 方式 | URL 路径 | 请求参数 | 说明 |
+|---|---|---|---|---|
+| 提示词预览 | POST | `/api/ai-render/prompt` | `{template_id\|drawing_data, style, time, scene}` | 返回 `{features(窗型特征), prompt(优化后完整提示词), size}` |
+| **云端生成** | POST | `/api/ai-render/generate` | `{template_id\|drawing_data, mode: quick\|photo, style, time, scene, photo_b64?}` | quick→`images/generations`；photo→`images/edits`（用户照片为参考图，AI 将窗型合成进照片窗洞）；返回 `{image_b64, prompt, size, features}`；耗时 30~90s（前端 timeout 200s） |
+
+- 未配置 Key 时接口返回 code=0，前端降级本地程序化渲染（"本地 4 方案"按钮）
+- 前端：AI 炫图页两个模式（照片合成/风格速览）主按钮均走云端生成，结果大图展示 + 提示词折叠面板可查看 + PNG 下载；照片模式参考图自动压缩 ≤1536px JPEG

frontend/scripts/verify-aicloud.mjs

@@ -0,0 +1,72 @@
+// 云端 AI 炫图(gpt-image)端到端验证: 注入测试照片 → 拖框 → AI 照片合成 → 截图
+// 用法: node verify-aicloud.mjs <token>
+import puppeteer from 'puppeteer-core'
+const browser = await puppeteer.launch({
+  executablePath: 'C:/Program Files (x86)/Microsoft/Edge/Application/msedge.exe',
+  headless: 'new',
+  args: ['--window-size=1440,900', '--enable-unsafe-swiftshader'],
+  defaultViewport: { width: 1440, height: 900 },
+})
+const page = await browser.newPage()
+page.on('pageerror', e => console.log(`[JS异常] ${String(e).slice(0, 200)}`))
+const token = process.argv[2]
+await page.goto(`http://localhost:8000/ai_render?token=${encodeURIComponent(token)}`, { waitUntil: 'networkidle2' })
+await new Promise(r => setTimeout(r, 2000))
+
+// 1) 选第一个样板
+await page.click('.controls .el-select')
+await new Promise(r => setTimeout(r, 800))
+await (await page.$('.el-select-dropdown__item')).click()
+await new Promise(r => setTimeout(r, 1000))
+
+// 2) 注入程序化"旧客厅"照片(带深色旧窗洞)
+await page.evaluate(() => {
+  const cv = document.createElement('canvas')
+  cv.width = 1200; cv.height = 800
+  const ctx = cv.getContext('2d')
+  const g = ctx.createLinearGradient(0, 0, 0, 800)
+  g.addColorStop(0, '#e8e0d2'); g.addColorStop(0.7, '#d8cdbb'); g.addColorStop(1, '#9c8b74')
+  ctx.fillStyle = g; ctx.fillRect(0, 0, 1200, 800)
+  ctx.fillStyle = 'rgba(120,100,80,0.25)'
+  ctx.fillRect(380, 260, 440, 340)
+  ctx.strokeStyle = 'rgba(90,70,50,0.5)'; ctx.lineWidth = 8
+  ctx.strokeRect(380, 260, 440, 340)
+  return new Promise(resolve => {
+    cv.toBlob(blob => {
+      const dt = new DataTransfer()
+      dt.items.add(new File([blob], 'room.png', { type: 'image/png' }))
+      const input = document.querySelector('.el-upload input[type=file]')
+      input.files = dt.files
+      input.dispatchEvent(new Event('change'))
+      resolve()
+    }, 'image/png')
+  })
+})
+await new Promise(r => setTimeout(r, 1200))
+
+// 3) 框选窗洞
+const stage = await page.$('.photo-cv')
+const b = await stage.boundingBox()
+await page.mouse.move(b.x + b.width * 0.3, b.y + b.height * 0.3)
+await page.mouse.down()
+await page.mouse.move(b.x + b.width * 0.7, b.y + b.height * 0.75, { steps: 8 })
+await page.mouse.up()
+await new Promise(r => setTimeout(r, 300))
+
+// 4) 切到"室外渲染"场景更像安装效果: 保持默认渲图, 点击 AI 照片合成
+const genBtn = await page.evaluateHandle(() => [...document.querySelectorAll('button')].find(x => x.textContent.includes('AI 照片合成')))
+await genBtn.asElement().click()
+console.log('已提交云端生成, 等待 AI 绘制(最长 150s)...')
+// 轮询等 ai-img 出现
+let ok = false
+for (let i = 0; i < 75; i++) {
+  await new Promise(r => setTimeout(r, 2000))
+  ok = await page.evaluate(() => !!document.querySelector('.ai-img'))
+  if (ok) break
+  if (i % 5 === 4) console.log(`  等待中 ${(i + 1) * 2}s`)
+}
+if (!ok) { console.log('✗ 超时未返回'); await browser.close(); process.exit(1) }
+const img = await page.$('.ai-result')
+await img.screenshot({ path: 'D:/Code/huamenchuang/shots/verify-aicloud.png' })
+console.log('✓ 云端 AI 照片合成验证完成')
+await browser.close()

frontend/src/api/index.ts

@@ -36,9 +36,9 @@ http.interceptors.response.use(
   },
 )
 
-/** 提取 envelope 的 data */
+/** 提取 envelope 的 data（config 可传 timeout 等覆盖） */
-export async function api<T = any>(method: string, url: string, body?: object): Promise<T> {
+export async function api<T = any>(method: string, url: string, body?: object, config?: object): Promise<T> {
-  const res = await http.request({ method, url, data: body })
+  const res = await http.request({ method, url, data: body, ...config })
   return res.data.data as T
 }
 

frontend/src/views/AIRenderView.vue

@@ -24,8 +24,9 @@
         <el-upload :show-file-list="false" accept="image/*" :http-request="onUpload">
           <el-button type="primary" plain>{{ photo ? '更换照片' : '上传场景照片' }}</el-button>
         </el-upload>
-        <el-button type="primary" :disabled="!photo || !drawing || !rect" :loading="generating" @click="generate">
+        <el-button type="primary" :disabled="!photo || !drawing || !rect" :loading="generating"
-          {{ generating ? '合成中…' : '✦ 生成炫图' }}</el-button>
+                   @click="generateCloud('photo')">
+          {{ generating ? 'AI 合成中…' : '✦ AI 照片合成' }}</el-button>
       </div>
       <div v-if="!photo" class="drop-hint" @click="triggerUpload">
         <el-icon size="40"><UploadFilled /></el-icon>
@@ -67,8 +68,9 @@
         <el-select v-model="form.scene" style="width:110px">
           <el-option value="indoor" label="室内" /><el-option value="outdoor" label="室外" /><el-option value="studio" label="影棚" />
         </el-select>
-        <el-button type="primary" :disabled="!drawing" :loading="generating" @click="generateQuick">
+        <el-button type="primary" :disabled="!drawing" :loading="generating" @click="generateCloud('quick')">
-          {{ generating ? '生成中…' : '✦ 生成 AI 炫图' }}</el-button>
+          {{ generating ? 'AI 绘制中…' : '✦ AI 云端生成' }}</el-button>
+        <el-button :disabled="!drawing" :loading="generating" @click="generateQuick">本地 4 方案</el-button>
       </div>
 
       <el-progress v-if="generating" :percentage="progress" :stroke-width="10" style="margin: 10px 0"
@@ -82,7 +84,21 @@
     </template>
 
     <div v-if="variants.length" class="gallery-foot">
-      <span class="tip">点击方案切换大图预览，图标下载 PNG</span>
+      <span class="tip">本地 4 方案（程序化渲染降级模式）· 点击图标下载</span>
+    </div>
+
+    <!-- 云端 AI 结果（两个模式共用） -->
+    <div v-if="aiImage" class="ai-result">
+      <img :src="aiImage" class="ai-img" />
+      <div class="ai-ops">
+        <el-button type="primary" size="small" plain @click="downloadAi">下载 PNG</el-button>
+      </div>
+      <el-collapse class="prompt-box">
+        <el-collapse-item title="查看优化后的提示词（由窗型数据结构化生成）">
+          <div class="prompt-text">{{ aiPrompt }}</div>
+          <div class="prompt-feat" v-if="aiFeatures">窗型特征：{{ aiFeatures }}</div>
+        </el-collapse-item>
+      </el-collapse>
     </div>
   </el-card>
 </template>
@@ -92,7 +108,7 @@ import { computed, nextTick, onMounted, reactive, ref } from 'vue'
 import { ElMessage } from 'element-plus'
 import { api } from '../api'
 import AIRenderCanvas from '../components/AIRenderCanvas.vue'
-import { PHOTO_SCENES, composePhoto, type PhotoScene } from '../components/photoCompose'
+import { PHOTO_SCENES, type PhotoScene } from '../components/photoCompose'
 
 const tab = ref<'photo' | 'quick'>('photo')
 const templates = ref<any[]>([])
@@ -122,8 +138,6 @@ const stageRef = ref<HTMLCanvasElement>()
 const rect = ref<{ x: number; y: number; w: number; h: number } | null>(null)
 let dragStart: { x: number; y: number } | null = null
 
-const sceneLabel = (s: string) => PHOTO_SCENES.find(x => x.key === s)?.label ?? s
-
 async function loadTemplate(tid: number | undefined) {
   drawing.value = ''
   variants.value = []
@@ -209,52 +223,65 @@ function drawStage() {
   }
 }
 
-// ---- 照片合成生成(4 方案: 当前场景 + 3 变化) ----
+// ---- 云端 AI 生成（gpt-image-2.5, 提示词由后端从窗型数据结构化构建） ----
-async function generate() {
+const aiImage = ref('')
-  if (!photo.value || !drawing.value || !rect.value) return
+const aiPrompt = ref('')
+const aiFeatures = ref('')
+
+async function generateCloud(mode: 'quick' | 'photo') {
+  if (!drawing.value) return
+  if (mode === 'photo' && (!photo.value || !rect.value)) return
   generating.value = true
+  aiImage.value = ''
   variants.value = []
-  const plan: { key: PhotoScene; label: string }[] = [
+  progress.value = 8
-    { key: scene.value, label: `${sceneLabel(scene.value)}` },
+  stageText.value = mode === 'photo' ? '上传照片与窗型特征' : '解析窗型结构'
-    { key: 'install', label: '安装合成' },
+  const timer = window.setInterval(() => {
-    { key: 'erase', label: '门窗擦除' },
+    progress.value = Math.min(92, progress.value + 2)
-    { key: scene.value === 'render' ? 'indoor' : 'render', label: scene.value === 'render' ? '室内装修' : '渲图' },
+    if (progress.value > 40) stageText.value = 'AI 绘制中(约 30~90 秒, 请勿关闭页面)'
-  ]
+  }, 1500)
   try {
-    for (let s = 0; s < STAGES.length; s++) {
+    let photo_b64: string | undefined
-      stageText.value = STAGES[s]
+    if (mode === 'photo') {
-      for (let p = 0; p <= 25; p += 5) {
+      photo_b64 = photoB64()
-        progress.value = Math.min(100, s * 25 + p)
-        await new Promise(r => setTimeout(r, 18))
-      }
-      await new Promise(r => setTimeout(r, 10))
     }
+    const d = await api<any>('post', '/api/ai-render/generate', {
+      template_id: templateId.value, mode,
+      style: form.ai_style, time: form.ai_time,
+      scene: mode === 'photo' ? scene.value : form.scene, photo_b64,
+    }, { timeout: 200000 })
+    aiImage.value = `data:image/png;base64,${d.image_b64}`
+    aiPrompt.value = d.prompt
+    aiFeatures.value = d.features?.summary ?? ''
     progress.value = 100
-    variants.value = plan.map((p, i) => ({
+    ElMessage.success('AI 炫图已生成')
-      label: p.label, style: String(p.key), time: String(i), seed: 7 + i * 13,
+  } catch { /* 拦截器已提示 */ } finally {
-    }))
+    window.clearInterval(timer)
-    pickIdx.value = 0
-    // 等 Vue 挂载画廊画布后再合成输出
-    await nextFrames()
-    // 渲染画廊: composePhoto 输出到各画布
-    for (let i = 0; i < plan.length; i++) {
-      const cv = variantCvs[i]
-      if (!cv) continue
-      const out = composePhoto(photo.value, rect.value, drawing.value, plan[i].key, 1280, 7 + i * 13)
-      cv.width = out.width
-      cv.height = out.height
-      cv.getContext('2d')!.drawImage(out, 0, 0)
-      stageText.value = `合成方案 ${i + 1}/4`
-      await new Promise(r => setTimeout(r, 30))
-    }
-    ElMessage.success('已生成 4 套炫图方案')
-  } finally {
     generating.value = false
     progress.value = 0
   }
 }
 
-// ---- 风格速览(原有程序化模式) ----
+/** 照片模式参考图: 全图压至 ≤1536px JPEG base64 */
+function photoB64(): string {
+  const img = photo.value!
+  const scale = Math.min(1, 1536 / img.naturalWidth)
+  const cv = document.createElement('canvas')
+  cv.width = Math.round(img.naturalWidth * scale)
+  cv.height = Math.round(img.naturalHeight * scale)
+  cv.getContext('2d')!.drawImage(img, 0, 0, cv.width, cv.height)
+  return cv.toDataURL('image/jpeg', 0.9).split(',')[1]
+}
+
+function downloadAi() {
+  if (!aiImage.value) return
+  const a = document.createElement('a')
+  a.href = aiImage.value
+  a.download = 'AI炫图.png'
+  a.click()
+}
+
+// ---- 风格速览(本地程序化 4 方案, 未配置云端 Key 时的降级模式) ----
 async function generateQuick() {
   generating.value = true
   variants.value = []
@@ -321,6 +348,13 @@ onMounted(async () => {
 .preview { height: 58vh; }
 .gallery { display: grid; grid-template-columns: repeat(auto-fill, minmax(240px, 1fr)); gap: 12px; margin-top: 14px; }
 .gallery-foot { margin-top: 8px; }
+.ai-result { margin-top: 14px; text-align: center; }
+.ai-img { max-width: 100%; max-height: 72vh; border-radius: 10px; box-shadow: 0 2px 14px rgba(0,0,0,.14); }
+.ai-ops { margin-top: 8px; }
+.prompt-box { margin-top: 10px; text-align: left; }
+.prompt-text { font-size: 12px; color: #606266; line-height: 1.7; font-family: Consolas, monospace;
+  background: #f7f8fa; padding: 10px 12px; border-radius: 6px; word-break: break-word; }
+.prompt-feat { margin-top: 8px; font-size: 12px; color: #909399; }
 .shot { position: relative; border: 2px solid transparent; border-radius: 10px; overflow: hidden; cursor: pointer;
   height: 200px; transition: border-color .2s, transform .2s; }
 .shot:hover { transform: translateY(-2px); }

pyproject.toml

@@ -14,6 +14,7 @@ dependencies = [
     "pyjwt>=2.8",
     "openpyxl>=3.1",
     "python-multipart>=0.0.32",
+    "httpx>=0.27",
 ]
 
 [project.scripts]

src/huamenchuang/api/ai_render.py

@@ -0,0 +1,107 @@
+"""AI 炫图路由：gpt-image 生图代理（Key 仅存后端），提示词由 drawing_data 结构化构建。
+
+- quick 模式: /v1/images/generations 纯提示词生成
+- photo 模式: /v1/images/edits 以用户照片为参考图，把窗型合成进照片
+"""
+import base64
+import json
+
+import httpx
+from fastapi import APIRouter
+from pydantic import BaseModel, Field
+
+from huamenchuang.core.config import settings
+from huamenchuang.core.deps import CurrentUser, DbSession
+from huamenchuang.core.response import fail, ok
+from huamenchuang.services.ai_prompt import build_prompt, describe_window, pick_size
+
+router = APIRouter(prefix="/ai-render", tags=["AI 炫图"])
+
+
+class AiRenderBody(BaseModel):
+    template_id: int | None = None
+    drawing_data: str | dict | None = None
+    mode: str = Field(default="quick", pattern="^(quick|photo)$")
+    style: str = Field(default="modern", pattern="^(modern|chinese|european|nordic)$")
+    time: str = Field(default="day", pattern="^(day|dusk|night)$")
+    # quick: indoor/outdoor/studio; photo: render/install/erase/indoor/villa/fence/blank
+    scene: str = "indoor"
+    photo_b64: str | None = Field(default=None, description="照片模式参考图(base64, 不带 dataURL 前缀)")
+
+
+async def _resolve_drawing(db, body: AiRenderBody) -> str | dict | None:
+    if body.drawing_data:
+        return body.drawing_data
+    if body.template_id:
+        from sqlalchemy import select
+        from huamenchuang.models.windoor_template import WindoorTemplate
+        t = (await db.execute(select(WindoorTemplate).where(
+            WindoorTemplate.id == body.template_id))).scalar_one_or_none()
+        return t.drawing_data if t else None
+    return None
+
+
+@router.post("/prompt")
+async def preview_prompt(body: AiRenderBody, user: CurrentUser, db: DbSession):
+    """提示词预览（前端可展开查看优化后的完整提示词）。"""
+    drawing = await _resolve_drawing(db, body)
+    if drawing is None:
+        return fail("未找到图纸数据")
+    try:
+        info = describe_window(drawing)
+        prompt = build_prompt(drawing, body.style, body.time, body.scene)
+    except Exception:
+        return fail("图纸数据解析失败")
+    return ok(data={"features": info, "prompt": prompt,
+                    "size": pick_size(info["width_mm"], info["height_mm"])})
+
+
+@router.post("/generate")
+async def generate(body: AiRenderBody, user: CurrentUser, db: DbSession):
+    """云端 AI 生图（gpt-image）：quick=生成 / photo=参考图编辑。"""
+    if not settings.ai_api_key:
+        return fail("AI 生图服务未配置(HMC_AI_API_KEY)")
+    drawing = await _resolve_drawing(db, body)
+    if drawing is None:
+        return fail("未找到图纸数据")
+    if body.mode == "photo" and not body.photo_b64:
+        return fail("照片模式需要上传参考照片")
+    try:
+        info = describe_window(drawing)
+        prompt = build_prompt(drawing, body.style, body.time, body.scene)
+        size = pick_size(info["width_mm"], info["height_mm"])
+    except Exception:
+        return fail("图纸数据解析失败")
+
+    base = settings.ai_base_url.rstrip("/")
+    headers = {"Authorization": f"Bearer {settings.ai_api_key}"}
+    timeout = httpx.Timeout(settings.ai_timeout, connect=15)
+    try:
+        async with httpx.AsyncClient(timeout=timeout) as client:
+            if body.mode == "photo":
+                img_bytes = base64.b64decode(body.photo_b64.split(",")[-1])
+                resp = await client.post(
+                    f"{base}/v1/images/edits", headers=headers,
+                    data={"model": settings.ai_model, "prompt": prompt, "size": size, "n": "1"},
+                    files={"image[]": ("photo.png", img_bytes, "image/png")})
+            else:
+                resp = await client.post(
+                    f"{base}/v1/images/generations", headers=headers,
+                    json={"model": settings.ai_model, "prompt": prompt, "n": 1, "size": size})
+        resp.raise_for_status()
+        data = resp.json()
+        item = (data.get("data") or [{}])[0]
+        b64 = item.get("b64_json")
+        if not b64:
+            return fail(f"生图服务返回异常: {str(data)[:200]}")
+    except httpx.TimeoutException:
+        return fail("AI 生图超时，请稍后重试")
+    except httpx.HTTPStatusError as e:
+        detail = e.response.text[:200] if e.response is not None else ""
+        return fail(f"生图服务错误({e.response.status_code if e.response else '?'}): {detail}")
+    except Exception as e:
+        return fail(f"生图失败: {str(e)[:150]}")
+    return ok(data={"image_b64": b64, "prompt": prompt, "size": size,
+                    "features": {"width_mm": info["width_mm"], "height_mm": info["height_mm"],
+                                 "summary": info["summary"]}},
+              message="生成成功")

src/huamenchuang/api/router.py

@@ -2,6 +2,7 @@
 from fastapi import APIRouter
 
 from huamenchuang.api import (
+    ai_render,
     auth,
     business,
     contracts,
@@ -29,6 +30,7 @@ api_router.include_router(quote.router)
 api_router.include_router(script.router)
 api_router.include_router(order_format.router)
 api_router.include_router(finance.router)
+api_router.include_router(ai_render.router)
 api_router.include_router(warehouse.router)
 api_router.include_router(business.router)
 api_router.include_router(platform.router)

src/huamenchuang/core/config.py

@@ -5,6 +5,7 @@ from pydantic_settings import BaseSettings
 
 _PROJECT_ROOT = Path(__file__).resolve().parents[3]
 _DEFAULT_DB = f"sqlite+aiosqlite:///{(_PROJECT_ROOT / 'huamenchuang.db').as_posix()}"
+_ENV_FILE = _PROJECT_ROOT / ".env"
 
 
 class Settings(BaseSettings):
@@ -19,8 +20,15 @@ class Settings(BaseSettings):
         "template", "customer", "contract", "warehouse", "order",
         "quote", "optimize", "render",
     ]
+    # AI 炫图（OpenAI Images 兼容中转站；未配置时前端降级本地程序化渲染）
+    ai_base_url: str = "https://geminiapi.ai"
+    ai_api_key: str = ""
+    ai_model: str = "gpt-image-2.5-1k"
+    ai_timeout: int = 180
 
-    model_config = {"env_prefix": "HMC_", "env_file": ".env"}
+    # .env 锚定项目根(与进程 cwd 无关); 不存在时不加载
+    model_config = {"env_prefix": "HMC_",
+                    "env_file": str(_ENV_FILE) if _ENV_FILE.exists() else None}
 
 
 settings = Settings()

src/huamenchuang/services/ai_prompt.py

@@ -0,0 +1,169 @@
+"""AI 炫图提示词构建器：从 drawing_data 提取窗型结构特征，生成专业英文生图提示词。
+
+设计原则（生图模型工程实践）:
+- 主体先行: 窗户的几何/材质/分格描述置前，场景氛围居后
+- 结构量化: 宽高 mm、分格数、扇数、开向全部显式给出，抑制模型自由发挥
+- 负面约束: 明确"保持窗型几何不变"（edits 模式关键）
+- 光影专项: 时间(日/黄昏/夜)与室内外视角分别给出摄影级光照词
+"""
+from huamenchuang.core.drawing import parse_drawing, replay_shape
+
+# 风格 → 型材材质/颜色描述
+STYLE_FINISH = {
+    "modern": ("matte anthracite grey aluminum frames, minimalist slim profiles", "modern minimalist"),
+    "chinese": ("dark bronze aluminum frames with wood-grain inner finish and subtle Chinese lattice grille", "new-Chinese style"),
+    "european": ("cream white aluminum frames with classic decorative architrave", "classic European"),
+    "nordic": ("white aluminum frames with light oak interior foil", "Scandinavian Nordic"),
+}
+
+# 时间 → 摄影光照
+TIME_LIGHT = {
+    "day": "bright midday daylight, soft natural shadows, clear sky",
+    "dusk": "warm golden-hour sunset light, long soft shadows, amber sky",
+    "night": "night scene, warm interior glow shining through the glass, cool blue ambience",
+}
+
+# 场景 → 场景描述（照片式 7 场景对齐源站 renderV2）
+SCENE_DESC = {
+    "render": "the window installed on a clean rendered exterior wall, straight-on elevation view",
+    "install": "the brand-new window just mounted into the wall opening, installation foam and fixing screws still visible at the edges, home renovation context",
+    "erase": "renovation scene: the old worn window has been removed from the masonry opening and this new window is being fitted in its place, dusty sill",
+    "indoor": "interior view, the window as the focal point of a tastefully furnished {style_word} living room with sofa, wooden floor, curtains and green plants",
+    "villa": "exterior view of a {style_word} villa facade featuring this large window, landscaped garden with lawn below, blue sky",
+    "fence": "view from inside looking out through the window, clear unobstructed garden and distant hills beyond the glass, no railings or bars outside",
+    "blank": "unfurnished rough-condition room with bare grey cement walls, the window as the only fitted element, pre-renovation state",
+}
+
+# 快速模式场景（无照片）
+QUICK_SCENE = {
+    "indoor": "interior view in a tastefully furnished {style_word} room, the window as the light source",
+    "outdoor": "exterior facade view on a residential building, mild urban greenery around",
+    "studio": "studio product shot on a seamless neutral background, softbox lighting, slight floor reflection",
+}
+
+
+def _sash_kind(sub: dict) -> str | None:
+    t = str(sub.get("type", ""))
+    is_door = bool(sub.get("isDoor"))
+    if "Screen" in t:
+        return None  # 纱扇不参与文字描述
+    if "Kfc" in t:
+        return "tilt-and-turn " + ("door leaf" if is_door else "sash")
+    if "DoubleSash" in t:
+        return "pair of French " + ("door leaves" if is_door else "casement sashes opening from the center")
+    if "Slide" in t:
+        return "sliding " + ("door panel" if is_door else "sash")
+    if t in ("Sash",):
+        return "side-hung casement " + ("door leaf" if is_door else "sash")
+    return None
+
+
+def _opening_words(sub: dict) -> str:
+    toward = str(sub.get("openToward") or "")
+    od = sub.get("od")
+    if "up" in str(sub.get("openDirection") or "").lower() or (isinstance(od, (int, float)) and 60 <= od <= 180 and "up" in str(sub.get("openDirection", "")).lower()):
+        return "top-hung"
+    return "inward-opening" if toward == "inward" else "outward-opening"
+
+
+def describe_window(drawing: str | dict) -> dict:
+    """从 drawing_data 提取窗型结构特征（供提示词与前端展示）。"""
+    dd = parse_drawing(drawing)
+    shapes = [s for s in dd.get("sm", []) if isinstance(s, dict)]
+    if not shapes:
+        return {"count": 0, "summary": "", "width_mm": 0, "height_mm": 0}
+
+    xs, ys = [], []
+    parts: list[str] = []
+    sash_total = 0
+    has_arc = has_screen = has_deco = False
+    for shape in shapes:
+        regions = replay_shape(shape)
+        pts = [[(seg["ps"]["x"], seg["ps"]["y"]) for seg in r] for r in regions]
+        if not pts:
+            continue
+        flat = [p for r in pts for p in r]
+        x0, y0 = min(p[0] for p in flat), min(p[1] for p in flat)
+        x1, y1 = max(p[0] for p in flat), max(p[1] for p in flat)
+        xs += [x0, x1]; ys += [y0, y1]
+        w, h = round(x1 - x0), round(y1 - y0)
+        n_region = len(regions)
+        # 扇描述
+        sashes = [sub for sub in (shape.get("sm") or []) if isinstance(sub, dict)]
+        kind_words: list[str] = []
+        for sub in sashes:
+            k = _sash_kind(sub)
+            if k is None:
+                if "Screen" in str(sub.get("type", "")):
+                    has_screen = True
+                continue
+            sash_total += 1
+            kind_words.append(f"{_opening_words(sub)} {k}")
+        if shape.get("rt"):
+            has_arc = True
+        if shape.get("decs"):
+            has_deco = True
+        part = f"{w}×{h}mm unit with {n_region} glass panes"
+        if kind_words:
+            part += ", " + " and ".join(dict.fromkeys(kind_words))
+        parts.append(part)
+
+    if not xs:
+        return {"count": 0, "summary": "", "width_mm": 0, "height_mm": 0}
+    width_mm, height_mm = round(max(xs) - min(xs)), round(max(ys) - min(ys))
+    multi = len(parts) > 1
+    summary = "; ".join(parts)
+    if multi:
+        summary = f"a combination of {len(parts)} window units joined together: " + summary
+    else:
+        summary = parts[0]
+    if has_arc:
+        summary += ", arched/curved top (heritage style)"
+    if has_screen:
+        summary += ", with retractable fly screen"
+    if has_deco:
+        summary += ", with decorative gliding bars (grilles) between the glass"
+    return {"count": len(parts), "summary": summary, "width_mm": width_mm,
+            "height_mm": height_mm, "sashes": sash_total}
+
+
+def build_prompt(drawing: str | dict, style: str = "modern", time: str = "day",
+                 scene: str = "render", ratio: str = "landscape") -> str:
+    """构建优化后的完整英文提示词。
+
+    ratio: portrait/landscape/square → 依据窗型宽高比的构图引导。
+    """
+    info = describe_window(drawing)
+    finish, style_word = STYLE_FINISH.get(style, STYLE_FINISH["modern"])
+    light = TIME_LIGHT.get(time, TIME_LIGHT["day"])
+    scene_tpl = SCENE_DESC.get(scene) or QUICK_SCENE.get(scene) or SCENE_DESC["render"]
+    scene_text = scene_tpl.replace("{style_word}", style_word)
+
+    window_desc = info["summary"] or "a modern aluminum window"
+    composition = {
+        "portrait": "vertical composition, window fills most of the frame height",
+        "landscape": "horizontal composition, generous wall context around the window",
+        "square": "balanced square composition, window centered",
+    }.get(ratio, "")
+
+    prompt = (
+        f"Photorealistic architectural photograph of {scene_text}. "
+        f"The window: {window_desc}; {finish}. "
+        f"Lighting: {light}. {composition} "
+        "Keep the window geometry, pane division and frame profiles exactly as described — do not add or remove sashes. "
+        "Realistic glass with soft sky reflection, physically correct shadows and contact shadows, "
+        "true-to-material aluminum texture, high dynamic range, ultra-detailed, 8k, professional architectural photography, no people, no text, no watermark."
+    )
+    return prompt
+
+
+def pick_size(width_mm: int, height_mm: int) -> str:
+    """按窗型宽高比选择生成尺寸（gpt-image 支持 1024x1024 / 1536x1024 / 1024x1536）。"""
+    if not width_mm or not height_mm:
+        return "1024x1024"
+    r = width_mm / height_mm
+    if r > 1.2:
+        return "1536x1024"
+    if r < 0.83:
+        return "1024x1536"
+    return "1024x1024"

uv.lock

@@ -42,6 +42,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/12/b8/4bd346e22b28902df4d651910f5242c28d84e4a5c2435ca5c3f797ed7e2e/anyio-4.15.1-py3-none-any.whl", hash = "sha256:6152fdbbf9a77fdec97731721bebf7c4c44f7c29b424b0065826173efc7ed101", size = 132079, upload-time = "2026-09-05T10:42:37.923Z" },
 ]
 
+[[package]]
+name = "certifi"
+version = "2026.7.22"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/a3/c2/24167ea9858356b47a87a50d39908bfdb72ceeefe0041586e704e5376b3a/certifi-2026.7.22.tar.gz", hash = "sha256:741e2c3b351ddf169a738da9f2c048608ff7f2c5cc02f1ebc6b118bb090d5d55", size = 138112, upload-time = "2026-07-22T03:35:12.644Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/0b/a7/71ac2cff56fec219ed242bb11b8efb69fcc4bec75db06fb7bfe35de520e6/certifi-2026.7.22-py3-none-any.whl", hash = "sha256:62f22742b58a1a33014a2b6b706588a8d7e2a88ae7bd1a6ebe8c992928483775", size = 136983, upload-time = "2026-07-22T03:35:11.276Z" },
+]
+
 [[package]]
 name = "click"
 version = "8.5.0"
@@ -124,6 +133,19 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/04/4b/29cac41a4d98d144bf5f6d33995617b185d14b22401f75ca86f384e87ff1/h11-0.16.0-py3-none-any.whl", hash = "sha256:63cf8bbe7522de3bf65932fda1d9c2772064ffb3dae62d55932da54b31cb6c86", size = 37515, upload-time = "2025-04-24T03:35:24.344Z" },
 ]
 
+[[package]]
+name = "httpcore"
+version = "1.0.9"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "certifi" },
+    { name = "h11" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/06/94/82699a10bca87a5556c9c59b5963f2d039dbd239f25bc2a63907a05a14cb/httpcore-1.0.9.tar.gz", hash = "sha256:6e34463af53fd2ab5d807f399a9b45ea31c3dfa2276f15a2c3f00afff6e176e8", size = 85484, upload-time = "2025-04-24T22:06:22.219Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/7e/f5/f66802a942d491edb555dd61e3a9961140fd64c90bce1eafd741609d334d/httpcore-1.0.9-py3-none-any.whl", hash = "sha256:2d400746a40668fc9dec9810239072b40b4484b640a8c38fd654a024c7a1bf55", size = 78784, upload-time = "2025-04-24T22:06:20.566Z" },
+]
+
 [[package]]
 name = "httptools"
 version = "0.8.0"
@@ -146,6 +168,21 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/48/63/b906c01e53f50d432c0defe43ce52764a111dc1bdd028bafbeb54dcfd008/httptools-0.8.0-cp314-cp314t-win_amd64.whl", hash = "sha256:384c17174464c8e873398b7af24f0b1f44d992c820328413951a625323155d77", size = 108209, upload-time = "2026-05-25T22:17:39.473Z" },
 ]
 
+[[package]]
+name = "httpx"
+version = "0.28.1"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "anyio" },
+    { name = "certifi" },
+    { name = "httpcore" },
+    { name = "idna" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/b1/df/48c586a5fe32a0f01324ee087459e112ebb7224f646c0b5023f5e79e9956/httpx-0.28.1.tar.gz", hash = "sha256:75e98c5f16b0f35b567856f597f06ff2270a374470a5c2392242528e3e3e42fc", size = 141406, upload-time = "2024-12-06T15:37:23.222Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/2a/39/e50c7c3a983047577ee07d2a9e53faf5a69493943ec3f6a384bdc792deb2/httpx-0.28.1-py3-none-any.whl", hash = "sha256:d909fcccc110f8c7faf814ca82a9a4d816bc5a6dbfea25d6591d6985b8ba59ad", size = 73517, upload-time = "2024-12-06T15:37:21.509Z" },
+]
+
 [[package]]
 name = "huamenchuang"
 version = "0.1.0"
@@ -153,6 +190,7 @@ source = { editable = "." }
 dependencies = [
     { name = "aiosqlite" },
     { name = "fastapi" },
+    { name = "httpx" },
     { name = "openpyxl" },
     { name = "pydantic" },
     { name = "pydantic-settings" },
@@ -166,6 +204,7 @@ dependencies = [
 requires-dist = [
     { name = "aiosqlite", specifier = ">=0.20" },
     { name = "fastapi", specifier = ">=0.115" },
+    { name = "httpx", specifier = ">=0.27" },
     { name = "openpyxl", specifier = ">=3.1" },
     { name = "pydantic", specifier = ">=2.7" },
     { name = "pydantic-settings", specifier = ">=2.3" },