diff --git a/docs/DEPLOYMENT_RUNBOOK_567001.md b/docs/DEPLOYMENT_RUNBOOK_567001.md
new file mode 100644
index 00000000..f11276e2
--- /dev/null
+++ b/docs/DEPLOYMENT_RUNBOOK_567001.md
@@ -0,0 +1,653 @@
+# GPT Image Playground 567001 部署与更新手册
+
+本文档整理当前这套服务在新环境上的可重复部署方式，以及后续更新流程。
+
+目标：
+- 部署 `gpt-image-playground`
+- 部署 `image-sse-proxy`
+- 配置 `systemd service`
+- 配置 `api.567001.xyz` 的 OpenResty/1Panel 代理规则
+- 提供后续更新步骤
+- 明确避免把任何 token 提交到 GitHub
+
+本文档不包含任何真实 token、PAT、密码、`gh hosts.yml` 内容。
+
+## 1. 当前架构
+
+当前线上链路分两段：
+
+1. `image.567001.xyz`
+   - 提供前端页面
+   - 反代到本机 `gpt-image-playground` Docker 容器
+   - 容器对外映射 `8081 -> 80`
+
+2. `api.567001.xyz`
+   - 提供 OpenAI 兼容接口
+   - 普通请求反代到本机 `newapi`，端口 `3000`
+   - `POST /v1/images/generations` 单独转到本机 `image-sse-proxy`，端口 `3101`
+
+这样做的原因：
+- `gpt-image-2` 的长图生成请求在 Cloudflare 链路上容易出现“上游最终成功，但浏览器提前断开”
+- `image-sse-proxy` 会立刻返回 `text/event-stream`，先发首包，再周期性发心跳，等上游最终 JSON 返回后再转成流式完成事件
+
+## 2. 目录与端口约定
+
+当前使用如下路径和端口：
+
+- 应用代码目录：`/opt/gpt_image_playground_update`
+- `image-sse-proxy` 目录：`/opt/image-sse-proxy`
+- systemd 文件：`/etc/systemd/system/image-sse-proxy.service`
+- 1Panel 站点配置目录：`/opt/1panel/www/sites/api.567001.xyz/proxy`
+- `gpt-image-playground` 容器映射端口：`8081`
+- `newapi` 端口：`3000`
+- `image-sse-proxy` 端口：`3101`
+
+如果新环境和这里不同，把下面命令里的路径和端口替换掉即可。
+
+## 3. 前置条件
+
+需要具备：
+
+- Docker
+- OpenResty 或 Nginx
+- systemd
+- Node.js
+- Git
+- GitHub CLI `gh`
+
+推荐 Node 版本：
+- `v22.x`
+
+当前机器使用：
+- `/root/.nvm/versions/node/v22.22.1/bin/node`
+
+如果新环境 Node 路径不同，记得同步修改 systemd 文件里的 `ExecStart`。
+
+## 4. Git 与仓库约定
+
+推荐使用你自己的 fork 作为部署源码来源。
+
+当前 fork：
+- `https://github.com/ccyy-lyx/gpt_image_playground`
+
+推荐新环境初始化方式：
+
+```bash
+gh auth login
+gh auth setup-git
+
+gh repo clone ccyy-lyx/gpt_image_playground /opt/gpt_image_playground_update
+cd /opt/gpt_image_playground_update
+git remote add upstream https://github.com/CookSleep/gpt_image_playground.git
+```
+
+不要做的事：
+
+- 不要把 PAT 写进 `git remote`
+- 不要把 `/root/.config/gh/hosts.yml` 提交到 GitHub
+- 不要把 `.env`、token、cookie、私钥写进仓库
+- 不要把临时命令里的 token 留在 shell history 或文档里
+
+如果你曾经误把 token 放进 remote URL，立刻改回普通 URL：
+
+```bash
+git remote set-url origin https://github.com/ccyy-lyx/gpt_image_playground.git
+```
+
+## 5. 部署 gpt-image-playground
+
+### 5.1 构建镜像
+
+在源码目录执行：
+
+```bash
+cd /opt/gpt_image_playground_update
+docker build -f deploy/Dockerfile -t gpt-image-playground:timeoutfix .
+```
+
+### 5.2 启动容器
+
+当前线上容器运行参数：
+
+- `DEFAULT_API_URL=https://api.567001.xyz/v1`
+- `ENABLE_API_PROXY=false`
+- `LOCK_API_PROXY=false`
+- 端口映射：`8081:80`
+- 网络：`sub2api_sub2api-network`
+- 重启策略：`unless-stopped`
+
+部署命令：
+
+```bash
+docker rm -f gpt-image-playground || true
+
+docker run -d \
+  --name gpt-image-playground \
+  --restart unless-stopped \
+  --network sub2api_sub2api-network \
+  -p 8081:80 \
+  -e DEFAULT_API_URL=https://api.567001.xyz/v1 \
+  -e ENABLE_API_PROXY=false \
+  -e LOCK_API_PROXY=false \
+  gpt-image-playground:timeoutfix
+```
+
+说明：
+
+- 这里显式关闭了容器内置 `API_PROXY`
+- 图片站前端默认直接请求 `https://api.567001.xyz/v1`
+- 这样可以避免再经过 `image.567001.xyz` 自己的 `/api-proxy` 链路
+
+## 6. 部署 image-sse-proxy
+
+### 6.1 创建目录
+
+```bash
+mkdir -p /opt/image-sse-proxy
+```
+
+### 6.2 写入 `/opt/image-sse-proxy/server.js`
+
+把以下完整内容写入：
+
+```js
+const http = require('http')
+const { URL } = require('url')
+
+const PORT = Number(process.env.PORT || 3101)
+const UPSTREAM_BASE = (process.env.UPSTREAM_BASE || 'http://127.0.0.1:3000').replace(/\/+$/, '')
+const HEARTBEAT_MS = Number(process.env.HEARTBEAT_MS || 15000)
+const UPSTREAM_TIMEOUT_MS = Number(process.env.UPSTREAM_TIMEOUT_MS || 600000)
+
+function writeJson(res, statusCode, payload) {
+  const body = JSON.stringify(payload)
+  res.writeHead(statusCode, {
+    'Content-Type': 'application/json; charset=utf-8',
+    'Content-Length': Buffer.byteLength(body),
+    'Cache-Control': 'no-store',
+  })
+  res.end(body)
+}
+
+function buildHeaders(req, bodyLength) {
+  const headers = { ...req.headers }
+  delete headers.host
+  headers.connection = 'keep-alive'
+  if (bodyLength != null) headers['content-length'] = String(bodyLength)
+  return headers
+}
+
+function normalizeImageResultPayload(payload) {
+  if (!payload || typeof payload !== 'object') return null
+  const data = Array.isArray(payload.data) ? payload.data : null
+  if (!data || data.length === 0) return null
+  const first = data[0] && typeof data[0] === 'object' ? data[0] : {}
+  const event = {
+    type: 'image_generation.completed',
+  }
+  if (typeof first.b64_json === 'string') event.b64_json = first.b64_json
+  if (typeof first.revised_prompt === 'string') event.revised_prompt = first.revised_prompt
+  if (typeof payload.size === 'string') event.size = payload.size
+  if (typeof payload.quality === 'string') event.quality = payload.quality
+  if (typeof payload.output_format === 'string') event.output_format = payload.output_format
+  if (typeof payload.moderation === 'string') event.moderation = payload.moderation
+  if (typeof payload.output_compression === 'number') event.output_compression = payload.output_compression
+  return event
+}
+
+async function readRequestBody(req) {
+  const chunks = []
+  for await (const chunk of req) chunks.push(chunk)
+  return Buffer.concat(chunks)
+}
+
+async function handleImageGeneration(req, res) {
+  let body
+  try {
+    body = await readRequestBody(req)
+  } catch (err) {
+    writeJson(res, 400, { error: 'failed to read request body' })
+    return
+  }
+
+  let payload
+  try {
+    payload = JSON.parse(body.toString('utf8') || '{}')
+  } catch {
+    writeJson(res, 400, { error: 'invalid json body' })
+    return
+  }
+
+  const requestPayload = { ...payload }
+  delete requestPayload.stream
+  delete requestPayload.partial_images
+
+  const upstreamUrl = new URL('/v1/images/generations', UPSTREAM_BASE)
+  const controller = new AbortController()
+  const timeout = setTimeout(() => controller.abort(), UPSTREAM_TIMEOUT_MS)
+
+  req.on('close', () => controller.abort())
+
+  res.writeHead(200, {
+    'Content-Type': 'text/event-stream; charset=utf-8',
+    'Cache-Control': 'no-cache, no-transform',
+    Connection: 'keep-alive',
+    'X-Accel-Buffering': 'no',
+    'Access-Control-Allow-Origin': req.headers.origin || '*',
+    'Access-Control-Allow-Credentials': 'true',
+  })
+  res.write(': stream-start\\n\\n')
+
+  const heartbeat = setInterval(() => {
+    if (!res.writableEnded) res.write(': ping\\n\\n')
+  }, HEARTBEAT_MS)
+
+  try {
+    const upstreamRes = await fetch(upstreamUrl, {
+      method: 'POST',
+      headers: buildHeaders(req, Buffer.byteLength(JSON.stringify(requestPayload))),
+      body: JSON.stringify(requestPayload),
+      signal: controller.signal,
+    })
+
+    if (!upstreamRes.ok) {
+      const errorText = await upstreamRes.text()
+      res.write(`data: ${JSON.stringify({ error: { message: errorText || `HTTP ${upstreamRes.status}` } })}\\n\\n`)
+      res.end('data: [DONE]\\n\\n')
+      return
+    }
+
+    const rawText = await upstreamRes.text()
+    let json
+    try {
+      json = JSON.parse(rawText)
+    } catch {
+      res.write(`data: ${JSON.stringify({ error: { message: 'upstream returned invalid json' } })}\\n\\n`)
+      res.end('data: [DONE]\\n\\n')
+      return
+    }
+
+    const event = normalizeImageResultPayload(json)
+    if (!event || !event.b64_json) {
+      res.write(`data: ${JSON.stringify({ error: { message: 'upstream did not return image data' } })}\\n\\n`)
+      res.end('data: [DONE]\\n\\n')
+      return
+    }
+
+    res.write(`data: ${JSON.stringify(event)}\\n\\n`)
+    res.end('data: [DONE]\\n\\n')
+  } catch (err) {
+    const message = err && typeof err === 'object' && 'name' in err && err.name === 'AbortError'
+      ? 'request aborted'
+      : (err instanceof Error ? err.message : String(err))
+    if (!res.writableEnded) {
+      res.write(`data: ${JSON.stringify({ error: { message } })}\\n\\n`)
+      res.end('data: [DONE]\\n\\n')
+    }
+  } finally {
+    clearInterval(heartbeat)
+    clearTimeout(timeout)
+  }
+}
+
+const server = http.createServer(async (req, res) => {
+  if (!req.url) {
+    writeJson(res, 400, { error: 'missing url' })
+    return
+  }
+
+  const url = new URL(req.url, `http://${req.headers.host || 'localhost'}`)
+
+  if (req.method === 'OPTIONS') {
+    res.writeHead(204, {
+      'Access-Control-Allow-Origin': req.headers.origin || '*',
+      'Access-Control-Allow-Credentials': 'true',
+      'Access-Control-Allow-Headers': req.headers['access-control-request-headers'] || '*',
+      'Access-Control-Allow-Methods': 'GET,POST,PUT,DELETE,OPTIONS',
+      'Access-Control-Max-Age': '43200',
+    })
+    res.end()
+    return
+  }
+
+  if (req.method === 'POST' && url.pathname === '/v1/images/generations') {
+    await handleImageGeneration(req, res)
+    return
+  }
+
+  const upstreamUrl = new URL(url.pathname + url.search, UPSTREAM_BASE)
+  const body = ['GET', 'HEAD'].includes(req.method || 'GET') ? undefined : await readRequestBody(req)
+  const controller = new AbortController()
+  const timeout = setTimeout(() => controller.abort(), UPSTREAM_TIMEOUT_MS)
+  req.on('close', () => controller.abort())
+
+  try {
+    const upstreamRes = await fetch(upstreamUrl, {
+      method: req.method,
+      headers: buildHeaders(req, body?.length),
+      body,
+      signal: controller.signal,
+    })
+    const headers = Object.fromEntries(upstreamRes.headers.entries())
+    res.writeHead(upstreamRes.status, headers)
+    if (upstreamRes.body) {
+      for await (const chunk of upstreamRes.body) res.write(chunk)
+    }
+    res.end()
+  } catch (err) {
+    writeJson(res, 502, { error: err instanceof Error ? err.message : String(err) })
+  } finally {
+    clearTimeout(timeout)
+  }
+})
+
+server.listen(PORT, '127.0.0.1', () => {
+  console.log(`image-sse-proxy listening on 127.0.0.1:${PORT}`)
+})
+```
+
+## 7. systemd service
+
+把以下内容写入 `/etc/systemd/system/image-sse-proxy.service`：
+
+```ini
+[Unit]
+Description=Image SSE Proxy for Cloudflare-safe image generation
+After=network.target
+
+[Service]
+Type=simple
+WorkingDirectory=/opt/image-sse-proxy
+ExecStart=/root/.nvm/versions/node/v22.22.1/bin/node /opt/image-sse-proxy/server.js
+Restart=always
+RestartSec=3
+Environment=PORT=3101
+Environment=UPSTREAM_BASE=http://127.0.0.1:3000
+Environment=HEARTBEAT_MS=15000
+Environment=UPSTREAM_TIMEOUT_MS=600000
+
+[Install]
+WantedBy=multi-user.target
+```
+
+启用并启动：
+
+```bash
+systemctl daemon-reload
+systemctl enable --now image-sse-proxy
+systemctl status --no-pager image-sse-proxy
+```
+
+如果 Node 路径不同，修改 `ExecStart=` 后再执行上面三条命令。
+
+## 8. api.567001.xyz 代理配置
+
+当前 1Panel 站点代理配置文件：
+
+- `/opt/1panel/www/sites/api.567001.xyz/proxy/root.conf`
+
+请确保内容至少包含以下规则：
+
+```nginx
+location = /v1/images/generations {
+    proxy_pass http://127.0.0.1:3101;
+    proxy_set_header Host $host;
+    proxy_set_header X-Real-IP $remote_addr;
+    proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+    proxy_set_header REMOTE-HOST $remote_addr;
+    proxy_set_header Upgrade $http_upgrade;
+    proxy_set_header Connection "";
+    proxy_set_header X-Forwarded-Proto $scheme;
+    proxy_set_header X-Forwarded-Port $server_port;
+    proxy_http_version 1.1;
+    proxy_buffering off;
+    proxy_request_buffering off;
+    gzip off;
+    add_header X-Accel-Buffering no always;
+    add_header Cache-Control "no-cache, no-transform" always;
+    add_header Strict-Transport-Security "max-age=31536000";
+    proxy_connect_timeout 60s;
+    proxy_send_timeout 600s;
+    proxy_read_timeout 600s;
+}
+
+location ^~ / {
+    proxy_pass http://127.0.0.1:3000;
+    proxy_set_header Host $host;
+    proxy_set_header X-Real-IP $remote_addr;
+    proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
+    proxy_set_header REMOTE-HOST $remote_addr;
+    proxy_set_header Upgrade $http_upgrade;
+    proxy_set_header Connection $http_connection;
+    proxy_set_header X-Forwarded-Proto $scheme;
+    proxy_set_header X-Forwarded-Port $server_port;
+    proxy_http_version 1.1;
+    add_header X-Cache $upstream_cache_status;
+    proxy_ssl_server_name off;
+    proxy_ssl_name $proxy_host;
+    proxy_buffering off;
+    proxy_request_buffering off;
+    gzip off;
+    add_header X-Accel-Buffering no always;
+    add_header Strict-Transport-Security "max-age=31536000";
+    proxy_connect_timeout 60s;
+    proxy_send_timeout 600s;
+    proxy_read_timeout 600s;
+}
+```
+
+重载 OpenResty：
+
+```bash
+docker exec 9203948e8da4 sh -lc 'openresty -t && openresty -s reload'
+```
+
+如果新环境不是这个容器 ID，请替换成你自己的 OpenResty/Nginx 容器或服务。
+
+## 9. image.567001.xyz 站点约定
+
+`image.567001.xyz` 站点应继续反代到：
+
+- `http://127.0.0.1:8081`
+
+不需要再启用 `gpt-image-playground` 容器内的 `/api-proxy` 作为主链路。
+
+## 10. 部署验证
+
+### 10.1 验证前端站点
+
+```bash
+curl -sk -o /dev/null -w '%{http_code}\n' https://image.567001.xyz/
+```
+
+期望：
+
+```text
+200
+```
+
+### 10.2 验证 `OPTIONS /v1/images/generations`
+
+```bash
+curl -isk \
+  -X OPTIONS https://api.567001.xyz/v1/images/generations \
+  -H 'Origin: https://image.567001.xyz' \
+  -H 'Access-Control-Request-Method: POST' \
+  -H 'Access-Control-Request-Headers: authorization,content-type'
+```
+
+期望：
+
+- `HTTP 204`
+- 存在 `Access-Control-Allow-Origin`
+
+### 10.3 验证 SSE 首包
+
+```bash
+timeout 12s curl -iskN \
+  https://api.567001.xyz/v1/images/generations \
+  -H 'Origin: https://image.567001.xyz' \
+  -H 'Authorization: Bearer INVALID' \
+  -H 'Content-Type: application/json' \
+  --data '{"model":"gpt-image-2","prompt":"test","size":"1024x1024","output_format":"png","moderation":"auto"}'
+```
+
+期望：
+
+- 返回 `HTTP 200`
+- `Content-Type: text/event-stream`
+- 首屏马上出现：
+
+```text
+: stream-start
+```
+
+只要首包能马上出来，说明 Cloudflare 链路已经不再等最终整包才出响应。
+
+## 11. 更新流程
+
+更新分两部分：
+
+1. 应用代码更新
+2. 部署链路确认
+
+### 11.1 更新你的 fork
+
+在本地开发机或服务器上：
+
+```bash
+cd /opt/gpt_image_playground_update
+git fetch upstream
+git checkout main
+git merge upstream/main
+git push origin main
+```
+
+如果你是先在修复分支上开发，就先合并修复分支到你自己的 `main`，再继续同步上游。
+
+### 11.2 服务器拉取最新代码
+
+```bash
+cd /opt/gpt_image_playground_update
+git fetch origin
+git checkout main
+git pull --ff-only origin main
+```
+
+### 11.3 重新构建并替换 `gpt-image-playground`
+
+```bash
+cd /opt/gpt_image_playground_update
+docker build -f deploy/Dockerfile -t gpt-image-playground:timeoutfix .
+
+docker rm -f gpt-image-playground || true
+
+docker run -d \
+  --name gpt-image-playground \
+  --restart unless-stopped \
+  --network sub2api_sub2api-network \
+  -p 8081:80 \
+  -e DEFAULT_API_URL=https://api.567001.xyz/v1 \
+  -e ENABLE_API_PROXY=false \
+  -e LOCK_API_PROXY=false \
+  gpt-image-playground:timeoutfix
+```
+
+### 11.4 如果 `image-sse-proxy` 改过，重启它
+
+```bash
+systemctl restart image-sse-proxy
+systemctl status --no-pager image-sse-proxy
+```
+
+### 11.5 检查 1Panel/OpenResty 配置有没有被覆盖
+
+重点检查：
+
+- `/opt/1panel/www/sites/api.567001.xyz/proxy/root.conf`
+
+因为 1Panel 面板操作、站点重建、代理规则调整都有可能覆盖这个文件。
+
+检查完后执行：
+
+```bash
+docker exec 9203948e8da4 sh -lc 'openresty -t && openresty -s reload'
+```
+
+### 11.6 最后做验证
+
+至少执行这三条：
+
+```bash
+curl -sk -o /dev/null -w '%{http_code}\n' https://image.567001.xyz/
+curl -sk -o /dev/null -w '%{http_code}\n' -X OPTIONS https://api.567001.xyz/v1/images/generations -H 'Origin: https://image.567001.xyz' -H 'Access-Control-Request-Method: POST'
+systemctl is-active image-sse-proxy
+```
+
+期望：
+
+- `200`
+- `204`
+- `active`
+
+## 12. 回滚流程
+
+如果新版本异常，按下面方式回滚：
+
+1. 回到上一个稳定 commit
+2. 重新构建 `gpt-image-playground`
+3. 保持 `image-sse-proxy` 和 OpenResty 配置不动
+
+示例：
+
+```bash
+cd /opt/gpt_image_playground_update
+git log --oneline -n 20
+git checkout <stable_commit>
+docker build -f deploy/Dockerfile -t gpt-image-playground:timeoutfix .
+docker rm -f gpt-image-playground || true
+docker run -d \
+  --name gpt-image-playground \
+  --restart unless-stopped \
+  --network sub2api_sub2api-network \
+  -p 8081:80 \
+  -e DEFAULT_API_URL=https://api.567001.xyz/v1 \
+  -e ENABLE_API_PROXY=false \
+  -e LOCK_API_PROXY=false \
+  gpt-image-playground:timeoutfix
+```
+
+## 13. 不要上传到 GitHub 的内容
+
+下面这些内容不要提交：
+
+- 任何 GitHub token
+- 任何 OpenAI / NewAPI / Cloudflare token
+- `/root/.config/gh/hosts.yml`
+- 带 token 的 remote URL
+- 临时导出的 `GITHUB_TOKEN`
+- 包含真实密钥的 `.env`
+- 面板导出的完整站点配置备份（如果里面含敏感信息）
+
+推荐做法：
+
+- 认证只用 `gh auth login`
+- 命令示例里只放占位符
+- 所有 secret 只保留在服务器本机
+- 提交前执行：
+
+```bash
+git diff --check
+git status
+```
+
+## 14. 建议
+
+这份文档可以安全放进 GitHub，因为它不包含 token。
+
+如果后面还要进一步固化，建议再补两样：
+
+1. 把 `image-sse-proxy` 独立成仓库内一个 `deploy-extra/` 目录
+2. 再做一个只含占位符的 `.env.example` 和 `docker-compose.example.yml`
+
+但就当前需求来说，这一个文件已经足够用于新环境重复部署与后续更新。
diff --git a/public/sw.js b/public/sw.js
index e2884c4f..4907b5d7 100644
--- a/public/sw.js
+++ b/public/sw.js
@@ -1,4 +1,4 @@
-const CACHE_NAME = 'gpt-image-playground-v0.1.5'
+const CACHE_NAME = 'gpt-image-playground-v0.4.6-direct-api-20260527b'
 const APP_SHELL = ['./', './index.html', './manifest.webmanifest', './pwa-icon.svg']
 
 self.addEventListener('install', (event) => {
@@ -27,7 +27,7 @@ self.addEventListener('fetch', (event) => {
 
   if (request.mode === 'navigate') {
     event.respondWith(
-      fetch(request)
+      fetch(request, { cache: 'reload' })
         .then((response) => {
           const copy = response.clone()
           caches.open(CACHE_NAME).then((cache) => cache.put('./index.html', copy))
@@ -42,7 +42,11 @@ self.addEventListener('fetch', (event) => {
     caches.match(request).then((cached) => {
       if (cached) return cached
 
-      return fetch(request).then((response) => {
+      const fetchRequest = url.pathname.startsWith('/assets/')
+        ? new Request(request, { cache: 'reload' })
+        : request
+
+      return fetch(fetchRequest).then((response) => {
         if (response.ok) {
           const copy = response.clone()
           caches.open(CACHE_NAME).then((cache) => cache.put(request, copy))
diff --git a/src/lib/openaiCompatibleImageApi.ts b/src/lib/openaiCompatibleImageApi.ts
index 544dd6e4..921d68da 100644
--- a/src/lib/openaiCompatibleImageApi.ts
+++ b/src/lib/openaiCompatibleImageApi.ts
@@ -19,6 +19,19 @@ import {
 } from './imageApiShared'
 
 const PROMPT_REWRITE_GUARD_PREFIX = 'Use the following text as the complete prompt. Do not rewrite it:'
+const MIN_LONG_RUNNING_OPENAI_TIMEOUT_SECONDS = 300
+
+function getLongRunningOpenAITimeoutMs(profile: Pick<ApiProfile, 'timeout'>): number {
+  return Math.max(profile.timeout, MIN_LONG_RUNNING_OPENAI_TIMEOUT_SECONDS) * 1000
+}
+
+function shouldStreamOpenAIImageRequest(profile: ApiProfile, customProvider?: CustomProviderDefinition | null): boolean {
+  return profile.streamImages || (!customProvider && profile.provider === 'openai')
+}
+
+function getOpenAIImagePartialImages(profile: ApiProfile): number {
+  return Math.max(getStreamPartialImages(profile), 3)
+}
 
 function getStreamPartialImages(profile: ApiProfile): number {
   return profile.streamPartialImages ?? DEFAULT_STREAM_PARTIAL_IMAGES
@@ -480,7 +493,7 @@ export async function callOpenAICompatibleImageApi(opts: CallApiOptions, profile
 
 async function callImagesApi(opts: CallApiOptions, profile: ApiProfile, customProvider?: CustomProviderDefinition | null): Promise<CallApiResult> {
   const n = opts.params.n > 0 ? opts.params.n : 1
-  if ((profile.codexCli || (profile.streamImages && n > 1)) && n > 1) {
+  if ((profile.codexCli || (shouldStreamOpenAIImageRequest(profile, customProvider) && n > 1)) && n > 1) {
     return callImagesApiConcurrent(opts, profile, n, customProvider)
   }
 
@@ -544,7 +557,7 @@ async function callImagesApiSingle(opts: CallApiOptions, profile: ApiProfile, cu
   const paths = createOpenAICompatiblePaths(customProvider)
 
   const controller = new AbortController()
-  const timeoutId = setTimeout(() => controller.abort(), profile.timeout * 1000)
+  const timeoutId = setTimeout(() => controller.abort(), getLongRunningOpenAITimeoutMs(profile))
 
   try {
     let response: Response
@@ -570,9 +583,9 @@ async function callImagesApiSingle(opts: CallApiOptions, profile: ApiProfile, cu
       if (profile.responseFormatB64Json) {
         formData.append('response_format', 'b64_json')
       }
-      if (profile.streamImages) {
+      if (shouldStreamOpenAIImageRequest(profile, customProvider)) {
         formData.append('stream', 'true')
-        formData.append('partial_images', String(getStreamPartialImages(profile)))
+        formData.append('partial_images', String(getOpenAIImagePartialImages(profile)))
       }
 
       const imageBlobs: Blob[] = []
@@ -632,9 +645,9 @@ async function callImagesApiSingle(opts: CallApiOptions, profile: ApiProfile, cu
       if (profile.responseFormatB64Json) {
         body.response_format = 'b64_json'
       }
-      if (profile.streamImages) {
+      if (shouldStreamOpenAIImageRequest(profile, customProvider)) {
         body.stream = true
-        body.partial_images = getStreamPartialImages(profile)
+        body.partial_images = getOpenAIImagePartialImages(profile)
       }
 
       response = await fetch(buildApiUrl(profile.baseUrl, paths.generationPath, proxyConfig, useApiProxy), {
@@ -653,7 +666,7 @@ async function callImagesApiSingle(opts: CallApiOptions, profile: ApiProfile, cu
       throw new Error(await getApiErrorMessage(response))
     }
 
-    if (profile.streamImages && isEventStreamResponse(response)) {
+    if (shouldStreamOpenAIImageRequest(profile, customProvider) && isEventStreamResponse(response)) {
       return parseImagesApiStreamResponse(response, mime, opts.onPartialImage)
     }
 
@@ -1005,7 +1018,7 @@ async function callResponsesImageApiSingle(opts: CallApiOptions, profile: ApiPro
   const useApiProxy = shouldUseApiProxy(profile.apiProxy, proxyConfig)
   const requestHeaders = createRequestHeaders(profile)
   const controller = new AbortController()
-  const timeoutId = setTimeout(() => controller.abort(), profile.timeout * 1000)
+  const timeoutId = setTimeout(() => controller.abort(), getLongRunningOpenAITimeoutMs(profile))
 
   try {
     if (opts.maskDataUrl) {
diff --git a/src/store.ts b/src/store.ts
index 33789b86..e9c10cf1 100644
--- a/src/store.ts
+++ b/src/store.ts
@@ -1590,16 +1590,19 @@ function failOpenAITaskIfStillRunning(taskId: string, error: string, now = Date.
   return true
 }
 
+const MIN_OPENAI_WATCHDOG_TIMEOUT_SECONDS = 300
+
 function scheduleOpenAIWatchdog(taskId: string, timeoutSeconds: number, profile?: TimeoutStreamingHintProfile | null) {
   clearOpenAIWatchdogTimer(taskId)
   const task = useStore.getState().tasks.find((item) => item.id === taskId)
   if (!task || !isRunningOpenAITask(task)) return
 
-  const timeoutMs = Math.max(0, timeoutSeconds * 1000)
+  const effectiveTimeoutSeconds = Math.max(timeoutSeconds, MIN_OPENAI_WATCHDOG_TIMEOUT_SECONDS)
+  const timeoutMs = Math.max(0, effectiveTimeoutSeconds * 1000)
   const remainingMs = Math.max(0, timeoutMs - (Date.now() - task.createdAt))
   const timer = setTimeout(() => {
     openAIWatchdogTimers.delete(taskId)
-    const failed = failOpenAITaskIfStillRunning(taskId, createOpenAITimeoutError(timeoutSeconds, profile))
+    const failed = failOpenAITaskIfStillRunning(taskId, createOpenAITimeoutError(effectiveTimeoutSeconds, profile))
     if (failed) useStore.getState().showToast('OpenAI 任务请求超时', 'error')
   }, remainingMs)
   openAIWatchdogTimers.set(taskId, timer)