gpt2api-openai-gateway

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

gpt2api OpenAI Gateway

gpt2api OpenAI 网关

Skill by ara.so — Daily 2026 Skills collection.

gpt2api

is a self-hosted OpenAI-compatible API gateway that reverse-engineers

chatgpt.com

to expose GPT Image 2 / DALL·E 3 / IMG2 grayscale capabilities via standard

/v1/images/generations

/v1/images/edits

, and

/v1/chat/completions

endpoints. It provides multi-account pooling, proxy pooling, rate limiting, credit billing, and an admin dashboard.

由 ara.so 开发的Skill — 每日2026技能合集。

gpt2api

是一款自托管的兼容OpenAI的API网关，通过逆向工程

chatgpt.com

，将GPT Image 2 / DALL·E 3 / IMG2灰度图生成能力通过标准的

/v1/images/generations

、

/v1/images/edits

和

/v1/chat/completions

端点暴露出来。它支持多账号池、代理池、速率限制、信用计费以及管理控制台。

Installation & Deployment

安装与部署

Docker Compose (Recommended)

Docker Compose（推荐）

bash

git clone https://github.com/432539/gpt2api.git
cd gpt2api/deploy
cp .env.example .env

Edit

.env

— these three are required:

env

JWT_SECRET=<generate with: openssl rand -base64 48 | tr -d '=/+' | cut -c1-48>
CRYPTO_AES_KEY=<generate with: openssl rand -hex 32>
MYSQL_ROOT_PASSWORD=<strong-password>
MYSQL_PASSWORD=<strong-password>

Generate secrets:

bash

undefined

bash

git clone https://github.com/432539/gpt2api.git
cd gpt2api/deploy
cp .env.example .env

编辑

.env

文件 — 以下三项为必填项：

env

JWT_SECRET=<使用以下命令生成: openssl rand -base64 48 | tr -d '=/+' | cut -c1-48>
CRYPTO_AES_KEY=<使用以下命令生成: openssl rand -hex 32>
MYSQL_ROOT_PASSWORD=<强密码>
MYSQL_PASSWORD=<强密码>

生成密钥：

bash

undefined

CRYPTO_AES_KEY (must be exactly 64 hex chars = 32 bytes AES-256)

CRYPTO_AES_KEY（必须恰好64个十六进制字符 = 32字节AES-256）

openssl rand -hex 32

JWT_SECRET (>=32 chars)

JWT_SECRET（≥32个字符）

openssl rand -base64 48 | tr -d '=/+' | cut -c1-48


Start services:

```bash
docker compose up -d --build
docker compose logs -f server

On startup the server automatically:

Waits for MySQL health check
Runs
```
goose up
```
database migrations
Starts HTTP on
```
:8080
```

Default admin credentials (change immediately):

URL:
```
http://<server-ip>:8080/
```
Email:
```
admin@example.com
```
Password:
```
admin123
```

openssl rand -base64 48 | tr -d '=/+' | cut -c1-48


启动服务：

```bash
docker compose up -d --build
docker compose logs -f server

启动时服务器会自动执行以下操作：

等待MySQL健康检查通过
运行
```
goose up
```
数据库迁移
在
```
:8080
```
端口启动HTTP服务

默认管理员凭据（请立即修改）：

地址：
```
http://<服务器IP>:8080/
```
邮箱：
```
admin@example.com
```
密码：
```
admin123
```

Configuration

配置

Main config file:

configs/config.yaml

(override with

GPT2API_*

env vars in Docker).

yaml

app:
  listen: ":8080"
  base_url: "https://your-domain.com"   # used for signed image proxy URLs

mysql:
  dsn: "user:pass@tcp(mysql:3306)/gpt2api?parseTime=true"
  max_open_conns: 500

redis:
  addr: "redis:6379"
  pool_size: 500                          # needed for distributed locks & rate limiting

jwt:
  secret: "${JWT_SECRET}"
  access_ttl_sec: 7200
  refresh_ttl_sec: 604800

crypto:
  aes_key: "${CRYPTO_AES_KEY}"           # 64-char hex, encrypts account AT/cookies

scheduler:
  min_interval_sec: 10                   # minimum seconds between uses of same account
  daily_usage_ratio: 0.8                 # daily usage cap ratio before circuit break
  cooldown_429_sec: 300                  # backoff when upstream returns 429

Environment variable override pattern (Docker):

env

GPT2API_APP_BASE_URL=https://api.example.com
GPT2API_SCHEDULER_MIN_INTERVAL_SEC=15
GPT2API_SCHEDULER_COOLDOWN_429_SEC=600

主配置文件：

configs/config.yaml

（可通过Docker中的

GPT2API_*

环境变量覆盖）。

yaml

app:
  listen: ":8080"
  base_url: "https://your-domain.com"   # 用于生成签名图片代理URL

mysql:
  dsn: "user:pass@tcp(mysql:3306)/gpt2api?parseTime=true"
  max_open_conns: 500

redis:
  addr: "redis:6379"
  pool_size: 500                          # 分布式锁和速率限制所需

jwt:
  secret: "${JWT_SECRET}"
  access_ttl_sec: 7200
  refresh_ttl_sec: 604800

crypto:
  aes_key: "${CRYPTO_AES_KEY}"           # 64字符十六进制字符串，用于加密账号AT/cookies

scheduler:
  min_interval_sec: 10                   # 同一账号两次使用的最小间隔秒数
  daily_usage_ratio: 0.8                 # 达到每日限额的该比例后触发熔断
  cooldown_429_sec: 300                  # 上游返回429错误后的冷却时间

环境变量覆盖规则（Docker）：

env

GPT2API_APP_BASE_URL=https://api.example.com
GPT2API_SCHEDULER_MIN_INTERVAL_SEC=15
GPT2API_SCHEDULER_COOLDOWN_429_SEC=600

Quick Start: First Image Generation

快速开始：首次图片生成

1. Add a Proxy

1. 添加代理

Via admin dashboard → Proxy Management → New Proxy, or via API:

bash

curl -X POST http://localhost:8080/api/admin/proxies \
  -H "Authorization: Bearer $ADMIN_JWT" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "proxy-us-01",
    "url": "http://user:pass@proxy.example.com:8080",
    "type": "http"
  }'

SOCKS5 example:

bash

-d '{"name":"socks-01","url":"socks5://user:pass@proxy.example.com:1080","type":"socks5"}'

通过管理控制台 → 代理管理 → 新增代理，或通过API：

bash

curl -X POST http://localhost:8080/api/admin/proxies \
  -H "Authorization: Bearer $ADMIN_JWT" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "proxy-us-01",
    "url": "http://user:pass@proxy.example.com:8080",
    "type": "http"
  }'

SOCKS5示例：

bash

-d '{"name":"socks-01","url":"socks5://user:pass@proxy.example.com:1080","type":"socks5"}'

2. Import ChatGPT Accounts

2. 导入ChatGPT账号

Batch import via admin dashboard → GPT Accounts → Batch Import.

Supported formats — JSON session:

json

[
  {
    "email": "user@example.com",
    "access_token": "$CHATGPT_ACCESS_TOKEN",
    "refresh_token": "$CHATGPT_REFRESH_TOKEN",
    "proxy_id": 1
  }
]

Or AT/RT/ST plain text (one per line in the UI).

通过管理控制台 → GPT账号 → 批量导入。

支持的格式 — JSON会话：

json

[
  {
    "email": "user@example.com",
    "access_token": "$CHATGPT_ACCESS_TOKEN",
    "refresh_token": "$CHATGPT_REFRESH_TOKEN",
    "proxy_id": 1
  }
]

或AT/RT/ST纯文本（UI中每行一个）。

3. Create an API Key

3. 创建API密钥

Admin dashboard → User Management → select user → API Keys → Create Key.

Or programmatically:

bash

curl -X POST http://localhost:8080/api/user/apikeys \
  -H "Authorization: Bearer $USER_JWT" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "my-app-key",
    "rpm_limit": 60,
    "daily_quota": 1000,
    "model_whitelist": ["gpt-image-2", "picture_v2"]
  }'

管理控制台 → 用户管理 → 选择用户 → API密钥 → 创建密钥。

或通过编程方式：

bash

curl -X POST http://localhost:8080/api/user/apikeys \
  -H "Authorization: Bearer $USER_JWT" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "my-app-key",
    "rpm_limit": 60,
    "daily_quota": 1000,
    "model_whitelist": ["gpt-image-2", "picture_v2"]
  }'

4. Generate an Image

4. 生成图片

bash

curl -X POST http://localhost:8080/v1/images/generations \
  -H "Authorization: Bearer sk-your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "A serene mountain lake at sunset, photorealistic",
    "n": 2,
    "size": "1024x1024"
  }'

Response:

json

{
  "created": 1713456789,
  "data": [
    {
      "url": "https://your-domain.com/p/img/task123/0?exp=1713460389&sig=abc123"
    },
    {
      "url": "https://your-domain.com/p/img/task123/1?exp=1713460389&sig=def456"
    }
  ]
}

bash

curl -X POST http://localhost:8080/v1/images/generations \
  -H "Authorization: Bearer sk-your-api-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "A serene mountain lake at sunset, photorealistic",
    "n": 2,
    "size": "1024x1024"
  }'

响应：

json

{
  "created": 1713456789,
  "data": [
    {
      "url": "https://your-domain.com/p/img/task123/0?exp=1713460389&sig=abc123"
    },
    {
      "url": "https://your-domain.com/p/img/task123/1?exp=1713460389&sig=def456"
    }
  ]
}

API Reference

API参考

Image Generation

图片生成

POST /v1/images/generations

json

{
  "model": "gpt-image-2",      // or "picture_v2" for IMG2 grayscale
  "prompt": "your prompt",
  "n": 1,                       // number of images (1-4 recommended)
  "size": "1024x1024",          // "1024x1024" | "1792x1024" | "1024x1792"
  "response_format": "url"      // "url" | "b64_json"
}

POST /v1/images/generations

json

{
  "model": "gpt-image-2",      // 或使用"picture_v2"生成IMG2灰度图
  "prompt": "your prompt",
  "n": 1,                       // 图片数量（推荐1-4）
  "size": "1024x1024",          // "1024x1024" | "1792x1024" | "1024x1792"
  "response_format": "url"      // "url" | "b64_json"
}

Image Edit (img2img)

图片编辑（img2img）

POST /v1/images/edits
Content-Type: multipart/form-data

bash

curl -X POST http://localhost:8080/v1/images/edits \
  -H "Authorization: Bearer sk-your-api-key" \
  -F "image=@input.png" \
  -F "prompt=Make the sky more dramatic" \
  -F "model=gpt-image-2" \
  -F "n=1"

POST /v1/images/edits
Content-Type: multipart/form-data

bash

curl -X POST http://localhost:8080/v1/images/edits \
  -H "Authorization: Bearer sk-your-api-key" \
  -F "image=@input.png" \
  -F "prompt=Make the sky more dramatic" \
  -F "model=gpt-image-2" \
  -F "n=1"

Check Task Status

检查任务状态

GET /v1/images/tasks/:id
Authorization: Bearer sk-your-api-key

GET /v1/images/tasks/:id
Authorization: Bearer sk-your-api-key

List Models

列出模型

GET /v1/models
Authorization: Bearer sk-your-api-key

GET /v1/models
Authorization: Bearer sk-your-api-key

Chat Completions (preserved, UI disabled)

聊天补全（已实现，UI入口禁用）

POST /v1/chat/completions

json

{
  "model": "gpt-4o",
  "messages": [{"role": "user", "content": "Hello"}],
  "stream": true
}

POST /v1/chat/completions

json

{
  "model": "gpt-4o",
  "messages": [{"role": "user", "content": "Hello"}],
  "stream": true
}

Using with OpenAI SDK

与OpenAI SDK配合使用

Python

python

from openai import OpenAI
import os

client = OpenAI(
    api_key=os.environ["GPT2API_KEY"],          # your sk- key from the dashboard
    base_url="http://localhost:8080/v1"          # or your production domain
)

python

from openai import OpenAI
import os

client = OpenAI(
    api_key=os.environ["GPT2API_KEY"],          # 从控制台获取的sk-密钥
    base_url="http://localhost:8080/v1"          # 或你的生产域名
)

Single image

单张图片

response = client.images.generate( model="gpt-image-2", prompt="A futuristic city skyline at night, cyberpunk style", n=1, size="1024x1024" ) print(response.data[0].url)

Batch images

批量图片

response = client.images.generate( model="gpt-image-2", prompt="Abstract watercolor painting, vibrant colors", n=4, size="1024x1792" # 9:16 portrait ) for img in response.data: print(img.url)

undefined

response = client.images.generate( model="gpt-image-2", prompt="Abstract watercolor painting, vibrant colors", n=4, size="1024x1792" # 9:16竖版 ) for img in response.data: print(img.url)

undefined

Node.js / TypeScript

typescript

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.GPT2API_KEY,
  baseURL: process.env.GPT2API_BASE_URL ?? "http://localhost:8080/v1",
});

async function generateImages(prompt: string, count: number = 2) {
  const response = await client.images.generate({
    model: "gpt-image-2",
    prompt,
    n: count,
    size: "1024x1024",
  });
  return response.data.map((img) => img.url);
}

// Image edit
async function editImage(imagePath: string, prompt: string) {
  const fs = await import("fs");
  const response = await client.images.edit({
    image: fs.createReadStream(imagePath),
    prompt,
    model: "gpt-image-2",
  });
  return response.data[0].url;
}

typescript

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: process.env.GPT2API_KEY,
  baseURL: process.env.GPT2API_BASE_URL ?? "http://localhost:8080/v1",
});

async function generateImages(prompt: string, count: number = 2) {
  const response = await client.images.generate({
    model: "gpt-image-2",
    prompt,
    n: count,
    size: "1024x1024",
  });
  return response.data.map((img) => img.url);
}

// 图片编辑
async function editImage(imagePath: string, prompt: string) {
  const fs = await import("fs");
  const response = await client.images.edit({
    image: fs.createReadStream(imagePath),
    prompt,
    model: "gpt-image-2",
  });
  return response.data[0].url;
}

Go

package main

import (
    "context"
    "fmt"
    "os"

    openai "github.com/sashabaranov/go-openai"
)

func main() {
    cfg := openai.DefaultConfig(os.Getenv("GPT2API_KEY"))
    cfg.BaseURL = os.Getenv("GPT2API_BASE_URL") // e.g. "http://localhost:8080/v1"
    client := openai.NewClientWithConfig(cfg)

    resp, err := client.CreateImage(context.Background(), openai.ImageRequest{
        Model:  "gpt-image-2",
        Prompt: "A tranquil Japanese garden with cherry blossoms",
        N:      2,
        Size:   openai.CreateImageSize1024x1024,
    })
    if err != nil {
        panic(err)
    }
    for _, img := range resp.Data {
        fmt.Println(img.URL)
    }
}

package main

import (
    "context"
    "fmt"
    "os"

    openai "github.com/sashabaranov/go-openai"
)

func main() {
    cfg := openai.DefaultConfig(os.Getenv("GPT2API_KEY"))
    cfg.BaseURL = os.Getenv("GPT2API_BASE_URL") // 例如 "http://localhost:8080/v1"
    client := openai.NewClientWithConfig(cfg)

    resp, err := client.CreateImage(context.Background(), openai.ImageRequest{
        Model:  "gpt-image-2",
        Prompt: "A tranquil Japanese garden with cherry blossoms",
        N:      2,
        Size:   openai.CreateImageSize1024x1024,
    })
    if err != nil {
        panic(err)
    }
    for _, img := range resp.Data {
        fmt.Println(img.URL)
    }
}

Key Concepts

核心概念

IMG2 Grayscale Hit

IMG2灰度图命中

IMG2 is a grayscale (A/B test) feature on

chatgpt.com

that returns multiple high-res final images in a single call. The gateway:

Sends the generation request
Detects
```
preview_only
```
in the tool message response
Automatically retries up to 3 times within the same conversation to capture IMG2 output
Returns all signed image URLs aggregated

Use model

picture_v2

to specifically target the IMG2 grayscale path:

json

{ "model": "picture_v2", "prompt": "...", "n": 1 }

Successful IMG2 hit shows in logs:

image runner result summary turns_used=1 is_preview=false signed_count=2

IMG2是

chatgpt.com

上的灰度图（A/B测试）功能，单次调用可返回多张高清最终图片。网关的处理流程：

发送生成请求
检测工具消息响应中的
```
preview_only
```
标记
在同一会话中自动重试最多3次以捕获IMG2输出
返回所有聚合后的签名图片URL

使用

picture_v2

模型可专门针对IMG2灰度图路径：

json

{ "model": "picture_v2", "prompt": "...", "n": 1 }

成功命中IMG2会在日志中显示：

image runner result summary turns_used=1 is_preview=false signed_count=2

Account Pool Scheduling

账号池调度

The scheduler uses Redis distributed locks to serialize account usage:

```
min_interval_sec
```
: minimum gap between consecutive uses of the same account
```
daily_usage_ratio
```
: if an account hits this fraction of its daily limit, it's circuit-broken
```
cooldown_429_sec
```
: account is cooled down after receiving a 429 from upstream
Accounts are bound 1:1 to
```
oai-device-id
```
/
```
oai-session-id
```
fingerprints (stored in DB)

调度器使用Redis分布式锁来序列化账号使用：

```
min_interval_sec
```
: 同一账号连续使用的最小间隔秒数
```
daily_usage_ratio
```
: 账号使用量达到每日限额的该比例后触发熔断
```
cooldown_429_sec
```
: 账号收到上游429错误后的冷却时间
账号与
```
oai-device-id
```
/
```
oai-session-id
```
指纹一对一绑定（存储在数据库中）

Image Proxy & Anti-Hotlink

图片代理与防盗链

All returned image URLs go through the built-in HMAC-signed proxy:

/p/img/:task_id/:index?exp=<unix_timestamp>&sig=<hmac_sha256>

This bypasses

chatgpt.com

estuary/content

403 hotlink protection. The proxy fetches and streams the image server-side when accessed.

所有返回的图片URL都经过内置的HMAC签名代理：

/p/img/:task_id/:index?exp=<unix_timestamp>&sig=<hmac_sha256>

这绕过了

chatgpt.com

的

estuary/content

403防盗链限制。代理会在访问时从服务器端获取并流式传输图片。

Credit Billing System

信用计费系统

Users have a credit wallet; credits are pre-deducted before generation
Group multipliers allow VIP / internal / reseller pricing tiers
Billing settled after successful generation; failed calls are refunded
EPay (易支付) integration for recharge

用户拥有信用钱包；生成前预扣除信用
分组倍率支持VIP/内部/经销商等不同定价层级
生成成功后结算账单；调用失败则退款
集成易支付（EPay）用于充值

Admin Operations

管理员操作

Account Pool Management

账号池管理

bash

undefined

bash

undefined

Check account statuses via API

通过API检查账号状态

curl http://localhost:8080/api/admin/accounts
-H "Authorization: Bearer $ADMIN_JWT"

Manually trigger account token refresh

手动触发账号令牌刷新

curl -X POST http://localhost:8080/api/admin/accounts/123/refresh
-H "Authorization: Bearer $ADMIN_JWT"

Bulk import accounts (JSON body)

批量导入账号（JSON请求体）

curl -X POST http://localhost:8080/api/admin/accounts/batch
-H "Authorization: Bearer $ADMIN_JWT"
-H "Content-Type: application/json"
-d @accounts.json

undefined

curl -X POST http://localhost:8080/api/admin/accounts/batch
-H "Authorization: Bearer $ADMIN_JWT"
-H "Content-Type: application/json"
-d @accounts.json

undefined

Database Backup & Restore

数据库备份与恢复

Via admin dashboard → Data Backup, or:

bash

undefined

通过管理控制台 → 数据备份，或使用命令：

bash

undefined

Trigger backup

触发备份

curl -X POST http://localhost:8080/api/admin/backups
-H "Authorization: Bearer $ADMIN_JWT"

List backups

列出备份

curl http://localhost:8080/api/admin/backups
-H "Authorization: Bearer $ADMIN_JWT"

Download backup

下载备份

curl http://localhost:8080/api/admin/backups/backup-2026-04-21.sql.gz
-H "Authorization: Bearer $ADMIN_JWT"
-o backup.sql.gz

undefined

curl http://localhost:8080/api/admin/backups/backup-2026-04-21.sql.gz
-H "Authorization: Bearer $ADMIN_JWT"
-o backup.sql.gz

undefined

User Credit Management

用户信用管理

bash

undefined

bash

undefined

Add credits to a user

为用户添加信用

curl -X POST http://localhost:8080/api/admin/users/42/credits
-H "Authorization: Bearer $ADMIN_JWT"
-H "Content-Type: application/json"
-d '{"amount": 10000, "note": "Manual top-up"}'

---

curl -X POST http://localhost:8080/api/admin/users/42/credits
-H "Authorization: Bearer $ADMIN_JWT"
-H "Content-Type: application/json"
-d '{"amount": 10000, "note": "Manual top-up"}'

---

Directory Structure

目录结构

gpt2api/
├── cmd/server/          # main entry point
├── configs/
│   ├── config.yaml      # main config
│   └── config.example.yaml
├── deploy/
│   ├── docker-compose.yml
│   ├── .env.example
│   └── README.md        # horizontal scaling guide
├── internal/
│   ├── api/             # Gin route handlers
│   ├── chatgpt/         # upstream client (utls, sentinel, SSE parser)
│   ├── scheduler/       # account pool scheduler + Redis leases
│   ├── billing/         # credit wallet + EPay integration
│   ├── proxy/           # proxy pool health checks
│   ├── imgproxy/        # HMAC-signed image proxy
│   └── model/           # DB models + migrations (goose)
├── web/                 # Vue 3 frontend source
└── docs/screenshots/

gpt2api/
├── cmd/server/          # 主入口
├── configs/
│   ├── config.yaml      # 主配置文件
│   └── config.example.yaml
├── deploy/
│   ├── docker-compose.yml
│   ├── .env.example
│   └── README.md        # 水平扩展指南
├── internal/
│   ├── api/             # Gin路由处理器
│   ├── chatgpt/         # 上游客户端（utls、sentinel、SSE解析器）
│   ├── scheduler/       # 账号池调度器 + Redis租赁
│   ├── billing/         # 信用钱包 + EPay集成
│   ├── proxy/           # 代理池健康检查
│   ├── imgproxy/        # HMAC签名图片代理
│   └── model/           # 数据库模型 + 迁移（goose）
├── web/                 # Vue 3前端源码
└── docs/screenshots/

Enabling Chat Completions

启用聊天补全

The

/v1/chat/completions

route is fully implemented but the UI entry is disabled via a feature flag due to upstream sentinel instability. To re-enable:

// internal/api/routes.go — find the feature flag:
const enableChatCompletions = false   // change to true

Rebuild and redeploy:

bash

docker compose up -d --build server

/v1/chat/completions

路由已完全实现，但由于上游sentinel不稳定，通过功能标志禁用了UI入口。如需重新启用：

// internal/api/routes.go — 找到功能标志:
const enableChatCompletions = false   // 修改为true

重新构建并部署：

bash

docker compose up -d --build server

Troubleshooting

故障排查

No accounts available / all accounts idle

无可用账号 / 所有账号闲置

bash

undefined

bash

undefined

Check account pool status

检查账号池状态

docker compose logs server | grep "scheduler"

Common causes:

常见原因:

1. min_interval_sec too high — reduce in config

1. min_interval_sec设置过高 — 在配置中降低

2. All accounts in cooldown — check for 429s upstream

2. 所有账号处于冷却状态 — 检查上游是否返回429错误

3. Proxy unhealthy — verify proxy health score in dashboard

3. 代理不健康 — 在控制台中查看代理健康评分

undefined

undefined

Images return 403 when accessed directly

直接访问图片返回403

Expected behavior — all image URLs must go through the

/p/img/

proxy. Ensure

base_url

in config matches your actual domain so signed URLs are correct.

预期行为 — 所有图片URL必须通过

/p/img/

代理访问。确保配置中的

base_url

与实际域名一致，以便生成正确的签名URL。

is_preview=true

in logs (IMG2 not hitting)

日志中显示

is_preview=true

（未命中IMG2）

The account doesn't have IMG2 grayscale enabled. Either:

Try different accounts (grayscale is account-specific)
Use
```
gpt-image-2
```
model instead of
```
picture_v2
```
The gateway retries up to 3 times automatically

该账号未启用IMG2灰度图功能。可尝试：

使用其他账号（灰度图功能是账号专属的）
使用
```
gpt-image-2
```
模型而非
```
picture_v2
```
网关会自动重试最多3次

Redis lock timeout errors

Redis锁超时错误

yaml

redis:
  pool_size: 500    # increase if seeing lock contention under high concurrency

yaml

redis:
  pool_size: 500    # 高并发下出现锁竞争时可增大该值

Database migration fails on startup

启动时数据库迁移失败

bash

docker compose exec server goose -dir /app/migrations mysql "$DSN" status
docker compose exec server goose -dir /app/migrations mysql "$DSN" up

bash

docker compose exec server goose -dir /app/migrations mysql "$DSN" status
docker compose exec server goose -dir /app/migrations mysql "$DSN" up

TLS fingerprint / JA3 detection

TLS指纹 / JA3检测

The gateway uses

refraction-networking/utls

to mimic Edge 143 browser TLS fingerprint. If accounts are getting flagged:

Ensure each account is bound to a dedicated proxy (avoid IP mixing)
Check
```
oai-device-id
```
binding is consistent (stored per-account in DB)
Rotate accounts that have been flagged

网关使用

refraction-networking/utls

模拟Edge 143浏览器的TLS指纹。如果账号被标记：

确保每个账号绑定专属代理（避免IP混用）
检查
```
oai-device-id
```
绑定是否一致（存储在数据库的账号记录中）
更换已被标记的账号

High memory usage

内存占用过高

Reduce connection pool sizes for smaller deployments:

yaml

mysql:
  max_open_conns: 50
redis:
  pool_size: 50

对于小型部署，可减小连接池大小：

yaml

mysql:
  max_open_conns: 50
redis:
  pool_size: 50

Security Checklist

安全检查清单

Change default admin password immediately after first login
Set strong
```
JWT_SECRET
```
(≥32 chars random)
Set
```
CRYPTO_AES_KEY
```
to exactly 64 hex chars
Set strong MySQL passwords
Configure
```
base_url
```
to your HTTPS domain
Set up nginx reverse proxy with TLS in front of
```
:8080
```
Configure API key IP whitelists for sensitive keys
Enable audit log review in admin dashboard

首次登录后立即修改默认管理员密码
设置强
```
JWT_SECRET
```
（≥32位随机字符）
设置
```
CRYPTO_AES_KEY
```
为恰好64位十六进制字符
设置强MySQL密码
将
```
base_url
```
配置为HTTPS域名
在
```
:8080
```
前端配置带TLS的nginx反向代理
为敏感API密钥配置IP白名单
在管理控制台中启用审计日志查看