基础生图：OpenAI 兼容

适合现有 OpenAI SDK 或 OpenAI 兼容客户端直接接入。

curl https://gpt222.shenfengwl.fun/v1/images/generations \
  -H "Authorization: Bearer <用户 API Key>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2",
    "prompt": "中国古代风格的庭院场景，细腻写实",
    "n": 1,
    "size": "1024x1024",
    "quality": "auto",
    "response_format": "b64_json"
  }'

Python SDK 示例

from openai import OpenAI
import base64

client = OpenAI(
    base_url="https://gpt222.shenfengwl.fun/v1",
    api_key="<用户 API Key>",
)

resp = client.images.generate(
    model="gpt-image-2",
    prompt="未来太空站内部，巨大舷窗外是地球和星空",
    n=1,
    size="1024x1024",
    quality="auto",
)

with open("out.png", "wb") as f:
    f.write(base64.b64decode(resp.data[0].b64_json))

图片编辑：单图 / 多图

编辑接口使用 multipart/form-data，单图和多图都重复使用 image 字段；遮罩图可用 mask 字段。

curl https://gpt222.shenfengwl.fun/v1/images/edits \
  -H "Authorization: Bearer <用户 API Key>" \
  -F "model=gpt-image-2" \
  -F "prompt=把背景改成日落海边，保持主体不变" \
  -F "response_format=b64_json" \
  -F "image=@input-1.png" \
  -F "image=@input-2.png"

异步任务队列

如果不想长时间占用 HTTP 连接，使用本站队列接口。提交后轮询 job 状态，成功后读取 images[].b64_json。

# 1. 提交任务
curl https://gpt222.shenfengwl.fun/v1/images/jobs \
  -H "Authorization: Bearer <用户 API Key>" \
  -H "Content-Type: application/json" \
  -d '{
    "endpoint": "generations",
    "model": "gpt-image-2",
    "prompt": "一只穿宇航服的柯基站在月球表面",
    "n": 4,
    "response_format": "b64_json"
  }'

# 2. 查询任务
curl https://gpt222.shenfengwl.fun/v1/images/jobs/job_xxx \
  -H "Authorization: Bearer <用户 API Key>"

Python 队列示例

遇到 Server disconnected without sending a response、Cloudflare/Nginx 超时、或图片生成时间较长时，使用这个方式最稳。

import base64
import time
import requests

BASE_URL = "https://gpt222.shenfengwl.fun"
API_KEY = "<用户 API Key>"
HEADERS = {"Authorization": f"Bearer {API_KEY}"}

submit = requests.post(
    f"{BASE_URL}/v1/images/jobs",
    headers={**HEADERS, "Content-Type": "application/json"},
    json={
        "endpoint": "generations",
        "model": "gpt-image-2",
        "prompt": "未来太空站内部，巨大舷窗外是地球和星空",
        "n": 1,
        "size": "1024x1024",
        "quality": "auto",
        "response_format": "b64_json",
    },
    timeout=30,
)
submit.raise_for_status()
job_id = submit.json()["id"]

while True:
    job = requests.get(
        f"{BASE_URL}/v1/images/jobs/{job_id}",
        headers=HEADERS,
        timeout=30,
    )
    job.raise_for_status()
    payload = job.json()

    if payload["status"] == "succeeded":
        for index, image in enumerate(payload["images"], start=1):
            with open(f"out-{index}.png", "wb") as f:
                f.write(base64.b64decode(image["b64_json"]))
        break

    if payload["status"] == "failed":
        raise RuntimeError(payload.get("error_message") or "image job failed")

    time.sleep(3)