JunFeiAI 图片接口文档：Responses API 方案

适用对象：需要通过 junfeiai.com 中转站调用图片生成模型的客户。

最后更新：2026-05-17

1. 推荐方案

当前推荐使用 Responses API + image_generation 工具 调用图片模型：

POST https://junfeiai.com/v1/responses

不再建议新接入客户优先使用 /v1/images/generations。Images API 属于兼容路径，适合已经按 OpenAI Images API 对接好的旧客户端继续使用；新项目建议直接走 /v1/responses，因为它能更直接暴露图片工具的流式事件，便于观察 generating、partial_image、completed 等阶段。

完整请求地址：

https://junfeiai.com/v1/responses

2. 准备 API Key

1. 登录 https://junfeiai.com
2. 进入控制台的 API Keys 页面
3. 创建或复制一个 API Key
4. 请求时放入 HTTP Header：

Authorization: Bearer sk-xxxx
Content-Type: application/json

注意：API Key 属于敏感凭证，不要放在前端网页、App 客户端、公开仓库或聊天截图里。建议由你的服务端调用君飞 AI 接口，再把生成结果返回给自己的前端。

3. Responses API 文生图

请求

POST https://junfeiai.com/v1/responses
Authorization: Bearer <你的 API Key>
Content-Type: application/json
Accept: text/event-stream

推荐请求体

{
  "model": "gpt-5.5",
  "input": [
    {
      "type": "message",
      "role": "user",
      "content": [
        {
          "type": "input_text",
          "text": "请生成一张简洁高级的科技产品宣传图，白色背景，主体是一台银色智能音箱，柔和棚拍光线，商业摄影风格"
        }
      ]
    }
  ],
  "stream": true,
  "store": false,
  "parallel_tool_calls": true,
  "tool_choice": {
    "type": "image_generation"
  },
  "reasoning": {
    "effort": "xhigh",
    "summary": "auto"
  },
  "tools": [
    {
      "type": "image_generation",
      "model": "gpt-image-2",
      "size": "1024x1024",
      "quality": "medium",
      "output_format": "png",
      "partial_images": 1
    }
  ]
}

常用参数

4. curl 示例

curl -N -X POST "https://junfeiai.com/v1/responses" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Accept: text/event-stream" \
  -d '{
    "model": "gpt-5.5",
    "input": [
      {
        "type": "message",
        "role": "user",
        "content": [
          {
            "type": "input_text",
            "text": "Create a premium product poster for a silver smart speaker on a clean white studio background, soft commercial lighting, realistic photography style, no text, no watermark."
          }
        ]
      }
    ],
    "stream": true,
    "store": false,
    "parallel_tool_calls": true,
    "tool_choice": { "type": "image_generation" },
    "reasoning": { "effort": "xhigh", "summary": "auto" },
    "tools": [
      {
        "type": "image_generation",
        "model": "gpt-image-2",
        "size": "1024x1024",
        "quality": "medium",
        "output_format": "png",
        "partial_images": 1
      }
    ]
  }'

curl -N 用于关闭 curl 的输出缓冲，便于实时看到 SSE 事件。

5. SSE 事件解析

开启 stream: true 后，接口返回 text/event-stream。正常情况下会看到类似事件：

event: response.created
data: {"type":"response.created","response":{...}}

event: response.image_generation_call.in_progress
data: {"type":"response.image_generation_call.in_progress","item_id":"..."}

event: response.image_generation_call.generating
data: {"type":"response.image_generation_call.generating","item_id":"..."}

event: response.image_generation_call.partial_image
data: {"type":"response.image_generation_call.partial_image","partial_image_index":0,"partial_image":"..."}

event: response.output_item.done
data: {"type":"response.output_item.done","item":{"type":"image_generation_call","result":"..."}}

event: response.completed
data: {"type":"response.completed","response":{...}}

客户端处理建议：

1. 按 SSE 协议逐段读取 event: 和 data:
2. 对每段 data: 做 JSON 解析
3. 遇到 response.image_generation_call.partial_image 时，可把 partial_image 或同类图片字段作为预览图
4. 遇到 response.output_item.done 或 response.completed 时，从 item.result 或 response.output[].result 里读取最终图片
5. 图片字段通常是 Base64 字符串，需要解码后保存为 .png、.jpeg 或 .webp

不同上游版本的图片字段名可能略有差异，建议客户端兼容以下字段：

result
b64_json
image
image_b64
partial_image
partial_image_b64

6. Node.js 流式示例

下面示例使用原生 fetch 读取 SSE。生产环境建议加上超时、重试和日志记录。

import fs from "node:fs";

const response = await fetch("https://junfeiai.com/v1/responses", {
  method: "POST",
  headers: {
    "Authorization": "Bearer YOUR_API_KEY",
    "Content-Type": "application/json",
    "Accept": "text/event-stream"
  },
  body: JSON.stringify({
    model: "gpt-5.5",
    input: [
      {
        type: "message",
        role: "user",
        content: [
          {
            type: "input_text",
            text: "请生成一张简洁高级的科技产品宣传图，白色背景，主体是一台银色智能音箱，柔和棚拍光线，商业摄影风格"
          }
        ]
      }
    ],
    stream: true,
    store: false,
    parallel_tool_calls: true,
    tool_choice: { type: "image_generation" },
    reasoning: { effort: "xhigh", summary: "auto" },
    tools: [
      {
        type: "image_generation",
        model: "gpt-image-2",
        size: "1024x1024",
        quality: "medium",
        output_format: "png",
        partial_images: 1
      }
    ]
  })
});

if (!response.ok || !response.body) {
  throw new Error(`HTTP ${response.status}: ${await response.text()}`);
}

const decoder = new TextDecoder();
const reader = response.body.getReader();
let buffer = "";
let eventName = "message";
let dataLines = [];
let finalImage = "";

function findImage(value) {
  if (!value || typeof value !== "object") return "";
  if (Array.isArray(value)) {
    for (const item of value) {
      const found = findImage(item);
      if (found) return found;
    }
    return "";
  }
  for (const [key, item] of Object.entries(value)) {
    if (["result", "b64_json", "image", "image_b64", "partial_image", "partial_image_b64"].includes(key)) {
      if (typeof item === "string" && item.length > 1000) return item;
    }
    if (item && typeof item === "object") {
      const found = findImage(item);
      if (found) return found;
    }
  }
  return "";
}

function flushEvent() {
  if (!dataLines.length) {
    eventName = "message";
    return;
  }
  const raw = dataLines.join("\n");
  dataLines = [];
  const payload = JSON.parse(raw);
  const image = findImage(payload);
  if (image) {
    finalImage = image;
    console.log("image event:", payload.type || eventName);
  } else {
    console.log("event:", payload.type || eventName);
  }
  eventName = "message";
}

while (true) {
  const { value, done } = await reader.read();
  if (done) break;
  buffer += decoder.decode(value, { stream: true });
  const lines = buffer.split(/\r?\n/);
  buffer = lines.pop() || "";
  for (const line of lines) {
    if (line === "") {
      flushEvent();
    } else if (line.startsWith("event:")) {
      eventName = line.slice(6).trim();
    } else if (line.startsWith("data:")) {
      const data = line.slice(5).trimStart();
      if (data !== "[DONE]") dataLines.push(data);
    }
  }
}
flushEvent();

if (!finalImage) throw new Error("No image returned");
fs.writeFileSync("output.png", Buffer.from(finalImage, "base64"));

7. Python 示例

下面示例使用 requests 读取 SSE。生产环境建议使用成熟 SSE 客户端库，并加入超时、重试和日志。

import base64
import json
import requests

url = "https://junfeiai.com/v1/responses"
headers = {
    "Authorization": "Bearer YOUR_API_KEY",
    "Content-Type": "application/json",
    "Accept": "text/event-stream",
}
payload = {
    "model": "gpt-5.5",
    "input": [
        {
            "type": "message",
            "role": "user",
            "content": [
                {
                    "type": "input_text",
                    "text": "请生成一张简洁高级的科技产品宣传图，白色背景，主体是一台银色智能音箱，柔和棚拍光线，商业摄影风格",
                }
            ],
        }
    ],
    "stream": True,
    "store": False,
    "parallel_tool_calls": True,
    "tool_choice": {"type": "image_generation"},
    "reasoning": {"effort": "xhigh", "summary": "auto"},
    "tools": [
        {
            "type": "image_generation",
            "model": "gpt-image-2",
            "size": "1024x1024",
            "quality": "medium",
            "output_format": "png",
            "partial_images": 1,
        }
    ],
}


def find_image(value):
    if isinstance(value, dict):
        for key, item in value.items():
            if key in {"result", "b64_json", "image", "image_b64", "partial_image", "partial_image_b64"}:
                if isinstance(item, str) and len(item) > 1000:
                    return item
            found = find_image(item)
            if found:
                return found
    elif isinstance(value, list):
        for item in value:
            found = find_image(item)
            if found:
                return found
    return ""


final_image = ""
event_name = "message"
data_lines = []

with requests.post(url, headers=headers, json=payload, stream=True, timeout=180) as resp:
    resp.raise_for_status()
    for raw_line in resp.iter_lines(decode_unicode=True):
        line = raw_line or ""
        if line == "":
            if data_lines:
                event = json.loads("\n".join(data_lines))
                data_lines = []
                image = find_image(event)
                if image:
                    final_image = image
                    print("image event:", event.get("type", event_name))
                else:
                    print("event:", event.get("type", event_name))
            event_name = "message"
        elif line.startswith("event:"):
            event_name = line[6:].strip()
        elif line.startswith("data:"):
            data = line[5:].lstrip()
            if data != "[DONE]":
                data_lines.append(data)

if not final_image:
    raise RuntimeError("No image returned")

with open("output.png", "wb") as f:
    f.write(base64.b64decode(final_image))

8. 常见尺寸与质量建议

说明：

• quality=low 适合快速预览和批量试稿
• quality=medium 适合多数业务场景
• quality=high 适合最终交付图
• partial_images 只是请求中间预览图，不代表最终图一定更快完成
• 大尺寸、复杂中文排版、长提示词通常会显著增加耗时

9. Images API 兼容说明

旧客户端如果已经按 OpenAI Images API 接入，可继续使用：

POST https://junfeiai.com/v1/images/generations

但新接入不建议优先使用该路径。原因是 Images API 在部分账号或渠道下会被中转服务转换为 Responses + image_generation 工具调用，再把上游事件翻译回 Images API 风格。这样便于兼容旧客户端，但排查耗时阶段时不如 /v1/responses 直观。

Images API 兼容请求体示例：

{
  "model": "gpt-image-2",
  "prompt": "一只白色陶瓷杯放在木桌上，自然光，真实摄影风格",
  "size": "1024x1024",
  "quality": "medium",
  "output_format": "png",
  "stream": true,
  "partial_images": 1
}

兼容路径可能返回类似事件：

event: image_generation.partial_image
event: image_generation.completed

如果你需要观察更原始的阶段，请改用推荐的 /v1/responses。

10. 图片编辑接口

图片编辑仍使用 Images Edits 兼容接口：

POST https://junfeiai.com/v1/images/edits
Authorization: Bearer <你的 API Key>
Content-Type: multipart/form-data

常用参数：

curl 示例：

curl -X POST "https://junfeiai.com/v1/images/edits" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "model=gpt-image-2" \
  -F "image[]=@input.png" \
  -F "prompt=保留人物姿势和服装，把背景换成干净的白色摄影棚，商业大片质感" \
  -F "size=1024x1024" \
  -F "quality=medium"

11. 错误码排查

12. 客户对接提醒

• 建议服务端调用，不建议浏览器前端直接调用，避免 API Key 泄露
• 图片生成可能需要几十秒到两分钟以上，复杂提示词或高分辨率可能更久
• 客户系统应设置较长超时时间，建议不少于 180 秒
• 生产环境建议记录请求 ID、模型、尺寸、质量、调用时间、SSE 阶段和错误信息，方便排查
• 对耗时排查最有价值的阶段是 response.image_generation_call.generating 到 partial_image / completed
• 生成结果需符合平台内容政策，不建议生成侵权、违法、色情、暴力、仿冒证件等内容

13. 资料来源

• 君飞 AI 中转站：https://junfeiai.com
• OpenAI Responses API：https://platform.openai.com/docs/api-reference/responses
• OpenAI 图片生成指南：https://platform.openai.com/docs/guides/image-generation