Gemini 3.1 Flash Image Preview（流式） - Starrise AI

Gemini 3.1 Flash Image Preview（流式）通过 Starrise AI 以原生 Gemini API 提供服务，支持实时 SSE 流式返回图片生成结果。思考 chunk 先行推送，最终图片 chunk 紧随其后。

核心能力

SSE 流式输出 — 实时推送思考 chunk 与图片 chunk
思考模式 — 内部推理 chunk（thought: true）在图片之前流式输出
文生图 — 根据文本描述生成图片
图片编辑 — 在 inline_data 中传入参考图，配合文字指令进行编辑
宽高比控制 — 1:1、4:3、3:4、16:9、9:16
分辨率控制 — 1K（~1024px）、2K（~2048px）、4K（~4096px，按长边）

SSE 响应格式

流式端点返回换行分隔的 SSE 数据行，每行以 data: 开头，后跟 JSON 对象。共有三种 chunk 类型：

思考 chunk — 最先到达；parts[0].thought 为 true
图片 chunk — 包含 parts[0].inlineData，含 mimeType 和 base64 data（注意：流式响应中为驼峰命名）
最终用量 chunk — 包含顶层 usageMetadata，含 thoughtsTokenCount 及各模态 token 详情

data: {"candidates":[{"content":{"role":"model","parts":[{"text":"...","thought":true}]}}],"usageMetadata":{"trafficType":"ON_DEMAND"},"modelVersion":"gemini-3.1-flash-image-preview","createTime":"...","responseId":"..."}

data: {"candidates":[{"content":{"role":"model","parts":[{"inlineData":{"mimeType":"image/png","data":"<base64>"}}]}}],...}

data: {"usageMetadata":{"promptTokenCount":8,"candidatesTokenCount":1120,"totalTokenCount":1392,"trafficType":"ON_DEMAND","promptTokensDetails":[{"modality":"TEXT","tokenCount":8}],"candidatesTokensDetails":[{"modality":"IMAGE","tokenCount":1120}],"thoughtsTokenCount":264}}

流式响应中图片字段名为 inlineData（驼峰），而请求体中字段名为 inline_data（下划线）。这是原生 Gemini API 的行为。

文生图示例

curl "https://ai.alad.com/v1beta/models/gemini-3.1-flash-image-preview:streamGenerateContent?key=YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [
      {
        "role": "user",
        "parts": [
          { "text": "生成一张山间日落的图片" }
        ]
      }
    ],
    "generationConfig": {
      "responseModalities": ["TEXT", "IMAGE"],
      "imageConfig": {
        "aspectRatio": "16:9",
        "imageSize": "1K"
      }
    }
  }'

import requests, base64, json

url = "https://ai.alad.com/v1beta/models/gemini-3.1-flash-image-preview:streamGenerateContent"
params = {"key": "YOUR_API_KEY"}
data = {
    "contents": [
        {
            "role": "user",
            "parts": [{"text": "生成一张山间日落的图片"}]
        }
    ],
    "generationConfig": {
        "responseModalities": ["TEXT", "IMAGE"],
        "imageConfig": {"aspectRatio": "16:9", "imageSize": "1K"}
    }
}

response = requests.post(url, params=params, json=data, stream=True)

for line in response.iter_lines():
    if not line:
        continue
    decoded = line.decode("utf-8")
    if not decoded.startswith("data:"):
        continue
    chunk = json.loads(decoded[len("data:"):].strip())

    candidates = chunk.get("candidates", [])
    if not candidates:
        continue
    parts = candidates[0].get("content", {}).get("parts", [])
    for part in parts:
        # 跳过思考 chunk
        if part.get("thought"):
            continue
        # 保存图片 chunk
        if "inlineData" in part:
            img_bytes = base64.b64decode(part["inlineData"]["data"])
            with open("output.png", "wb") as f:
                f.write(img_bytes)
            print("图片已保存至 output.png")

图片编辑示例（传入参考图）

在同一个 parts 数组中同时传入 text 指令和 inline_data 参考图。

# 先将图片转为 base64：
# BASE64=$(base64 -i your_photo.jpg)
#
# 然后发送请求：
curl "https://ai.alad.com/v1beta/models/gemini-3.1-flash-image-preview:streamGenerateContent?key=YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [
      {
        "role": "user",
        "parts": [
          {
            "text": "这是我的一张照片，请在我旁边加一只羊驼"
          },
          {
            "inline_data": {
              "mime_type": "image/jpeg",
              "data": "<YOUR_BASE64_ENCODED_IMAGE>"
            }
          }
        ]
      }
    ],
    "generationConfig": {
      "responseModalities": ["TEXT", "IMAGE"],
      "imageConfig": {
        "aspectRatio": "1:1",
        "imageSize": "1K"
      }
    }
  }'

import requests, base64, json

# 读取并编码参考图片
with open("your_photo.jpg", "rb") as f:
    image_b64 = base64.b64encode(f.read()).decode("utf-8")

url = "https://ai.alad.com/v1beta/models/gemini-3.1-flash-image-preview:streamGenerateContent"
params = {"key": "YOUR_API_KEY"}
data = {
    "contents": [
        {
            "role": "user",
            "parts": [
                {
                    "text": "这是我的一张照片，请在我旁边加一只羊驼"
                },
                {
                    "inline_data": {
                        "mime_type": "image/jpeg",
                        "data": image_b64
                    }
                }
            ]
        }
    ],
    "generationConfig": {
        "responseModalities": ["TEXT", "IMAGE"],
        "imageConfig": {"aspectRatio": "1:1", "imageSize": "1K"}
    }
}

response = requests.post(url, params=params, json=data, stream=True)

for line in response.iter_lines():
    if not line:
        continue
    decoded = line.decode("utf-8")
    if not decoded.startswith("data:"):
        continue
    chunk = json.loads(decoded[len("data:"):].strip())

    candidates = chunk.get("candidates", [])
    if not candidates:
        continue
    parts = candidates[0].get("content", {}).get("parts", [])
    for part in parts:
        if part.get("thought"):
            continue
        if "inlineData" in part:
            img_bytes = base64.b64decode(part["inlineData"]["data"])
            with open("output.png", "wb") as f:
                f.write(img_bytes)
            print("图片已保存至 output.png")

参数说明

参数	类型	必填	说明
`key`	string	是	API 密钥（查询参数）
`alt`	string	否	设为 `sse` 可显式开启 SSE 模式（可选，流式为默认行为）
`contents[].parts[].text`	string	是	文字提示或指令
`contents[].parts[].inline_data.mime_type`	string	否	参考图类型：`image/jpeg`、`image/png`、`image/webp`
`contents[].parts[].inline_data.data`	string	否	Base64 编码的参考图数据
`generationConfig.responseModalities`	array	是	`["IMAGE"]` 或 `["TEXT", "IMAGE"]`
`generationConfig.imageConfig.aspectRatio`	string	否	`1:1` / `4:3` / `3:4` / `16:9` / `9:16`
`generationConfig.imageConfig.imageSize`	string	否	`1K` / `2K` / `4K`（默认 `1K`）

API 参考

查看 Gemini 3.1 Flash Image Preview（流式）的交互式 API Playground。