banana-image

Banana Image Generation Skill

密钥

从 research_config.jsonc 的 banana_image 字段读取所有 ${...} 变量。

---

生成图像

API 概览

• 格式：OpenAI DALL-E 兼容
• 模型：Nano-banana-3.1-Flash（Generations，推荐）
• 请求方式：POST ${image_gen_url}/v1/images/generations

请求参数

Header

参数	类型	必需	说明
Authorization	string	否	默认值：Bearer ${image_generation_api_key}

Body（application/json）

参数	类型	必需	说明
model	string	是	模型名称，如 gemini-3.1-flash-image-preview
prompt	string	是	图像描述提示词
aspect_ratio	enum	否	可选值：4:3 3:4 16:9 9:16 2:3 3:2 1:1 4:5 5:4 21:9 1:4 4:1 8:1 1:8
response_format	string	否	url 或 b64_json
image	array[string]	否	参考图数组（url 或 b64_json）
image_size	enum	否	仅 nano-banana-2 支持，可选值：512 1K 2K 4K

请求示例

import http.client
import json

conn = http.client.HTTPSConnection("${image_gen_url}'s host")  # read from research_config.jsonc -> banana_image.image_gen_url
payload = json.dumps({
   "prompt": "cat",  # include image url here if you want to edit one image
   "model": "gemini-3.1-flash-image-preview"
})
headers = {
   'Authorization': 'Bearer ${image_generation_api_key}',  # read from research_config.jsonc -> banana_image.image_generation_api_key
   'Content-Type': 'application/json'
}
conn.request("POST", "/v1/images/generations", payload, headers)
res = conn.getresponse()
data = res.read()
print(data.decode("utf-8"))

返回响应

• 200：成功，返回 application/json 对象

---

Prompting 指南

Interactive prompt crafting for Nano Banana Pro image generation. This skill guides users through a structured process to create effective prompts by clarifying intent and applying proven techniques.

Step 1: Gather Reference Materials

Before asking questions, check if the user has provided:

• Reference images - Photos to use for character consistency, style, or composition
• Existing prompts - Previous attempts to improve upon
• Visual references - Screenshots or examples of desired output