Nano Banana 2 (APIYI)

Generate images from text descriptions using APIYI's Gemini 3.1 Flash Image Preview API.

Quick Start

当用户说“生成一张xxx图片 / 帮我生成xxx”时，必须先走中文提示词选择流程（先给选择、再生成；此阶段绝不调用生成 API）：

先调用脚本候选输出（不生成）：
- python {baseDir}/scripts/generate_image.py --suggest-zh "<用户需求>"
- 把返回的 A/B/C 三套“增强中文提示词 + 反向词”发给用户让其选择
用户选择 A/B/C
- 若用户选择 A（北欧极简方向），再调用： python {baseDir}/scripts/generate_image.py --suggest-zh-a "<用户需求>" 并把 A1/A2/A3 发给用户二次选择
用户最终确认：版本（A/B/C 或 A1/A2/A3）+ 比例（16:9/9:16/1:1）+ 清晰度（1K/2K）
仅在用户确认后，才调用生成脚本：
- python {baseDir}/scripts/generate_image.py --prompt "<最终提示词>" --aspect-ratio "<比例>" --size "<1K/2K>"

说明：APIYI 对英文更友好；如用户追求更稳定/更像参考图，可在第 4 步把“最终中文提示词”再改写为英文 prompt 后生成。

中文提示词候选（不生成，只给选择）

A/B/C 三选一：

python {baseDir}/scripts/generate_image.py --suggest-zh "简约独居冷色治愈"

若用户选 A（北欧极简方向），再给 A1/A2/A3：

python {baseDir}/scripts/generate_image.py --suggest-zh-a "简约独居冷色治愈"

Basic Generation

python {baseDir}/scripts/generate_image.py --prompt "你的最终提示词" --filename "output.png" --aspect-ratio "16:9" --size "2K"

Advanced: Prompt Restructuring Workflow

For better results, restructure user description into optimized prompts:

Step 1: Understand User Input

Extract from user's description:

Subject: What object/person/scene
Style: Illustration, realistic, watercolor, oil painting, minimal, tech, Japanese, Chinese retro, etc.
Mood/Atmosphere: Warm, cold, nostalgic, futuristic, cozy, etc.
Technical: HD, detailed, close-up, bokeh, etc.

Step 2: Restructure Prompt

Combine elements into English prompt:

Template:

[Subject], [Style], [Mood/Atmosphere], [Lighting], [Material], [Composition], [Technical params]

Example:

User: "一只可爱的柴犬坐在樱花树下，水彩画风格，高清细节"
Restructured: "A cute柴犬 sitting under a cherry blossom tree, watercolor painting style, soft pastel colors, delicate details, HD quality, illustration aesthetic"

Step 3: Generate

python3 {baseDir}/scripts/generate_image.py \
  --prompt "A cute dog sitting under cherry blossoms, watercolor style, soft pastel colors, spring atmosphere, HD quality" \
  --filename "2026-03-12-16-00-00-sakura-dog.png" \
  --aspect-ratio "16:9" \
  --size "2K"

Parameters

Parameter	Options	Default
`--aspect-ratio`	`1:1`, `16:9`, `9:16`	`16:9`
`--size`	`1K`, `2K`	`2K`

API Key Configuration

Set in ~/.openclaw/openclaw.json:

{
  "skills": {
    "nano-banana2-apiyi": {
      "apiKey": "sk-your-api-key"
    }
  }
}

Or use --api-key flag directly.

Examples

Example 1: Tech Poster

User: "生成一张科技感的数据可视化海报"

Restructured: "Data visualization dashboard with futuristic UI, neon blue and purple color scheme, holographic interface, clean lines, tech poster style, high definition"

Example 2: Vintage Photography

User: "生成一张复古胶片风格的咖啡店照片"

Restructured: "Cozy coffee shop interior, vintage film photography style, warm golden lighting, analog film grain, nostalgic atmosphere, 35mm film look, detailed"

Example 3: Product Shot

User: "帮我生成一个简约的香水瓶商业摄影"

Restructured: "Elegant perfume bottle on clean background, commercial product photography, minimalist style, soft studio lighting, luxury branding, high-end packaging, professional advertising photo"

Notes

Always use English prompts for better APIYI performance
Use timestamps in filenames: yyyy-mm-dd-hh-mm-ss-name.png
The script prints MEDIA: line for auto-attach on supported chat providers
Do not read the image back; report the saved path only