minimax image t2i

SKILL.md

---
name: minimax-image-t2i
description: MiniMax AI text-to-image generation tool. Use when the user wants to generate, create, or produce images (pictures/photos/illustrations) via AI. Triggers on phrases like "生成图片", "创建图片", "画一张图", "生成一张图片", "帮我画", "text-to-image", "文生图", "文生图", "text to image", or any request that asks for AI image generation. Requires only a text prompt describing the desired image content.
---

# MiniMax Text-to-Image (T2I)

Generate images from text prompts using MiniMax AI.

## Quick Usage

When user asks to generate an image, extract the image description from their prompt and run:

```bash
cd /root/.openclaw/workspace && python3 /root/.openclaw/skills/minimax-image/scripts/minimax-image.py "用户想要的图片描述" -n 数量 -a 宽高比 -o 输出文件.png
```

**Examples:**
- User: "帮我生成一张可爱的卡通螃蟹" → `python3 ... "一只可爱的卡通螃蟹" -n 1 -a 1:1 -o image.png"`
- User: "生成3张海边日落风景图" → `python3 ... "海边日落风景图" -n 3 -a 16:9 -o image.png"`

## Prompt Expansion (提示词扩写)

**自动行为**：当用户输入的提示词较短（<30字符）时，会自动调用 MiniMax 模型进行扩写，生成更详细专业的描述后再生成图片。

**扩写示例**：
- 输入："一只猫" → 输出："A fluffy adorable cat sitting on a windowsill, soft natural lighting, warm afternoon sun, photorealistic, 4K, high detail, front view"
- 输入："海边日落" → 输出："Breathtaking ocean sunset scene with vibrant orange and pink sky, calm waves reflecting golden light, silhouetted palm trees, cinematic composition, photorealistic, 8K, masterpiece"

**跳过扩写**：用户明确说"不扩写"、"不用扩写"时，添加 `--no-expand` 参数：
```bash
python3 ... --no-expand "一只猫"
```

## Parameters

| Parameter | Description | Default |
|-----------|-------------|---------|
| `-n` | Number of images (1-9) | 1 |
| `-a` | Aspect ratio | 1:1 |
| `-m` | Model (image-01 / image-01-live) | image-01 |
| `-o` | Output filename | image.png |
| `--expand` | Enable prompt expansion (default) | enabled |
| `--no-expand` | Disable prompt expansion | disabled |

**Aspect ratios:** 1:1, 16:9, 4:3, 3:2, 2:3, 3:4, 9:16, 21:9

## Workflow

1. Extract image description from user's prompt
2. **Check if expansion is needed** (short prompt + user didn't say "不扩写")
3. If expansion needed: call MiniMax model to expand → print expanded prompt
4. Generate image using expanded or original prompt
5. Send the generated image back to user via 飞书 with detailed message:

   **For group chat (群里 @ 你的情况)**:
   ```bash
   openclaw message send --channel feishu --target chat:oc_a386f54f877bb01f548c41f521e65d81 --message "🎨 图片生成完成！\n\n📝 原始提示词：{原始提示词}\n\n✨ 扩写后提示词：\n{完整的扩写后英文提示词}\n\n---\nGenerated by MiniMax AI" --media /root/.openclaw/workspace/minimax-images/输出文件.png
   ```

   **For direct message (私聊)**:
   ```bash
   openclaw message send --channel feishu --target user:ou_c767b11bf84f718816a6919f3d7a08cf --message "🎨 图片生成完成！\n\n📝 原始提示词：{原始提示词}\n\n✨ 扩写后提示词：\n{完整的扩写后英文提示词}\n\n---\nGenerated by MiniMax AI" --media /root/.openclaw/workspace/minimax-images/输出文件.png
   ```

   **重要**：消息中必须包含完整的扩写后提示词，不要省略或简化！这是用户想要看到的重要内容。

## Output Location

All images are saved to `/root/.openclaw/workspace/minimax-images-t2i/` directory.

- If user specifies a filename (e.g., `-o cat.png`), it will be saved as `/root/.openclaw/workspace/minimax-images-t2i/cat.png`
- If no filename specified, auto-generates `image_1.png`, `image_2.png`, etc.