linkfox-multimodal-generate-image

Dieser Auszug wurde bei der Prüfung gespeichert. Die externe Quelle enthält die vollständige und aktuelle Version.

---
name: linkfox-multimodal-generate-image
description: AI驱动的图片生成与编辑工具，用于制作高质量产品图。当用户要求生成图片、制作图片、编辑照片、文生图、图生图、换背景、变换风格、替换图片中的物体、将产品合成到场景中、换模特、制作任何类型的AI生成视觉内容、AI drawing, image generation, text-to-image, image-to-image, background replacement, style transfer, product image creation, AI image editing时触发此技能。即使用户未明确说"AI图片"，只要其请求涉及生成、修改或变换图片，也应触发此技能。
---

# AI Image Generation

This skill guides you on how to generate and edit images using the AI image generation service, helping users create high-quality product images, modify existing images, and perform creative visual transformations.

## Core Concepts

The AI Image Generation tool produces new images based on a text prompt and optional reference images. It supports a wide range of use cases:

- **Text-to-image**: Generate a brand-new image purely from a text description.
- **Image-to-image**: Provide one or more reference images and a prompt to generate a new image that preserves elements from the references.
- **Image editing**: Modify specific elements, colors, backgrounds, or styles in an existing image.
- **Product compositing**: Place a product from one image into a scene from another image.
- **Model swapping**: Replace the model or mannequin in a product photo.

**Reference images are strongly recommended** when the user wants the output to closely resemble an existing product or scene. Up to 3 reference image URLs can be provided, separated by commas.

## Parameter Guide

| Parameter | Required | Description | Default |
|-----------|----------|-------------|---------|
| prompt | Yes | Text description of the desired image. Supports text-to-image, image-to-image, editing, model swapping, and more. Max 1000 characters. | -- |
| referenceImageUrl | No | URL(s) of reference image(s). Separate multiple URLs with commas. Up to 3 images supported. Max 1000 characters. | -- |
| aspectRatio | No | Aspect ratio of the output image. | 1:1 |

### Supported Aspect Ratios

| Value | Description |
|-------|-------------|
| 1:1 | Square (default) |
| 3:4 | Portrait |
| 4:3 | Landscape |
| 9:16 | Vertical fullscreen |
| 16:9 | Horizontal fullscreen |

### Prompt Writing Tips

1. **Be specific and descriptive**: Clearly describe the subject, scene, lighting, style, and mood you want.
2. **Reference images by number**: When using reference images, refer to them as "image 1", "image 2", etc., in the order they appear in `referenceImageUrl`.
3. **State the operation explicitly**: Use clear action verbs like "replace", "change", "put", "combine", "generate".
4. **Keep within 1000 characters**: Prompts have a maximum length of 1000 characters.

### Prompt Examples by Scenario

**Object replacement**:
```
Replace the vase on the table in image 1 with a potted plant
```

**Background color change**:
```
Change the background color of image 1 to pure white
```

**Product compositing**:
```
Place the product from image 2 onto the marble countertop in image 1
```

**Style transfer**:
```
Transform image 1 into the artistic style shown in image 2
```

**Text-to-image (no reference)**:
```
A professional product photo of a sleek black wireless headphone on a gradient blue background, studio lighting, 8K quality
```

**Model swapping**:
```
Replace the model in image 1 with a different model while keeping the same clothing and pose
```

## Local Image Upload

This tool requires **publicly accessible image URLs** for reference images. If the user provides a local image file path (e.g., `C:\Users\...\photo.png`, `/home/.../image.jpg`), you must upload it first to obtain a public URL.

Run the upload script:
```bash
python scripts/upload_image.py /path/to/local/image.png
```

The script will return a public URL (valid for 24 hours) that can be used as the reference image URL parameter.

## 调用方式

- **API 端点**：`POST /multimodal/generateImage`（完整参数/响应/错误码见 `references/api.md`）
- **Python 脚本**：`python scripts/multimodal_generate_image.py '<JSON 参数>' [--inline]`
- **成本约束**：本工具会消耗算力；同一会话同一参数组合默认只调用一次，脚本带 24