# Image Generation

BonzAI offers four image generation pipelines, each optimized for different use cases. All run locally via the Flask server on port 65000.

## Pipelines

| Pipeline     | Model         | Speed  | Quality           | VRAM  |
| ------------ | ------------- | ------ | ----------------- | ----- |
| **Turbo**    | FLUX Klein    | Fast   | Good              | 8 GB  |
| **Standard** | SDXL + LoRA   | Medium | High (uncensored) | 6 GB  |
| **Quality**  | FLUX Krea     | Slow   | Best              | 12 GB |
| **Advanced** | Z-Image Turbo | Fast   | High              | 8 GB  |

## API Endpoints

### Turbo

```
POST http://localhost:65000/image/turbo
```

### Standard

```
POST http://localhost:65000/image/standard
```

### Quality

```
POST http://localhost:65000/image/quality
```

### Advanced

```
POST http://localhost:65000/image/advanced
```

## Request Format

```json
{
  "prompt": "A cyberpunk cityscape at sunset",
  "negative_prompt": "blurry, low quality",
  "width": 1024,
  "height": 1024,
  "num_inference_steps": 30,
  "guidance_scale": 7.5
}
```

## Companion Portraits

When generating images for companions, BonzAI uses a multi-step pipeline:

1. **GLM Flash** (primary) or **LLaMA** (fallback) enhances the text prompt with visual details
2. The enhanced prompt is sent to the selected image pipeline
3. Generated images are uploaded to IPFS (Pinata) for on-chain metadata

## Minting as NFTs

Generated images can be minted as NFTs on Base (ERC-721) or LUKSO (LSP8). Requires **LVL3** (10,000 BONZAI held).

* Base mint fee: 0.0025 ETH
* LUKSO mint fee: 15 LYX