|
|
--- |
|
|
license: mit |
|
|
base_model: |
|
|
- Laxhar/noobai-XL-Vpred-1.0 |
|
|
--- |
|
|
# γ·γγ γ©γ’γγ's PixelArt Style Model |
|
|
|
|
|
This is the pixel art style model I used on X (Twitter), account: [@shimamuraa11451](https://twitter.com/shimamuraa11451). This model is designed to generate pixel-style images. |
|
|
|
|
|
## π§ Recommended Settings |
|
|
|
|
|
- **CFG (Classifier-Free Guidance Scale)**: 3 |
|
|
- **Step**: 20 \~ 30 |
|
|
- **Scheduler**: SGM\_UNIFORM |
|
|
- **Sampler**: Euler series / DPM series |
|
|
|
|
|
## π Pixel Block Size & Rescaling Method |
|
|
|
|
|
This model generates approximately `8Γ8` pixel blocks. To ensure proper alignment, you must use the **"NEAREST downsample to 1/8, then upsample back to the original size"** method. |
|
|
|
|
|
- **ComfyUI**: This method can be implemented directly. |
|
|
- **Automatic1111 WebUI**: Related plugins are available for automatic processing. |
|
|
|
|
|
### Image Resolution Scaling |
|
|
|
|
|
Since the model works with `8Γ8` pixel blocks, you should set the height (H) and width (W) according to the final pixel art resolution you want. Refer to the table below: |
|
|
|
|
|
| Target Pixel Art Size | Input Resolution | |
|
|
| --------------------- | ---------------- | |
|
|
| 32Γ32 | 256Γ256 | |
|
|
| 128Γ128 | 1024Γ1024 | |
|
|
| 256Γ256 | 2048Γ2048 | |
|
|
|
|
|
## βοΈ Model Training Information |
|
|
|
|
|
This model was fine-tuned using **noob vpred-1.0**. |
|
|
|
|
|
Since it is vpred-based, additional steps may be required to ensure proper usage. |