README.md

---
license: mit
tags:
  - face-swap
  - face-enhancement
  - deep-learning
  - computer-vision
  - morphstream
---

# MorphStream Models

Models for real-time face processing used by the MorphStream project.

## Structure

```
/
├── inswapper_128.onnx          # Standard face swap
├── inswapper_128_fp16.onnx     # FP16 optimized (default)
├── hyperswap_1a_256.onnx       # HyperSwap variant A
├── hyperswap_1b_256.onnx       # HyperSwap variant B
├── hyperswap_1c_256.onnx       # HyperSwap variant C
└── gfpgan/
    ├── GFPGANv1.4.pth                  # Main GFPGAN model
    └── weights/
        ├── detection_Resnet50_Final.pth # Face detection
        └── parsing_parsenet.pth         # Face parsing
```

## Face Swap Models

| Model | Description | Size |
|-------|-------------|------|
| inswapper_128.onnx | Standard quality face swap | ~250MB |
| inswapper_128_fp16.onnx | Optimized FP16 model (default) | ~125MB |
| hyperswap_1a_256.onnx | High quality 256px - variant A | - |
| hyperswap_1b_256.onnx | High quality 256px - variant B | - |
| hyperswap_1c_256.onnx | High quality 256px - variant C | - |

## GFPGAN Models

| Model | Description |
|-------|-------------|
| gfpgan/GFPGANv1.4.pth | Main GFPGAN model |
| gfpgan/weights/detection_Resnet50_Final.pth | Face detection (ResNet50) |
| gfpgan/weights/parsing_parsenet.pth | Face segmentation/parsing |

## Usage

### Python (huggingface_hub)

```python
from huggingface_hub import hf_hub_download

# Face swap model
model_path = hf_hub_download(
    repo_id="latark/MorphStream",
    filename="inswapper_128_fp16.onnx"
)

# GFPGAN model
gfpgan_path = hf_hub_download(
    repo_id="latark/MorphStream",
    filename="gfpgan/GFPGANv1.4.pth"
)
```

### GPU Worker

```bash
./scripts/download_models.sh /models
```

## License

MIT License