UniRig Models (Safetensors)

This repository contains the UniRig model weights converted to safetensors format for safer and faster loading.

Attribution

These models are converted from the original VAST-AI/UniRig repository by VAST-AI-Research.

Original Work:

• Paper: UniRig: A Unified Framework for 3D Character Rigging
• GitHub: VAST-AI-Research/UniRig
• Authors: VAST AI Research Team

All credit for the model architecture and training goes to the original authors.

Models

Why Safetensors?

• Safer: No arbitrary code execution risk (no pickle)
• Faster: Memory-mapped loading for quick startup
• Portable: Standard format supported across frameworks

Usage

Loading with safetensors

from safetensors.torch import load_file

# Load skeleton model
skeleton_weights = load_file("skeleton.safetensors")

# Load skin model
skin_weights = load_file("skin.safetensors")

Using with UniRig

These weights are compatible with the original UniRig codebase. Replace the .ckpt loading code:

# Original (ckpt)
# checkpoint = torch.load("model.ckpt")
# state_dict = checkpoint['state_dict']

# With safetensors
from safetensors.torch import load_file
state_dict = load_file("model.safetensors")

Conversion Details

Converted using the following process:

import torch
from safetensors.torch import save_file

checkpoint = torch.load("model.ckpt", map_location='cpu', weights_only=False)
state_dict = checkpoint['state_dict']
save_file(state_dict, "model.safetensors")

The conversion preserves all tensor values, shapes, and dtypes exactly.

License

This work is licensed under the MIT License, following the original UniRig license.

Citation

If you use these models, please cite the original work:

@article{unirig2024,
  title={UniRig: A Unified Framework for 3D Character Rigging},
  author={VAST AI Research},
  journal={arXiv preprint arXiv:2405.02986},
  year={2024}
}