README.md

---
license: mit
language:
- en
base_model:
- microsoft/TRELLIS-image-large
pipeline_tag: image-to-3d
tags:
- animatable
- rigging
- 3D
- Tripo
- VAST
---
# AniGen_Weights

Pretrained checkpoints for [AniGen](https://github.com/VAST-AI-Research/AniGen), a unified framework for generating animatable 3D assets from a single image.

<p align="center">
  <a href="https://arxiv.org/pdf/2604.08746"><img src="https://img.shields.io/badge/arXiv-Paper-red?logo=arxiv&logoColor=white" alt="arXiv"></a>
  <a href="https://yihua7.github.io/AniGen_web/"><img src="https://img.shields.io/badge/Project_Page-Website-green?logo=googlechrome&logoColor=white" alt="Project Page"></a>
  <a href="https://www.tripo3d.ai"><img src="https://img.shields.io/badge/Tripo-AI_3D_Workspace-orange" alt="Tripo"></a>
  <a href="https://huggingface.co/spaces/VAST-AI/AniGen"><img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Live_Demo-blue" alt="Hugging Face Demo"></a>
  <a href="https://github.com/VAST-AI-Research/AniGen"><img src="https://img.shields.io/badge/GitHub-Repository-black?logo=github&logoColor=white" alt="GitHub"></a>
</p>

This repository stores the contents of the `ckpts/` directory used by the AniGen codebase, including:

- AniGen stage checkpoints
- DINOv2 vision encoder weights
- DSINE normal estimation weights
- VGG backbone weights

## What Is Included

The repository is organized exactly like the `ckpts/` folder expected by the main AniGen repo:

```text
ckpts/
├── anigen/
│   ├── ss_dae/
│   ├── slat_dae/
│   ├── ss_flow_duet/
│   ├── ss_flow_epic/
│   ├── ss_flow_solo/
│   ├── slat_flow_auto/
│   ├── slat_flow_control/
│   └── slat_flow_gsn_auto/
├── dinov2/
├── dsine/
└── vgg/
```

Approximate total size: about 23 GB.

## Recommended Checkpoints

For most users, we recommend:

- `ss_flow_duet` for sparse structure generation
- `slat_flow_auto` for structured latent generation

This combination is also the default setup used by the AniGen inference example.

## Checkpoint Overview

### Core AniGen checkpoints

| Folder | Purpose |
| --- | --- |
| `ckpts/anigen/ss_dae` | Sparse Structure autoencoder |
| `ckpts/anigen/slat_dae` | Structured Latent autoencoder |
| `ckpts/anigen/ss_flow_duet` | SS flow model with stronger skeleton detail |
| `ckpts/anigen/ss_flow_epic` | SS flow model balancing geometry and skeleton quality |
| `ckpts/anigen/ss_flow_solo` | SS flow model with stronger geometry generalization |
| `ckpts/anigen/slat_flow_auto` | SLAT flow model with automatic joint-count prediction |
| `ckpts/anigen/slat_flow_control` | SLAT flow model with controllable joint density |
| `ckpts/anigen/slat_flow_gsn_auto` | Additional SLAT variant included in the release |

### Dependency checkpoints

| Folder | Purpose |
| --- | --- |
| `ckpts/dinov2` | DINOv2 encoder files and pretrained ViT-L/14 weights |
| `ckpts/dsine` | DSINE normal estimation weights |
| `ckpts/vgg` | VGG weights used by the pipeline |

## How To Use

Clone the main AniGen repository first:

```bash
git clone --recurse-submodules https://github.com/VAST-AI-Research/AniGen.git
cd AniGen
```

Then download this weights repository so that the folder structure is preserved under the project root.

### Option 1: Download with `huggingface_hub`

```bash
python -c "from huggingface_hub import snapshot_download; snapshot_download(repo_id='VAST-AI/AniGen_Weights', repo_type='model', local_dir='.', local_dir_use_symlinks=False)"
```

After download, you should have paths like:

```text
ckpts/anigen/ss_flow_duet/ckpts/denoiser.pt
ckpts/anigen/slat_flow_auto/ckpts/denoiser.pt
ckpts/dsine/dsine.pt
ckpts/vgg/vgg16-397923af.pth
```

### Option 2: Download from the web UI

You can also download this repository from the Hugging Face file browser and place the entire `ckpts/` folder at the root of the AniGen project.

## Run AniGen With These Weights

Once the `ckpts/` folder is in place, you can run:

```bash
python example.py --image_path assets/cond_images/trex.png
```

Or launch the Gradio demo:

```bash
python app.py
```

## Notes

- Keep the directory names unchanged. The AniGen code expects the exact `ckpts/...` layout shown above.
- The code repository may automatically fetch missing files in some setups, but this weights repository is the recommended way to pre-download and manage checkpoints explicitly.
- `slat_flow_control` supports joint density control, while `slat_flow_auto` is the best default for general use.

## Related Links

- Best AI 3D studio -- Tripo: https://www.tripo3d.ai
- Main code repository: https://github.com/VAST-AI-Research/AniGen
- Project page: https://yihua7.github.io/AniGen-web/
- Demo: https://huggingface.co/spaces/VAST-AI/AniGen
- Paper: https://arxiv.org/pdf/2604.08746

## Citation

```bibtex
@article{huang2026anigen,
  title     = {AniGen: Unified $S^3$ Fields for Animatable 3D Asset Generation},
  author    = {Huang, Yi-Hua and Zhou, Zi-Xin and He, Yuting and Chang, Chirui
               and Pu, Cheng-Feng and Yang, Ziyi and Guo, Yuan-Chen
               and Cao, Yan-Pei and Qi, Xiaojuan},
  journal   = {ACM SIGGRAPH},
  year      = {2026}
}
```