---
license: other
license_name: mixed-upstream-licenses
license_link: https://huggingface.co/lefreud/GimbalDiffusion/blob/main/LICENSE.md
pipeline_tag: text-to-video
base_model:
- THUDM/CogVideoX-2b
- Wan-AI/Wan2.1-T2V-1.3B
- Wan-AI/Wan2.2-TI2V-5B
datasets:
- YOUSIKI/PanoWan
tags:
- video-generation
- camera-control
- camera-trajectory
- gravity
- gimbaldiffusion
---

# GimbalDiffusion

Paper-release checkpoints for **GimbalDiffusion: Gravity-Aware Camera Control for Video Generation** (ECCV 2026). GimbalDiffusion is implemented across CogVideoX-2B, WAN 2.1, and WAN 2.2 using both Plücker-ray and UCPE camera encodings.

- [Project page](https://lvsn.github.io/GimbalDiffusion/)
- [Paper](https://arxiv.org/abs/2512.09112)
- [Code](https://github.com/lvsn/GimbalDiffusion)

## Checkpoints

| Path | Backbone | Variant | Size (GB) |
| --- | --- | --- | ---: |
| `gimbal_ac3d/controlnet.pt` | CogVideoX-2b / AC3D | Main | 0.022 |
| `gimbal_ucpe/dit.safetensors` | Wan 2.1 / UCPE | Main | 2.909 |
| `gimbal_ucpe_no_np/dit.safetensors` | Wan 2.1 / UCPE | Without null-pitch conditioning | 2.909 |
| `gimbal_ucpe_no_absolute/dit.safetensors` | Wan 2.1 / UCPE | Relative poses | 2.909 |
| `gimbal_wan_2_1/dit.safetensors` | Wan 2.1 | Main | 2.942 |
| `gimbal_wan_2_2/dit.safetensors` | Wan 2.2 | Main | 10.007 |
| `gimbal_wan_2_2_no_np/dit.safetensors` | Wan 2.2 | Without null-pitch conditioning | 10.007 |

## Download

```bash
hf download lefreud/GimbalDiffusion --local-dir checkpoint/gimbaldiffusion
hf download THUDM/CogVideoX-2b --local-dir checkpoint/backbones/THUDM/CogVideoX-2b
hf download Wan-AI/Wan2.1-T2V-1.3B --local-dir checkpoint/backbones/Wan-AI/Wan2.1-T2V-1.3B
hf download Wan-AI/Wan2.2-TI2V-5B --local-dir checkpoint/backbones/Wan-AI/Wan2.2-TI2V-5B
```

For setup, inference, and training instructions, see the [GimbalDiffusion README](https://github.com/lvsn/GimbalDiffusion#readme).

## License

This repository contains artifacts governed by different upstream terms. In particular, the AC3D checkpoint is restricted to non-commercial research use. Read [LICENSE.md](LICENSE.md) and [LICENSE-AC3D.txt](LICENSE-AC3D.txt) before use. Base-model weights are not redistributed here and retain their own licenses.

## Citation

```bibtex
@inproceedings{fortierchouinard2026gimbaldiffusion,
  title = {GimbalDiffusion: Gravity-Aware Camera Control for Video Generation},
  author = {Fortier-Chouinard, Fr{\'e}d{\'e}ric and Hold-Geoffroy, Yannick and Deschaintre, Valentin and Gadelha, Matheus and Lalonde, Jean-Fran{\c{c}}ois},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year = {2026}
}
```