README.md

---
license: apache-2.0
base_model: THUDM/CogVideoX-5b-I2V
tags:
  - video-generation
  - cogvideox
  - motion-control
  - driving
  - physics
---

# Motion Forcing: A Decoupled Framework for Robust Video Generation in Motion Dynamics

[![arXiv](https://img.shields.io/badge/arXiv-2603.10408-b91c1c?style=flat&logo=arxiv)](https://arxiv.org/abs/2603.10408)
[![Project Page](https://img.shields.io/badge/Project%20Page-Motion%20Forcing-2d6a4f?style=flat&logo=github)](https://tianshuo-xu.github.io/Motion-Forcing/)
[![GitHub](https://img.shields.io/badge/GitHub-Repo-24292e?style=flat&logo=github)](https://github.com/Tianshuo-Xu/Motion-Forcing)


- **Base model**: [THUDM/CogVideoX-5b-I2V](https://huggingface.co/THUDM/CogVideoX-5b-I2V)
- **Task**: Image-to-video generation with motion control

## Usage

```bash
git clone https://github.com/Tianshuo-Xu/Motion-Forcing
cd Motion-Forcing
pip install -r requirements.txt
python gradio_demo.py
```

The model weights are downloaded automatically from this repo via `diffusers`. See the [GitHub repository](https://github.com/Tianshuo-Xu/Motion-Forcing) for full setup instructions and CLI options.

## Demo & Results

Visit the [project page](https://tianshuo-xu.github.io/Motion-Forcing/) for video comparisons across driving scenes, physics simulations, and embodied AI tasks.

## Citation

```bibtex
@misc{xu2026motion,
      title={Motion Forcing: A Decoupled Framework for Robust Video Generation in Motion Dynamics}, 
      author={Tianshuo Xu and Zhifei Chen and Leyi Wu and Hao Lu and Ying-cong Chen},
      year={2026},
      eprint={2603.10408},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2603.10408}, 
}
```