You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
🥺 This is a minimal re-implementation of MotionCanvas based on Wan-I2V-1.3B with limited resources.
🤗 MotionCanvas can generate short video clips from a static image with specified camera motion and object (global and local) motion. Please check our project page and paper for more information.
📝 Changelog
Add gradio demo and inference code for camera and object motion control.
[2025.07.26]: Release the minimal re-implementation code.
[2025.02.26]: Launch the project page and update the arXiv preprint.
Download the pre-trained Wan2.1-Fun-1.3B-InP model weights and our pre-trained weights.
This re-implementation of MotionCanvas supports generating videos of up to 49 frames with a resolution of 832x480. The inference time can be reduced by using fewer denoising steps.
⚙️ Setup
Install Environment via Anaconda (Recommended)
Please follow the instruction of installation in DiffSynth-Studio.
😉 Citation
Please consider citing our paper if our code is useful:
@article{xing2025motioncanvas,
title={Motioncanvas: Cinematic shot design with controllable image-to-video generation},
author={Xing, Jinbo and Mai, Long and Ham, Cusuh and Huang, Jiahui and Mahapatra, Aniruddha and Fu, Chi-Wing and Wong, Tien-Tsin and Liu, Feng},
journal={arXiv preprint arXiv:2502.04299},
year={2025}
}
🙏 Acknowledgements
We would like to thank Yujie for providing partial implementation, DiffSynth-Studio for offering an awesome codebase and Wan-AI for GPU support.
📢 Disclaimer
This project strives to impact the domain of AI-driven video generation positively. Users are granted the freedom to create videos using this tool, but they are expected to comply with local laws and utilize it responsibly. The developers do not assume any responsibility for potential misuse by users.
About
[SIGGRAPH 2025] MotionCanvas: Cinematic Shot Design with Controllable Image-to-Video Generation