Jun Hao Liew* · Hanshu Yan* · Jianfeng Zhang · Zhongcong Xu · Jiashi Feng
ByteDance Inc.

Paper PDF Project Page

MagicEdit explicitly disentangles the learning of appearance and motion to achieve high-fidelity and temporally coherent video editing. It supports various editing applications, including video stylization, local editing, video-MagicMix and video outpainting.

For avatar-centric video generation and animation, please also check our latest work MagicAvatar!

If you find our work useful, please consider citing:

@inproceedings{liew2023magicedit,
 author = {Liew, Jun Hao and Yan, Hanshu and Zhang, Jianfeng and Xu, Zhongcong and Feng, Jiashi},
 title = {MagicEdit: High-Fidelity and Temporally Coherent Video Editing},
 booktitle = {arXiv},
 year = {2023}
}