We propose a first truly non-local video editor. Unlike other video editing methods, which either avoid making large spatial changes or do so by corrupting video content, our method can manipulate video content in both space and time. MotionV2V can control where objects appear, how fast they move and in which direction, when they appear, and even the camera position. All of these edits are done with a simple interface: a point-based control mechanism, allowing these edits to be made with simple point trajectory modifications (no rotoscoping or masks!). Please read our paper and see our videos to learn more!

If you use this work in your research, please cite:

@misc{burgert2025motionv2veditingmotionvideo,
      title={MotionV2V: Editing Motion in a Video},
      author={Ryan Burgert and Charles Herrmann and Forrester Cole and Michael S Ryoo and Neal Wadhwa and Andrey Voynov and Nataniel Ruiz},
      year={2025},
      eprint={2511.20640},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2511.20640},
}