ShadowDancer

Teaching Video World Models Any Action from a Video and Its Shadow

ShadowDancer: Teaching Video World Models Any Action by Learning Unified Dynamics Representations from a Video and Its Shadow

Jin Cao, Zian Meng, Kaipeng Zhang† († corresponding author)

ShadowDancer gives interactive video world models an any-action, frame-level control.

The code and model weights are being cleaned up and are undergoing internal review before release.🚧

Citation

@misc{cao2026shadow,
  title={ShadowDancer: Teaching Video World Models Any Action by Learning Unified Dynamics Representations from a Video and Its Shadow},
  author={Jin Cao and Zian Meng and Kaipeng Zhang},
  year={2026},
  eprint={2607.28362},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2607.28362},
}