Kirin: Animal Motion Generation from In-the-Wild Video

Paper | Project page | Code

Kirin reconstructs 3D animal motion from in-the-wild videos and uses it to learn to generate realistic quadruped motion from text and images. This repository hosts the pretrained Kirin motion generation models, trained on AiM3D, a large-scale dataset of aligned video, text and 3D motion for quadruped animals.

• kirin_text_image/: generates motion from text, an image of the animal, or both
• kirin_text/: generates motion from text
• animal_unified/: the evaluator used for the AiM3D metrics

See the code repository for usage.

Citation

@inproceedings{zhao2026kirin,
  title={Kirin: Animal Motion Generation from In-the-Wild Video},
  author={Zhao, Brian Nlong and Pan, Zhuoyang and Rehg, James M. and Wu, Jiajun and Wu, Shangzhe},
  booktitle={European Conference on Computer Vision (ECCV)},
  year={2026}
}