FrankenMotion

Part-level human motion generation and composition diffusion model.

• 📄 Paper: https://huggingface.co/papers/2601.10909
• 💻 Code: https://github.com/Coral79/FrankenMotion-Code
• 🧟 Dataset: Coral79/frankenstein-dataset
• 📊 Eval encoders: Coral79/frankenmotion-eval-model

This repo hosts the trained FrankenMotion diffusion checkpoint used for inference, training resume, and evaluation in the CVPR 2026 paper FrankenMotion: Part-level Human Motion Generation and Composition.

Contents

config.json                  # training config (consumed by src.config.read_config)
frankenmotion.ckpt           # diffusion model weights (~170 MB)
motion_stats/{mean,std}.pt   # normaliser stats for the 613-dim smplrifke features
text_stats/{mean,std}.pt     # normaliser stats for the per-part CLIP text embeddings

Quick start

from huggingface_hub import snapshot_download
snapshot_download(repo_id="Coral79/frankenmotion", local_dir="pretrained/frankenmotion")

Then follow the main repo README for inference, training, and evaluation commands.

Citation

@inproceedings{li2026frankenmotion,
  title={{FrankenMotion}: Part-level Human Motion Generation and Composition},
  author={Li, Chuqiao and Xie, Xianghui and Cao, Yong and Geiger, Andreas and Pons-Moll, Gerard},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2026}
}