CoDance Model Card

CoDance is a research model for pose-driven character image animation. It is designed to improve robustness when the driving pose sequence is spatially mismatched with the reference image and when the reference contains multiple subjects.

Model Weights

The CoDance checkpoint is hosted on Hugging Face:

https://huggingface.co/Shuaishuai0219/codance/blob/main/codance.ckpt

The checkpoint contains the CoDance LoRA weights, Pose Shift Encoder, and Mask Encoder. It should be used together with Wan2.1-I2V-14B-720P and DWPose ONNX models.

Intended Use

This release is intended for academic research, reproducibility, and non-commercial experimentation in controllable video generation and character animation.

Out-of-Scope Use

Do not use this model for non-consensual identity manipulation, impersonation, harassment, fraud, deceptive media generation, or any application that violates the rights or consent of depicted people.

Data and Ethics

The project page notes that demo images are used for academic purposes. Users are responsible for ensuring that their own reference images, masks, and driving videos are used with proper rights and consent.

Citation

Please cite CoDance if you use this repository or checkpoint in your research.