README.md

---
license: mit
pipeline_tag: image-to-video
library_name: diffusers
---

# CharacterShot: Controllable and Consistent 4D Character Animation

[CharacterShot](https://arxiv.org/abs/2508.07409) is a controllable and consistent 4D character animation framework that enables any individual designer to create dynamic 3D characters from a single reference character image and a 2D pose sequence.

- **Paper:** [CharacterShot: Controllable and Consistent 4D Character Animation](https://arxiv.org/abs/2508.07409)
- **Code:** [GitHub Repository](https://github.com/Jeoyal/CharacterShot)
- **Authors:** [Junyao Gao](https://huggingface.co/Gaojunyao), [Jiaxing Li](https://huggingface.co/LiJiaxing), Wenran Liu, [Yanhong Zeng](https://huggingface.co/zengyh1900), Fei Shen, Kai Chen, Yanan Sun, Cairong Zhao

## Introduction

CharacterShot begins by pretraining a powerful 2D character animation model based on a DiT-based image-to-video model (CogVideoX). It lifts the animation model from 2D to 3D through introducing dual-attention module together with camera prior to generate multi-view videos with spatial-temporal and spatial-view consistency. Finally, it employs a novel neighbor-constrained 4D gaussian splatting optimization on these multi-view videos, resulting in continuous and stable 4D character representations.

## Citation

```bibtex
@article{gao2025charactershot,
  title={CharacterShot: Controllable and Consistent 4D Character Animation},
  author={Gao, Junyao and Li, Jiaxing and Liu, Wenran and Zeng, Yanhong and Shen, Fei and Chen, Kai and Sun, Yanan and Zhao, Cairong},
  journal={arXiv preprint arXiv:2508.07409},
  year={2025},
}
```

## Acknowledgements
The code is built upon [CogVideo](https://github.com/THUDM/CogVideo).