README.md

---
license: apache-2.0
pipeline_tag: image-to-video
---

# ReImagine: Rethinking Controllable High-Quality Human Video Generation via Image-First Synthesis

[**Project Page**](https://keruzheng.github.io/ReImagine-Project/) | [**Paper (arXiv)**](https://arxiv.org/abs/2604.19720) | [**Code**](https://github.com/Taited/ReImagine) | [**Demo**](https://taited-reimagine.hf.space/)

**ReImagine** is a framework for controllable high-quality human video generation. It revisits the problem from an image-first perspective, where high-quality human appearance is learned via image generation and used as a prior for video synthesis. This approach decouples appearance modeling from temporal consistency.

The system utilizes a pose- and viewpoint-controllable pipeline that combines a pretrained image backbone with SMPL-X-based motion guidance, followed by a training-free temporal refinement stage based on a pretrained video diffusion model.

## Getting Started

### Installation

```bash
conda create -n reimagine python=3.10
conda activate reimagine
pip install torch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1 --index-url https://download.pytorch.org/whl/cu124
pip install -e .
```

### Pretrained Weights

ReImagine utilizes base models and specific LoRA weights. You can download the weights using the Hugging Face CLI:

```bash
# Download base FLUX.1 model
hf download black-forest-labs/FLUX.1-Kontext-dev \
  --local-dir ./models/FLUX.1-Kontext-dev \
  --exclude "flux1-kontext-dev.safetensors" \
  --exclude "vae/**"

# Download ControlNet
hf download jasperai/Flux.1-dev-Controlnet-Surface-Normals \
  --local-dir ./models/Flux.1-dev-Controlnet-Surface-Normals

# Download ReImagine LoRA Weights
hf download taited/ReImagine-Pretrained --local-dir ./models/ReImagine-Pretrained
```

## Inference

To perform image-first synthesis, use the provided inference script:

```bash
python inference_img.py
```
This script requires a wide reference image (front and back views) and a normal map generated from SMPL-X. For video synthesis, the temporal-refinement stage is used to ensure consistency across frames.

## Citation

If you find this project useful, please consider citing the paper:

```bibtex
@article{sun2025rethinking,
  title={ReImagine: Rethinking Controllable High-Quality Human Video Generation via Image-First Synthesis},
  author={Sun, Zhengwentai and Zheng, Keru and Li, Chenghong and Liao, Hongjie and Yang, Xihe and Li, Heyuan and Zhi, Yihao and Ning, Shuliang and Cui, Shuguang and Han, Xiaoguang},
  journal={arXiv preprint arXiv:2604.19720},
  year={2026},
  url={https://arxiv.org/abs/2604.19720v1}
}
```