Zhen Yang1*
·
Guibao Shen1*
·
Liang Hou2
·
Mushui Liu4
·
Luozhou Wang1
·
Xin Tao2
·
Pengfei Wan2
·
Di Zhang2
·
Yingcong Chen1,3✉
1HKUST(GZ) · 2KuaiShou Research · 3HKUST · 4Zhejiang University
*Both authors contributed equally.
✉Corresponding author.
- Create the environment and install the dependencies by running:
conda create -n RectifiedHR python=3.9
conda activate RectifiedHR
pip install diffusers==0.34.0
pip install transformers==4.46.2
pip install opencv-python>4.10.0
pip install torch==2.5.1
pip install accelerate==1.1.1
pip install einops==0.8.0
pip install ftfy==6.3.1
- Run with SDXL
python run_sdxl.py
- Run with WAN
python run_wan.py
- Change the hyperparameters
Edit the parameters in configs_sdxl.py or configs_wan.py to generate higher-quality images.
@article{yang2025rectifiedhr,
title={Rectifiedhr: Enable efficient high-resolution image generation via energy rectification},
author={Yang, Zhen and Shen, Guibao and Hou, Liang and Liu, Mushui and Wang, Luozhou and Tao, Xin and Wan, Pengfei and Zhang, Di and Chen, Ying-Cong},
journal={arXiv e-prints},
pages={arXiv--2503},
year={2025}
}