https://github.com/williechai/speedup-plugin-for-stable-diffusions
https://github.com/williechai/speedup-plugin-for-stable-diffusions
Last synced: 7 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/williechai/speedup-plugin-for-stable-diffusions
- Owner: williechai
- License: mit
- Created: 2023-12-13T08:28:00.000Z (almost 2 years ago)
- Default Branch: main
- Last Pushed: 2023-12-22T06:02:44.000Z (almost 2 years ago)
- Last Synced: 2023-12-23T04:23:09.406Z (almost 2 years ago)
- Language: Python
- Homepage:
- Size: 8.53 MB
- Stars: 49
- Watchers: 4
- Forks: 2
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
- awesome-diffusion-categorized - [Code
README
# speedup-plugin-for-stable-diffusions
This repo is the official implementation of SpeedUpNet(SUN) in PyTorch.
Paper: [SpeedUpNet: A Plug-and-Play Hyper-Network for Accelerating Text-to-Image Diffusion Models](https://arxiv.org/pdf/2312.08887.pdf)
Project Page: [SpeedUpNet](https://williechai.github.io/speedup-plugin-for-stable-diffusions.github.io/)
## 10x speed up on stable diffusions
Introducing SUN as a plug-in, a pre-trained SD can generate high-quality images in only 4 steps. We can test on MacBook Pro(M1 Pro):
DPM-Solver++ 20 steps, 16 seconds (baseline)
+SUN, 4 steps, 2 seconds
See more on our [webpage](https://williechai.github.io/speedup-plugin-for-stable-diffusions.github.io/)
## Realtime Controllable Generation
SUN is compatible with controllable tools.
Real-time rendering can be achieved on high-end consumer-grade graphics cards.
## Usage
```
cd demo
# prepare models
python controlnet_lora.py
```
## Download SUN adapter
https://huggingface.co/Williechai/SpeedUpNet/tree/main
## Update
**`2023.12.15`**: Readme.
## Citation
If you find this work is helpful in your research, please cite our work:
```
@misc{chai2023speedupnet,
title={SpeedUpNet: A Plug-and-Play Hyper-Network for Accelerating Text-to-Image Diffusion Models},
author={Weilong Chai and DanDan Zheng and Jiajiong Cao and Zhiquan Chen and Changbao Wang and Chenguang Ma},
year={2023},
eprint={2312.08887},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
## Contact
If you have any questions, feel free to open an issue or directly contact me via: `weilong.cwl@antgroup.com`.