ChenyangSi/FreeU
A training-free technique that enhances diffusion model sample quality by strategically reweighting U-Net skip connections and feature maps.
Not currently ranked — collecting fresh signals.
star history
FreeU is a research method from S-Lab at Nanyang Technological University published at CVPR 2024 Oral. It improves diffusion model outputs without additional training or fine-tuning by modulating skip connections and backbone features in the U-Net. The technique can be integrated into existing diffusion models including Stable Diffusion to boost generation quality at no computational cost, making it applicable to both text-to-image and image-to-image generation pipelines.
Frequently asked
- What is ChenyangSi/FreeU?
- A training-free technique that enhances diffusion model sample quality by strategically reweighting U-Net skip connections and feature maps.
- Is FreeU open source?
- Yes — ChenyangSi/FreeU is open source, released under the MIT license.
- How popular is FreeU?
- ChenyangSi/FreeU has 1.9k stars on GitHub.
- Where can I find FreeU?
- ChenyangSi/FreeU is on GitHub at https://github.com/ChenyangSi/FreeU.