Lossy Compression With Pretrained Diffusion Models

We present a lossy compression method that can leverage state-of-the-art diffusion models for entropy coding. Our method works zero-shot, requiring no additional training of the diffusion model or any ancillary networks. We apply the DiffC algorithm¹ to Stable Diffusion 1.5, 2.1, XL, and Flux-dev. We demonstrate that our method is competitive with other state-of-the-art generative compression methods at ultra-low bitrates.

Select an image and comparison settings

DiffC

MS-ILLM

0.0057 bpp0.0077 bpp0.0400 bpp0.0789 bpp

DiffEIC

0.0198 bpp0.0341 bpp0.0570 bpp0.0872 bpp0.1187 bpp

PerCo

0.0030 bpp0.0320 bpp0.1260 bpp

The following video shows lossy reconstructions at each of Flux’s 1,000 timesteps:

Citation

    @inproceedings{
  vonderfecht2025lossy,
  title={Lossy Compression with Pretrained Diffusion Models},
  author={Jeremy Vonderfecht and Feng Liu},
  booktitle={The Thirteenth International Conference on Learning Representations},
  year={2025},
  url={https://openreview.net/forum?id=raUnLe0Z04}
}