---
license: apache-2.0
---
These are two proof-of-concept `compact` (`SRVGGNetCompact`) models meant to be used with anime images.

- `RevDiff1_x1_soft.pth` - x1 model, cleans up linework/random spots fairly well, recommended.
- `RevDiff1_x2_sharp.pth` - x2 model, upscales clean lineart reasonably well, but has issues with artifacting.

They were trained against images that have been processed by stable diffusion at a low denoise. The goal was to try and revert the artifacts introduced by the VAE/UNET.

Release post soon.