Trained on SDXL 1.0_vae_fix with an image size of 1024px

Managed again to reduce the file size even further without compromising (too much) on likeness/quality. No model merging/mixing or other fancy stuff. No resizing the File size afterwards..

Just pure training...

Works best with Dreamshaper XL therefore all example images were created with Dreamshaper XL 1.0 and are raw outputs of the used checkpoint.

No additional magic was used.