Trained on 35 images, some mirrored, mostly nsfw, sometimes generates just one ear, hi res fix sometimes adds some details related to a dog's face instead of a skull.