Basically feralizer for your krea + small-scale uncensor. Unbalanced dataset, overtrained on canines. Anus included. Recommended weight: 0.8. Works in the range 0.5-1. 1.1 causes style drift due to overfitting. No special tags, works out from the box. Pussy is enough. In case you want to enforce special pussy on anthro, you can add "anatomically correct SPECIE pussy". Due to that tagging strategy the steering towards special combination is kind of hard. Breaks anatomy in some cases. If so, lower the weight. Can do: Canines - very well. Dogs, foxes and some pokemons. Equines - well. Mares, ponies and even centaurs. Cervines - good most of the times. Deers, deer-centaurs and so on. Bovines - very poor dataset but suprisingly good. Cows. Implements furry anthro scenes, despite dataset cover on this topic is very low. Works bad with: Dragons, lizards: completely broken anatomy. Cats and felines - no images was in dataset so it draws default canine pussy. Realistic style - messes with anatomy often. you could try though. Any sex scene. This is presentation! Was trained on RTX3080TI on 512 resolution with a whopping 20000 steps on 149 images dataset. Had no experience with natural language models yet, so could screw up captioning easily. NO COMMERCIAL USAGE - the images for this dataset was not mine.

Files with SHA256: 808e5816637c6435e75f89d493b573bf4f014b82dd8984e77627a5bf9d2f3fd9