>>11336477
It's SDXL, just a more focused dataset. No Rule34 or ponies, way better Danbooru knowledge and some e621. Can actually do characters and artists at around a hundred booru images, and handles concepts better too.