>>3019781Nobody has the hardware for actual nsfw finetunes, the latest base is still noob v-pred 1.0. What you think of checkpoints are just merges of illustrious and/or noob together with style loras and inferior models.
The only meaningful attempt since then was bluvoll's rectified flow, trained on top of latest noob. But I have no idea if he got a hold of a newer dataset or if he trained enough for the knowledge to stick, and there's the downside of it making everything grey again because he "fixed" v-pred.
Chroma was supposed to be this, but lodestone lied about including anime in the dataset so it has very limited danbooru knowledge. Same deal with Pony v7, it seems to know even less than v6.
