Brand new dataset, I basically had to remake everything from scratch and start anew. I've been trying to make this style as consistent and manageable as possible, and it looks like this, for a first version is pretty spot on so I'll be releasing it as an early access till I test it a bit more and if I encounter any strange behaviour or get more feedback from the community, I'll be able to work on it a bit more, the only reason why this is behind Early Access is to be able to gather buzz in order to retrain it as soon as possible!If you want to support me, you can either donate, like or follow for more updates, either one helps. Thank you so much, every single buzz or support that I receive is going straight into the LoRAs.RecommendationsStrength 0.9, lower for more flexibility. Same goes for your cfg, anything below 4.0 is perfect or recommended. Also try not to use too many LoRAs together as it can diminish the quality of your image fairly quickly, or a high cfg will also do that.I'd also recommend avoid using tags such as; curvy, plump, "huge", mature female, wide hips, huge ass, puffy nipples or thick, since the dataset is already trained to generate thick girls as default, and if you want to go a bit more on the safe side of things, you can also put those in the negative and whatnot, or use weights to emphasize certain things like; ((small_breasts)), f.e.

Purumomo ぷる桃 Style | Illustrious

Still testing this one out, I've tested the last epoch and it's a bit overfitted, this one seems to be a bit more manageable and less artifacts, atleast for now.

Files with SHA256: 7d2ec5630b5530a02f7985a0c2fb274e4c529da055bb3ec225976e1641c2dd39