V2-1I noticed that there was a need to generate multi-person images, so I tested some models. I found that the previous "overfitted" model generated multi-person images well.Put the label "anime screenshot" into the negation for a better style adaptation.Recommend using ADetailer.<hr />V2Put the label "anime screenshot" into the negation for a better style adaptation.The generation of the "club track jacket" is not stable.<s>While I have a relatively stable version of the clothing generation, I think there is some overfitting and the hands are a bit broken.</s><s>But I want to ask if you care about the clothing generation? Generally speaking, there are more NSFW content generated, and these are usually nude.</s><hr />Three roles in re-main,minato kiyomizu, jou joujima and takekazu ejiri.The example images were generated using AutismMix SDXL

Files with SHA256: 8d0a165e466a9b8586a585b1e79e6b090313b2e3ad427392624cc9f4e933219e