Meloetta LoRA trained on NovelAI with Kohya training scripts using the DAdaptAdam optimizer at 512x512 resolution. Trained on lewd Meloetta pics without the black dress to make it easier to see the tid (and put other clothes on her). Works fairly consistently, but I had issues with YiffyMix trying to add a nose to her. Adding (noseless) to the positive and (nose) to the negative helped with that. Tends to go cross-eyed if you don't prompt for eye position, likely due to the high number of ahegao images in the dataset. Should have wide compatibility with model mixes based on NovelAI, which is most of them. Trained on an all female dataset.