>>8692687
it's not about being happy with it, i've trained versions of the lora which look good. i keep retraining it because it's my largest dataset with the most alternate outfits and i ported/made some of them myself. it's good for figuring out how shading can effect style burn in or testing out how large details on outfits need to be for the vae to work with them on top of figuring out hyperparameter settings between the two loras