For best results use 0.4 to 0.8 at 0.9 the bodies become much worse and you may get artifacts.Use "Kisaki" in prompt or she may become a blonde
Also do not forget that the model has been trained on quite explicit art, be careful using it
Description
I just decided to improve the previous version because a lot of time has passed, the loRA itself has become a little more flexible, and I think that because I retrained it on a larger number of artifacts and on a better model the results of loRA is sharper and more contrast