The model works fine on its own, but for some reason, when I train the models I create in Anima Preview 3—even though the dataset is originally from anime—the results can look like they were created with digital painting. I don't know why, but the good news is that it's easy to fix: you just have to add the word “anime” when this happens. I’ve tested the model with a strength of 0.5 to 1; it works well if it’s just this model. If you’re going to use other LORA models with it, I recommend starting to lower the intensity. With 0.5, I’ve been able to use between two and three LORAs without compromising the aesthetics or causing errors to appear.
