Trained on the same dataset as the SD1.5 versions of the model. Generates quite high quality images but seems to trend toward a 2.5d pixar style for some reason. I havent tested in pure anime checkpoints so maybe its more due to the checkpoints I am using but you may need to use keywords to get it to look more anime. I also recommend using lower strength for this version if you dont like the 2.5D style.