Trained on 36 images of them from the internet. It is biased because I manually picked the images from the anime.