Trained from 500+ artworks, mostly from the anime artworks and some video game adaptations that use the same art style. Anime screencaps were NOT included in the dataset, so you can expect better and sharper outlines.