Trained on 58 images,mostly of "on-model" sources like the games,renders and upscaled anime screencaps
