Trained on 64 images from the anime. Used frames of the anime to train it.

Credit goes to Cybermeow for the program used to gather the dataset

https://github.com/cyber-meow/anime_screenshot_pipeline