Trained on 100 screenshots of them from the anime. It is now better because of a new dataset, and It was trained on LoCon instead Of Lora

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

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