Trained on 1000 screenshots of Raika Ken-Style Anime.

The dataset is derived from screenshots of animations, where characters are cropped using the YOLOv8 model. Due to the prevalent blur effects on the top and bottom borders in animations, the model tends to generate images with these blurred borders. This issue seems to be mitigated when using Highres.Fix at resolutions above 1024 pixels.

Still working on creating a better training set.