Yolov8 medium model trained on about 8k Manga, Webtoon, Manhua and Western Comic style Images for speech bubble detection. Training Imgsize = 1024. Training Images were resized, not cropped. It can handle the extreme Image ratios often found in Korean webtoons.