If you didn't already know these, I give you a few tips I've found work when dealing with comics for data set
1: Blank out speech bubbles (just cover the text with white) and use the tags "speech bubble, blank speech bubble". It's way easier for the trainer to understand than unaltered (and gives way more options in data than avoiding stuff with them) and prevents any speech bubble/weird text from showing up in the final
2: Monochrome/greyscale images work fine as training data just tag them "monochrome, greyscale". I've never tried it with anything quite as simple in drawing as this though (maybe add chibi to the interior images?).
3: For such simple drawings on white backgrounds with a good scan, it's often quite easy to cut out stuff that isn't part of the character from your training data (provided the character isn't behind it). Don't consider it necessary for every image, but it's a nice trick to have for getting certain data as perfect as possible.