The second picture messing up those highlights is such a a mood. It really struggles so much with such things. I tried making a style and it did pretty great until I tried to do the eyes. It kept adding too many highlights or not enough.

I theorize if we made a checkpoint without starting with an existing base, we might be able to get it to understand the concept, but older anime has issues with staying on model, so to fix it, we would have to edit every picture to make sure the eyes are always the exact same within every style.