how the fuck do you get prompt adherance in flux/chroma, especially with facial features? anyone know what words it recognizes? i try and prompt a short person and it just ignores it, even half the time if i add the clip5 tokens at the end. it really likes to stick to a couple of basic faces and no matter how much your try to change its mind it ignores you. most of the articles i read seem to think that have short hair is a facial feature. do i actually have to train a model?