I took about 2000 images that I felt were the best quality representations of 'cute' and slapped them into a trainer. This seems to work best at nudging outcomes to the cuter side at around half strength. Full strength works too, but often pushes outcomes to a more 2D art style even if you want something more 2.5D