I wonder if it would be possible to train them to specifically avoid generating images that look scary to people. Eg prefer blur or other artifacts over distort

I wonder if it would be possible to train them to specifically avoid generating images that look scary to people. Eg prefer blur or other artifacts over distorted bodies and faces.