The answer could be in between. Who said delusion models are limited to 2d pixel generations?