I think if it gets all the way to a computer model rendering, where the raw data input is in no way shaped like an image “yet”, the distinction between traditional and new-generative-model approaches sounds more like a difference in degree.