Exactly. The whole image generation bit, at this stage, is noise. But you could easily describe what someone is looking at. Even that is very interesting. And perhaps this will scale up to get high quality, accurate, outputs.