GR-3 Technical Report

I was expecting this to build on top of GR-2, but this model doesn’t seem to predict images based on what I see?