Anybody knows if there is a technical report for this, or for other models that generate images in a similar way? I'd really like to understand the architecture behind 4o-like image gen.