[doug_durham]

How much of that is new detail versus repeated patterns? A good approximation of all human visual reality fits in a 5GB diffusion model.