There’s a new paper: “Autonomous language-image generation loops converge to generic visual motifs”  — diffusion models have just 12 standard templates. [Cell; Cell, with supplements, PDF; press re…