A minimalist text-to-image model — pick the B/16 or L/16 variant below. Both are preloaded. Images are generated at 512×512.