i1 shows that strong text-to-image models do not have to rely on opaque recipes, inaccessible data, or highly specialized new components. A simple, fully open recipe built from public datasets and…