Compositional Text-to-Image Generation Made Easy