blurgyy/CoMPaSS

Enhancing spatial understanding in text-to-Image diffusion models