Can Text-to-Image Models Draw from the Right Frame of Reference? Paper • 2608.03357 • Published Aug 4 • 23