understanding human preference in text-to-image synthesis - View it on GitHub