Rich Human Feedback For Text To Image