Ideogram 4

Post-Training State-of-the-Art Text-to-Image Models through Composed Preference and Rubric Rewards

This article describes a text-to-image post-training recipe that combines human preference rewards, rubric rew…