Can We Predict The Human Preference For Text-to-Image Content Prior To Generation And Is It Even Useful To Do So?
Researchers have developed a method to predict human preference scores for text-to-image generations before they are created. This approach aims to reduce computational waste by identifying promising generations early. The study found that predicting these scores is feasible with minimal hardware overhead and can be used to improve the quality of generated images. AI
IMPACT Enables more efficient training and generation of AI art by predicting quality before compute is spent.