News without the noise
Li and colleagues report that a video-scoring model trained to identify defects can predict human preferences and guide improvements to both open-weight and closed-source generators.
Southeast University · Opus AI Research · Institute of Automation, Chinese Academy of Sciences · University of Illinois Urbana-Champaign · University of California, Los Angeles
ArtiVidRM scores video by first listing its visible defects as a structured checklist, then using that defect list as a quality rubric.
Current methods for judging AI video quality give an opaque score or explanations not tied to the final judgment, making model improvement difficult.
ArtiVidRM forces the model to identify specific visible defects and organize them into a structured list, using this list as an explicit rubric for the quality score.
This grounds the score in concrete evidence, making it interpretable and more accurately reflecting what humans consider good quality, enabling targeted refinement.
The authors built a reward model, a system that evaluates generated outputs and supplies feedback for improving them.
The Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.