Tag
VA-Judger is the first reward model for joint video-audio generation that uses human preference feedback to evaluate holistic quality, including a dataset and benchmark, and demonstrates significant improvements in human preference rates when applied to models like LTX-2.