MSAVBench: Towards Comprehensive and Reliable Evaluation of Multi-Shot Audio-Video Generation

Researchers introduced MSAVBench, the first comprehensive benchmark for multi-shot audio-video generation.

The framework evaluates models across video, audio, shot, and reference dimensions with adaptive methods.

Its evaluation is more robust and aligns closely with human judgments, including strong Spearman correlation.

Results reveal clear gaps in current state-of-the-art models, especially in control and audio-video synchronization.