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

2026-05-19

AI Summary

Existing audiovisual generation evaluation frameworks suffer from limited data diversity, insufficient task coverage, and inadequate flexibility, hindering their applicability to complex, multi-shot narrative scenarios. This work proposes the first comprehensive benchmark and adaptive hybrid evaluation framework tailored for multi-shot audiovisual generation, encompassing four dimensions—video, audio, shot structure, and reference alignment—and supporting up to 15 shots, including non-photorealistic content. The framework introduces novel components: self-correcting shot segmentation, instantiated scoring rules, and a tool-driven evidence extraction mechanism. It achieves a Spearman rank correlation of 91.5% with human judgments and systematically evaluates 19 state-of-the-art models, revealing that modular or agent-based generation pipelines can effectively narrow the performance gap between open- and closed-source models while exposing critical limitations in director-level control and fine-grained audiovisual synchronization.

Abstract

Video generation is rapidly evolving from single-shot synthesis to complex multi-shot audio-video (MSAV) narratives to meet real-world demands. However, evaluating such frontier models remains a fundamental challenge. Existing benchmarks are limited in scope and data diversity, and rely on rigid evaluation pipelines, preventing systematic and reliable assessment of modern MSAV models. To bridge these gaps, we introduce MSAVBench, the first comprehensive benchmark and adaptive hybrid evaluation framework for multi-shot audio-video generation. Our benchmark spans four key dimensions, video, audio, shot, and reference, covering diverse task settings, varying shot counts of up to 15, and challenging non-realistic scenarios. Our evaluation framework improves robustness through an adaptive self-correction mechanism for shot segmentation, instance-wise rubrics for subjective metrics, and tool-grounded evidence extraction for complex judgments. Furthermore, MSAVBench achieves high alignment with human judgments, reaching a Spearman rank correlation of 91.5%. Our systematic evaluation of 19 state-of-the-art closed- and open-source models shows that current systems still struggle with director-level control and fine-grained audio-visual synchronization, while modular or agentic generation pipelines offer a promising path toward narrowing the gap between open- and closed-source models. We will release the benchmark data and evaluation code to facilitate future research.

Problem

• multi-shot audio-video generation
• evaluation benchmark
• audio-visual synchronization
• video generation
• model evaluation

Innovation

• multi-shot audio-video generation
• adaptive evaluation framework
• shot segmentation self-correction
• tool-grounded evidence extraction
• human-aligned benchmarking