MSVBench: Towards Human-Level Evaluation of Multi-Shot Video Generation

TL;DR

MSVBench introduces a comprehensive multi-shot video generation benchmark with hierarchical scripts and a hybrid evaluation framework, revealing current models' limitations and enabling scalable, human-aligned improvements.

Contribution

It presents the first multi-shot video generation benchmark with hierarchical scripts and a hybrid evaluation method, advancing beyond single-shot paradigms.

Findings

Current models act as visual interpolators, not true world models.

MSVBench achieves 94.4% correlation with human judgments.

Fine-tuning on MSVBench's signals yields human-aligned performance.

Abstract

The evolution of video generation toward complex, multi-shot narratives has exposed a critical deficit in current evaluation methods. Existing benchmarks remain anchored to single-shot paradigms, lacking the comprehensive story assets and cross-shot metrics required to assess long-form coherence and appeal. To bridge this gap, we introduce MSVBench, the first comprehensive benchmark featuring hierarchical scripts and reference images tailored for Multi-Shot Video generation. We propose a hybrid evaluation framework that synergizes the high-level semantic reasoning of Large Multimodal Models (LMMs) with the fine-grained perceptual rigor of domain-specific expert models. Evaluating 20 video generation methods across diverse paradigms, we find that current models--despite strong visual fidelity--primarily behave as visual interpolators rather than true world models. We further validate the reliability of our benchmark by demonstrating a state-of-the-art Spearman's rank correlation of 94.4% with human judgments. Finally, MSVBench extends beyond evaluation by providing a scalable supervisory signal. Fine-tuning a lightweight model on its pipeline-refined reasoning traces yields human-aligned performance comparable to commercial models like Gemini-2.5-Flash.