Mayu Otani

Narrative intent dramatically alters video edits, yet current models struggle to match human quality, revealing a critical gap in AI video editing capabilities.

Current multimodal benchmarks are full of single-modality shortcuts, but this paper offers a way to prune them, yielding more reliable and efficient evaluations of true cross-modal reasoning.