InstructAV2AV: Instruction-Guided Audio-Video Joint Editing

Researchers introduced InstructAV2AV, the first end-to-end system for instruction-guided joint audio-video editing.

They also built InsAVE-80K, a large-scale dataset of source-target audio-video pairs for training and evaluation.

The method improves alignment between audio and visuals while better preserving content from the original media.

It outperforms prior approaches on multiple benchmarks, addressing a key gap in realistic and controllable media editing.