VINA: A Unified Framework for AI-Generated Image and Video Detection

A new research paper proposes VINA (Video as Natural Augmentation), a unified framework for detecting AI-generated images and videos. The authors identify a critical failure mode: state-of-the-art AI image detectors often collapse when applied to video frames due to cross-modal gaps from video processing shifts and model-specific fingerprints. VINA jointly trains on image and video data, using video frames as natural augmentations, and introduces cross-modal supervised contrastive learning to bridge the gap. The paper is available on arXiv under ID 2605.21977.