Smaller Deepfakes May Be the Bigger Threat

A new dataset called MultiFakeVerse reveals how AI can subtly alter images to change narratives without swapping faces. These edits, which modify gestures, props, and backgrounds, fool both AI detectors and humans, with people only correctly identifying fakes 62% of the time. The study exposes a blind spot in deepfake detection, as current systems struggle to identify these subtle, narrative-driven manipulations.