Secret lighting codes could make spotting deepfake videos easier

Cornell researchers have developed a novel light-based watermarking technique to combat the growing threat of deepfake videos, which have become increasingly convincing due to advances in generative AI. Unlike traditional digital watermarks that require cooperation from cameras or AI models, this method embeds nearly invisible codes directly into the lighting environment during video recording. By subtly varying the brightness of light sources—such as computer screens or lamps equipped with small computer chips—the system creates a hidden signature that is imperceptible to the human eye but can later be used to verify video authenticity. This “noise-coded” lighting approach leverages natural light fluctuations, making the embedded codes difficult to detect or remove without knowledge of the secret pattern. Each light source carries a unique code, enabling forensic analysts to identify manipulated or missing footage by comparing the original lighting pattern with recovered “code videos,” which reveal inconsistencies in altered sections. The technique supports multiple simultaneous codes within a scene, increasing the difficulty for adversaries who would need to replicate all codes consistently to