Cornell researchers unveil optical watermarks using noise-coded illumination that embeds authentication in light, exposing edits and deepfakes. Trusted video.
© A. Krivonosov
Researchers at Cornell University have introduced a new way to safeguard video from tampering—“optical watermarks.” Rather than embedding a digital tag in the file, the marker is woven into the very light used during shooting. The idea is elegant: authenticity shifts from the file to the illumination, where post-production tricks have less room to hide.
The method, known as noise-coded illumination, enables the creation of a unique watermark code that any camera can capture. These codes remain in the footage as temporal markers, allowing quick detection of edits or forgeries.
To put it into practice, one can use programmable lighting fixtures or ordinary bulbs equipped with a microchip that subtly modulates brightness within thresholds invisible to the human eye. That practicality matters: during verification, experts can readily flag inconsistencies—removed scenes, inserted objects, or fully synthetic clips.
At SIGGRAPH 2025, the researchers showed that up to three different light codes can operate simultaneously within a single scene, multiplying the effort required for would-be forgers. The authors said the technology could become a key tool against deepfakes and help restore trust in video as a source of truth. Given how the light-coded approach complicates tampering, that assessment sounds well grounded.