Cornell researchers have developed noise-coded illumination (NCI), a way to place a hidden, time-varying watermark in the light illuminating a scene. A camera recording that scene captures the signal without needing to add a mark to the video file. Analysts with the secret code can then reconstruct low-fidelity, time-stamped views that may expose edits or synthetic additions. It is a research-stage forensic aid—not a universal deepfake detector, proof that footage is genuine, or a consumer product currently identified by Cornell.
What Cornell’s light-based watermark is
NCI encodes a secret signal in controlled illumination rather than in the pixels of a video file. The light changes subtly according to a code; a camera filming the illuminated scene records those changes as part of the captured light. The watermark is therefore associated with the scene’s lighting, not a particular camera’s watermarking feature.
After recording, an analyst who has the relevant code can recover low-fidelity, time-stamped “code videos” showing the scene under slightly different illumination. The coded views give the analyst another signal to compare with the suspect footage. They are not ordinary high-quality copies of the video, and the technique depends on access to both the controlled lighting and the code used to verify it. Cornell’s account explains the approach in its July 30, 2025 report.
How the signal could reveal edits or synthetic footage
The researchers say inconsistencies in the recovered code videos can point to changes in the recorded scene. A cut may create a gap in the time-stamped signal. Objects added or replaced in manipulated footage generally appear black in the recovered code videos, according to Cornell’s report. These are clues for forensic examination, not an automatic verdict that a clip is fake.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →#1 Best Overall
- Used Book in Good Condition
The idea is to make a convincing forgery harder by giving an analyst several related, coded views to check. Abe Davis, the Cornell assistant professor who first conceived the idea, described the challenge as requiring an attacker to fake each code video separately while making all of those fakes agree. That is a proposed source of friction for an adversary; Cornell also cautions that misinformation defense is an arms race and that attackers will keep finding ways to deceive.
How NCI differs from other watermarking and detection
| Approach | Where the signal comes from | What cooperation it requires |
|---|---|---|
| Direct video-file watermarking | Small changes embedded in video pixels. | The camera or AI system creating the video must embed the mark. |
| Noise-coded illumination (NCI) | A secret code encoded in controlled light illuminating the scene. | The scene must be lit by coded illumination, and verification requires access to the relevant code. |
| Post-capture deepfake detection | Analysis of the video after it has been recorded. | Does not create a capture-time lighting signal; the detector analyzes the video itself. |
NCI is intended to leave a signal in footage from different cameras recording the lit scene, unlike file watermarking that relies on a particular capture or generation system. It is also a proactive capture-environment measure, whereas post-capture detection examines footage after the fact. Cornell’s report and the ACM paper record do not establish a head-to-head performance comparison with other methods, so neither approach can be described as categorically more effective on this evidence.
What Cornell says has been demonstrated
Cornell’s university report says the team demonstrated up to three separate light codes in one scene, verified the approach in some outdoor settings, and tested it on people with different skin tones. These are research demonstrations reported by the university, not independent certification or evidence of broad operational deployment. The report gives no headline accuracy or reliability figure, so there is no supported percentage for how often NCI detects manipulated video.
The project was led by computer science graduate student Peter Michael, with Davis, Zekun Hao, and Serge Belongie as co-authors. Michael was scheduled to present “Noise-Coded Illumination for Forensic and Photometric Video” at SIGGRAPH 2025 on August 10 in Vancouver. The ACM record identifies the paper and describes NCI as a forensic and photometric video-analysis technique: ACM paper record.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Rank #3
Can you buy or set up NCI now?
Cornell’s report mentions possible applications in computer screens, photography lamps, and built-in lighting. It says programmable lights could be coded with software and that some older lamps could be adapted with a small chip. But the sources do not identify a compatible retail fixture, controller, or ready-to-use consumer installation. Buying a programmable LED, studio light, camera, or video-analysis program alone does not make it implement this specific research method.
For now, NCI is best understood as a research approach that could be built into controlled capture environments if the method becomes available for practical use. A creator or newsroom cannot assume that ordinary footage already contains this watermark, and a viewer cannot use the concept as a standalone authenticity check.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
