UCLA Tool Spots Deepfake Videos Using Light

Researchers at the University of California, Los Angeles have developed a method that flags manipulated videos by analyzing the light they contain.

Updated on Nov. 3, 2026 in Artificial Intelligence

Photograph of a prism splitting white light into a rainbow band across a dark lab bench beside a strip of blank film frames and a lens assembly. — Researchers at the University of California, Los Angeles developed a tool that flags deepfake videos by analyzing the light they contain, reaching 99.86% accuracy on manipulated footage. AI Illustration. Upload story photo >

Live Poll

Do you trust that videos you see online are real?

A technique developed by UCLA researchers detects deepfake videos by analyzing light in a video, identifying manipulated videos with 99.86% accuracy and real videos with 95.72% accuracy. The system processes 15 videos at a time.

Why it matters

Humans identify deepfake videos only about 50% of the time, according to research published in Nature Communications, making automated detection a practical need. A method that reaches high accuracy on both manipulated and authentic footage could help viewers and institutions judge video authenticity.

The UCLA technique reports 99.86% accuracy identifying manipulated videos and 95.72% accuracy identifying real videos, versus about 50% accuracy for humans spotting deepfakes. The extraction does not specify how the system performs on larger volumes or on newer forms of manipulation.

The players

University of California, Los Angeles

UCLA is a public research university in Los Angeles, California, and the institution where the deepfake detection research was conducted.

The details

The method works by analyzing the light present in a video to determine whether the footage is a deepfake. The system processes 15 videos at a time, according to the research.

Article on the UCLA deepfake light-detection technique published November 3, 2026.

Roadmap

Detecting manipulated video has become a central concern as synthetic media grows more convincing and human viewers struggle to tell real footage from fake. The UCLA method adds a physics-based signal, light, to the set of tools researchers are developing to meet that challenge.

Viewers who encounter videos online may eventually have access to detection tools that flag manipulated footage. Until such tools are widely available, people are reminded that human judgment alone catches deepfakes only about half the time.

The takeaway

Automated detection may become an important complement to human judgment as manipulated video becomes harder to spot by eye. Readers should treat unverified video with caution, since people identify deepfakes only about half the time.