Researchers Create Tool to Trace Fake Videos to the AI System That Made Them

This viral CCTV-style AI video in 2025 had many believing the bunnies are real.

Researchers have developed a new tool that can identify fake videos by tracing them back to the AI system that made them, using distinct visual “fingerprints” that these generative models leave behind.

AI technology can generate videos so realistic that distinguishing them from authentic footage is becoming increasingly difficult. But a computer science team led by researchers at the University of California, Riverside have developed a tool that moves beyond simply identifying whether a video is fake. It also determines which AI system created it.

Never knew how much I needed to see bunnies jumping on a trampoline pic.twitter.com/1zn7uaPSHD

— greg (@greg16676935420) July 28, 2025

The new framework identifies visual patterns within fake video frames that were unintentionally introduced by AI video generators, providing a forensic path to identifying the systems that created the videos. The new framework is called SAGA — short for Source Attribution of Generative AI Videos. It represents one of the first large-scale efforts to trace AI-generated videos back to their origins.

“The patterns are like fingerprints that the generative model leaves behind, and our goal here was to find out if the signatures are distinct amongst different generators. It turns out that, yes, there are distinct fingerprints there,” UCR doctoral student Rohit Kundu, who led the research under the guidance of Professor Amit Roy-Chowdhury in collaboration with researchers at YouTube and Google DeepMind, says in a statement.

“Knowing that a video is fake is often not enough,” the researchers write. “The critical need has shifted from whether it’s fake to what is its source?”

How It Identifies AI systems Used to Make Fakes

Unlike still images, videos contain motion and temporal information — how visual elements change from one frame to the next over time. Different AI video generators produce subtle patterns in those changes, creating distinctive artifacts. To detect those artifacts, the team analyzed both spatial details within individual frames and temporal relationships across entire video sequences. Rather than examining a single frame, the system studies how visual information evolves from moment to moment.

A key innovation is a technique called Temporal Attention Signatures, or T-Sigs. These signatures visualize the unique patterns associated with different video generators. By averaging patterns across many videos produced by the same AI system, SAGA generates a characteristic profile that helps distinguish one generator from another.

In a recent paper, the team tested SAGA on public datasets containing videos created by 19 different AI video generators. The systems included both text-to-video models, which generate videos from written prompts, and image-to-video models, which turn still images into moving footage.

The researchers found that SAGA could identify whether a video was real or AI-generated, determine whether it was created from text or an image, tell apart different versions of AI models, and even trace videos back to the team that developed the model.