New detector spots deepfakes beyond face-swaps with 99% accuracy

Researchers at UC Riverside have developed a “universal” deepfake detector that achieved record-breaking accuracy rates of 95-99% across multiple types of AI-manipulated videos. Unlike existing tools that focus primarily on face-swap detection, this new system can identify completely synthetic videos, background manipulations, and even realistic video game footage that might be mistaken for real content.

What you should know: The detector represents a significant breakthrough in combating the growing threat of synthetic media across various applications.

• It monitors multiple background elements and facial features simultaneously, spotting subtle spatial and temporal inconsistencies that reveal AI manipulation.
• The system can detect inconsistent lighting on artificially inserted people, discrepancies in background details of AI-generated videos, and signs of manipulation in synthetic content without human faces.
• Several Google researchers participated in the development, though the company hasn’t indicated whether it will deploy this technology on platforms like YouTube.

Why this matters: The proliferation of cheap AI-powered deepfake tools has created an “out-of-control online spread” of synthetic videos used for non-consensual pornography, election misinformation, and financial scams targeting both consumers and executives.

• Many deepfakes depict women—including celebrities and schoolgirls—in non-consensual pornography.
• Scammers have begun using deepfakes during live video conferencing calls, creating new challenges for verification.

How it works: The universal detector takes a comprehensive approach that goes beyond traditional face-focused detection methods.

• “We assume that the entire video may be generated synthetically,” explains Rohit Kundu, the lead researcher at UC Riverside.
• The AI monitors spatial and temporal inconsistencies across all video elements rather than concentrating solely on facial features.
• It successfully flags realistic-looking scenes from video games like Grand Theft Auto V that aren’t necessarily AI-generated but could be mistaken for real footage.

The competitive advantage: This detector outperformed all other published methods for identifying face-manipulated deepfakes and achieved superior accuracy for completely synthetic videos.

• “Most existing methods handle AI-generated face videos—such as face-swaps, lip-syncing videos or face reenactments that animate a face from a single image,” notes Siwei Lyu at the University at Buffalo.
• “This method has a broader applicability range,” Lyu adds.

What’s next: The research team is already working on real-time detection capabilities for live video calls.

• “How do you know that the person on the other side is authentic, or is it a deepfake generated video, and can this be determined even as the video travels over a network?” asks Amit Roy-Chowdhury at UC Riverside.
• The researchers presented their findings at the 2025 IEEE Conference on Computer Vision and Pattern Recognition in Nashville on June 15.