New Tool Identifies Sources of AI-Generated Videos
A new tool has been developed to identify the sources of fake videos, a growing concern in today’s digital landscape. The innovation comes as artificial intelligence systems are rapidly improving and synthetic videos are increasingly being used in entertainment, advertising, education, social media, and politics.
The AI-generated video problem is not just about determining whether a video is real or fake; it’s also about understanding where the content originated from. A team of researchers at UC Riverside has developed a tool called SAGA (Source Attribution of Generative AI Videos) that can identify which AI system created a particular video.
SAGA works by analyzing visual patterns within individual frames and temporal relationships across entire video sequences. 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 that can be detected.
The researchers used a technique called Temporal Attention Signatures (T-Sigs) to visualize these 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.
SAGA was tested using public datasets containing videos generated by 19 different AI video generators, including text-to-video systems and image-to-video systems. The tool can determine whether a video is real or synthetic, identify whether it was created from text or images, and even identify the development team behind a model.
The ability to identify the source of AI-generated content could help digital forensic investigators track misinformation campaigns, assist regulators in enforcing transparency requirements, and aid technology companies in understanding how synthetic media spreads online. This is particularly important when it comes to free AI video generators that are increasingly being used by businesses for advertising and other purposes.
According to Rohit Kundu, the lead researcher on the project, ‘The patterns are like fingerprints that the generative model leaves behind.’ The researchers’ goal was to find out if these signatures were distinct amongst different generators. Their findings show that yes, there are distinct fingerprints left by each AI system.
The development of SAGA is part of an ongoing effort to make generative AI safer as increasingly sophisticated AI video tools become widely available. As Kundu noted, ‘It’s a cat-and-mouse game for sure.’ The researchers’ work aims to stay ahead of the curve and provide a comprehensive solution to identifying the sources of fake videos.
The research is described in the paper ‘SAGA: Source Attribution of Generative AI Videos,’ which includes contributions from Kundu, Professor Amit Roy-Chowdhury, and other collaborators. Their current work builds on previous research by the same team that developed an artificial intelligence model to detect video tampering.