As AI-generated videos become increasingly convincing and accessible, UC Riverside researchers have partnered with Google to develop a powerful new weapon against digital deception.
Their system, called the Universal Network for Identifying Tampered and synthEtic videos (UNITE), addresses a critical vulnerability in current deepfake detection technology. While existing tools primarily focus on facial anomalies, UNITE examines entire video frames, including backgrounds, motion patterns, and subtle spatial-temporal inconsistencies that reveal manipulation.
"Deepfakes have evolved," explains Rohit Kundu, a UC Riverside doctoral candidate who led the research. "They're not just about face swaps anymore. People are now creating entirely fake videos—from faces to backgrounds—using powerful generative models. Our system is built to catch all of that."
The collaboration, which included Professor Amit Roy-Chowdhury and Google researchers Hao Xiong, Vishal Mohanty, and Athula Balachandra, was presented at the 2025 Conference on Computer Vision and Pattern Recognition in Nashville. Their innovation comes as text-to-video and image-to-video generation platforms have made sophisticated video forgeries accessible to virtually anyone.
UNITE employs a transformer-based deep learning model built on a foundation called SigLIP, which extracts features not bound to specific people or objects. A novel training method dubbed "attention-diversity loss" forces the system to monitor multiple visual regions in each frame, preventing over-reliance on faces.
Though still in development, UNITE could soon become essential for social media platforms, newsrooms, and fact-checkers working to prevent manipulated videos from going viral. As deepfakes increasingly threaten public trust, democratic processes, and information integrity, universal detection tools like UNITE represent a crucial line of defense against digital misinformation.