Goodbye deepfakes? This new system can spot any AI-made video
AI is battling against AI: is this the start of the war?


Ever scrolled over a video on social media that appears too crazy to be true? Unless it’s Trump talking about the life of grass, it probably is a deepfake. These AI-generated videos appear incredibly lifelike and have the potential to completely upend someone’s reputation if used in the wrong way.
These days, spotting a deepfake isn’t always about checking faces: generative AI has already advanced far beyond basic lip-syncs or facial swaps. Entire scenes can now be created from scratch, all with convincing backgrounds, lighting, and motion. That’s exactly why researchers at UC Riverside, along with Google scientists, have fought AI vs AI and introduced a new tool designed to counter this growing threat.
Called UNITE (Universal Network for Identifying Tampered and synthEtic videos), this system is a departure from traditional detectors that merely focus on faces. Instead, UNITE casts a much wider net, analysing full video frames for subtle inconsistencies in motion, texture, colours, and surroundings.
Deepfake porn: the UK celebrity victims – including me.
— Channel 4 News (@Channel4News) March 21, 2024
Watch @cathynewman report. pic.twitter.com/YH51ReGhcF
“Deepfakes have evolved... people are now creating entirely fake videos”
“Deepfakes have evolved,” Rohit Kundu, a doctoral candidate in computer engineering at UC Riverside, told The Brighter Side. “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 model’s inner workings rest on a powerful transformer capable of interpreting spatial and temporal information. UNITE’s approach also works with “attention-diversity loss,” a part of the program that encourages the system to spread its focus across different parts of a frame rather than zeroing in on faces alone.
This broader vision delivers impressive results. Tested against benchmark datasets that include fully synthetic footage, altered backgrounds, and manipulated faces, UNITE consistently outperforms existing models that try to do the same job.
Related stories
“It’s scary how accessible these tools have become,” Kundu warned. “Anyone with moderate skills can bypass safety filters and generate realistic videos of public figures saying things they never said.”
Get your game on! Whether you’re into NFL touchdowns, NBA buzzer-beaters, world-class soccer goals, or MLB home runs, our app has it all. Dive into live coverage, expert insights, breaking news, exclusive videos, and more – plus, stay updated on the latest in current affairs and entertainment. Download now for all-access coverage, right at your fingertips – anytime, anywhere.


Complete your personal details to comment