Nvidia Just Built an AI to Catch Other AIs Lying in Video
Nvidia is throwing fire at its own house with the new Synthetic Video Detector NIM, designed to spot deepfakes before they ruin reality.
Engineers at Nvidia released a microservice called Synthetic Video Detector NIM that scans video frames to rate whether content is real or computer-generated. The system allows media editorial teams to automatically flag suspicious footage and send it down the pipeline for deeper inspection.
Instead of completely replacing human fact-checkers, the tool acts as a rapid-response triage system. The model achieves 92% accuracy on pristine uncompressed video, though that drops to 87% with moderate 15% compression and down to 82% when video gets squished by 50%. Processing a 1080p video clip takes just 22 milliseconds on consumer RTX hardware and around 30 milliseconds on enterprise L40 GPUs.
In benchmark tests like AI GVD, the model landed in top positions, while internal tests yielded an impressive AUC score of 0.9614. Nvidia is already partnering with streaming platform Wowza to integrate the tool into broader media workflows across 170 countries.
The tech ecosystem has officially entered a perpetual motion machine of chaos, selling the hardware to generate realistic deepfakes and then turning around to sell the software needed to catch them.
Source: Wccftech
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.