NVIDIA launches AI video detector with 92% accuracy

NVIDIA has introduced Synthetic Video Detector, an AI-powered microservice designed to help media organisations identify potentially AI-generated video.

The system analyses footage frame by frame and produces a probability score indicating whether it contains synthetic content.

Editorial teams can use the result to prioritise clips for review, quarantine suspicious footage or escalate it for more detailed verification.

NVIDIA said the tool is intended to supplement established fact-checking and forensic practices rather than determine authenticity on its own.

In company testing, the detector achieved up to 92% accuracy on uncompressed video. Accuracy declined to 87% at 15% compression and 82% at 50% compression.

The system can process 1080p video in as little as 22 milliseconds on NVIDIA RTX systems and around 30 milliseconds on NVIDIA L40 GPUs.

Its architecture uses an ensemble of DINOv2 and DINOv3 vision transformer backbones and focuses on artefacts introduced by generative video systems.

Synthetic Video Detector forms part of the NVIDIA AI for Media platform and can be deployed in on-premises, edge, hybrid and approved air-gapped environments.

NVIDIA said the flexible deployment options could help broadcasters, government agencies and other organisations analyse sensitive footage while retaining control over their data.

Why does it matter?

Synthetic video is becoming harder to identify during fast-moving news events, increasing pressure on media organisations to verify footage before publication. NVIDIA’s detector could provide an additional screening signal and help editorial teams focus limited verification resources. Still, its declining performance on compressed material shows why automated detection cannot replace source checks, forensic analysis and human judgement.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our chatbot!