NVIDIA Synthetic Video Detector Hits 92% Accuracy on Uncompressed Footage

NVIDIA has released the Synthetic Video Detector, a software tool designed to identify AI-generated and fake news videos. This release is significant because synthetic media is increasingly difficult to distinguish from authentic footage, and the tool provides a technical method for verification. The detector analyzes video frame-by-frame to produce a classifier score for synthetic content. It aims to help platforms and users verify the authenticity of video content.

New microservice analyzes 1080p video in 22ms on RTX systems

The product is a NIM microservice, which means it is designed for easy integration into existing software workflows. NVIDIA is partnering with Wowza to embed this microservice in its Intelligence Video framework. This partnership enables video streaming and management platforms to integrate detection capabilities directly into their workflows. The tool is designed for developers and enterprise users who need to manage video integrity at scale.

The detector achieves 92% accuracy on uncompressed video, 87% accuracy at 15% compression, and 82% accuracy at 50% compression. Processing latency is 22ms for 1080p footage on NVIDIA RTX systems and 30ms on L40 GPUs. These metrics demonstrate that the tool can analyze video in real-time on standard RTX hardware. The internal benchmark accuracy is 0.9453 with an AUC of 0.9614 on NVIDIA's test set.

The Synthetic Video Detector is already topping the leaderboards on the AI GVD bench. This performance indicates that it is currently one of the most effective tools for this specific task. The microservice will soon be available to over 35,000 deployments across 170 countries. This broad availability suggests that NVIDIA is targeting global adoption for enterprise and platform integrations.

While NVIDIA previously focused on hardware acceleration for AI tasks, this release highlights a shift toward specialized software solutions. The Synthetic Video Detector adds a specific utility to the NIM ecosystem for content verification. Users and platforms can now integrate this detection capability into their video workflows. The tool is available for deployment in the near future across a global network of partners.