Nvidia launches deepfake video detector at SIGGRAPH 2026

Nvidia ↘ 0.06% introduced the Synthetic Video Detector NIM microservice at SIGGRAPH 2026 in Los Angeles, offering newsrooms and broadcasters a tool to identify AI-generated video with up to 92% accuracy on uncompressed footage. The announcement came during the annual computer graphics conference, running July 19–23 at the Los Angeles Convention Center.s2026.siggraph+1

How It Works

The microservice analyzes video frame by frame, producing a classifier score that indicates the probability a clip is synthetic rather than issuing a binary verdict. Editorial teams can use the score to prioritize clips for review, flag questionable footage, or escalate it for deeper analysis. On compressed video, accuracy drops to 87% at 15% compression and 82% at 50% compression, according to Nvidia's testing. The latest model revision reported an area under the curve (AUC) of 0.9614 on Nvidia's internal benchmark.wccftech+1

Processing speeds reach as low as 22 milliseconds for 1080p video on Nvidia RTX systems and roughly 30 milliseconds on Nvidia L40 GPUs. The detector focuses on low-level statistical and frequency-domain traces characteristic of generative video pipelines rather than analyzing semantic content.vp-land+2

Wowza Integration

Streaming infrastructure company Wowza announced its Video Intelligence Framework on July 20, embedding Nvidia's Synthetic Video Detector for real-time authenticity checks across live video feeds. Wowza's streaming software powers more than 35,000 video deployments across over 170 countries, bringing the detection capability to broadcasters, public safety agencies, and enterprise customers already running its infrastructure.globenewswire+3

The framework, built directly into Wowza Streaming Engine, continuously analyzes live streams, detects events of interest, and initiates automated responses through alerts and annotations. Nvidia's detector is available through the framework for customers running supported Nvidia AI infrastructure.securityinfowatch+1

Broader Context

The tool arrives as AI-generated video has grown more convincing and accessible, raising concerns about disinformation in news and social media. Nvidia positioned the detector as part of its AI for Media platform, targeting organizations that must make rapid decisions about footage authenticity. The microservice is designed to be robust against typical video compression, a common challenge for forensic tools that encounter re-encoded content across distribution platforms.build.nvidia+2