NVIDIA Detects AI Video at 92% in 22ms
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- NVIDIA unveiled the Synthetic Video Detector NIM microservice, which analyzes video frame by frame to produce a classifier score indicating whether footage is synthetic, letting editorial teams prioritize clips for review, flag questionable content, or escalate for deeper analysis
- The Synthetic Video Detector delivers up to 92% accuracy on uncompressed video, falling to 87% at 15% compression and 82% at 50% compression, and does not replace existing verification practices but adds a verification layer for time-sensitive decisions
- On NVIDIA RTX systems the detector processes 1080p video in as little as 22 milliseconds, or roughly 30ms on L40 GPUs
- The latest model revision posted an AUC of 0.9614 and 94.53% accuracy on NVIDIA's internal test set, with thresholds configurable to support conservative review postures that minimize missed synthetic video
- NVIDIA is partnering with Wowza to embed the microservice in its Intelligent Video framework, putting the tool in front of more than 35,000 deployments across 170 countries
- The detector currently tops the AI-Generated Video Detection benchmark with a 98.7 mean AUC score, the highest among open-source generators tested
Why it matters: Newsrooms and media platforms handling user-submitted or breaking footage get a 22ms first-pass filter that runs at 92% accuracy on RTX hardware already common in editing suites. The Wowza integration puts that capability directly in front of 35,000 streaming deployments across 170 countries, turning what was custom AI forensics into a drop-in pipeline tool.
