NVIDIA has launched a new artificial intelligence-powered tool designed to detect AI-generated or manipulated videos in real time.
The technology, called the NVIDIA Synthetic Video Detector (SVD), will be distributed through video streaming company Wowza’s Video Intelligence Framework (VIF), allowing organisations to analyse live video streams across on-premises, edge, hybrid and air-gapped computing environments using existing NVIDIA AI infrastructure.
The announcement comes amid rising concern over the misuse of synthetic media. According to cybersecurity company Surfshark, fraud linked to AI-generated content resulted in an estimated USD 410 million in losses during the first half of 2025. Separately, research cited by Wowza from Gartner found that 62 percent of organisations experienced a synthetic media attack in the previous 12 months.
Unlike systems that rely on identifying visible flaws in AI-generated content, the detector analyses statistical patterns left behind during the AI video generation process. Wowza said the model can operate on compressed and streamed video, producing a probability score for individual frames to indicate whether content may have been artificially generated.
The technology won the 2025 SAFE Challenge at the International Conference on Computer Vision (ICCV) and was later presented at the NeurIPS 2025 conference.
The detector is delivered as an NVIDIA NIM microservice, enabling organisations to deploy it on supported NVIDIA hardware without transferring sensitive video to external cloud services. Wowza said the system supports fully air-gapped deployments, allowing organisations to process video entirely within their own networks.
