Nvidia Launches Synthetic Video Detector to Identify AI-Generated Videos with Up to 92% Accuracy
Nvidia has launched the Synthetic Video Detector (SVD) service, integrated as a NIM microservice, which can identify AI-generated videos with up to 92% accuracy. The tool analyzes videos frame-by-frame using Meta's DINOv2 and DINOv3 vision transformer models. As deepfakes and AI-generated media become increasingly realistic and widespread, this tool provides a high-accuracy, low-latency solution for platforms and developers to verify video authenticity. Its integration into Nvidia's NIM ecosystem makes it highly accessible for deployment in real-time applications. The SVD service crops video frames to 504x504 resolution and scores them from 0 to 1 before averaging the results. While it achieves 92% accuracy on uncompressed videos, performance drops to 87% at 15% compression and 82% at 50% compression, though it maintains a low latency of 22 milliseconds for 1080p videos on Nvidia RTX GPUs.
## BACKGROUND
Nvidia NIM (Nvidia Inference Microservices) is a set of easy-to-use microservices designed to accelerate the deployment of generative AI models across cloud, data center, and workstation environments. Meta's DINOv2 and DINOv3 are self-supervised vision transformer models trained on massive image datasets to learn visual representations without requiring human-annotated labels.