NVIDIA Launches Synthetic Video Detector NIM with Up to 92% Accuracy
NVIDIA has introduced the Synthetic Video Detector NIM, an AI-powered microservice designed to analyze videos frame-by-frame to detect AI-generated content. Announced at Siggraph as part of the NVIDIA AI for Media platform, the tool can identify synthetic videos with up to 92% accuracy. As deepfakes and AI-generated misinformation become increasingly sophisticated, this tool provides media organizations and newsrooms with a fast, reliable way to verify video authenticity. Its integration into NVIDIA's NIM ecosystem allows for easy deployment and low-latency processing, which is crucial for real-time editorial workflows. While the detector achieves 92% accuracy on uncompressed videos, its accuracy drops to 85% at 15% compression and 82% at 50% compression. In terms of performance, it can process a 1080p video in 22 milliseconds on an RTX GPU and approximately 30 milliseconds on an enterprise L40 GPU.
## BACKGROUND
NVIDIA NIM (NVIDIA Inference Microservice) is a set of accelerated inference microservices designed to help developers deploy AI models efficiently across various GPU infrastructures. With the rapid rise of generative AI tools that can create highly realistic synthetic media, the industry has faced a growing need for automated, high-throughput detection systems to combat deepfakes.