NVIDIA Unveils 4B-Parameter Cosmos 3 Edge and 92%-Accurate Video Detector at SIGGRAPH
Updated
Updated · NVIDIA Blog · Jul 21
NVIDIA Unveils 4B-Parameter Cosmos 3 Edge and 92%-Accurate Video Detector at SIGGRAPH
3 articles · Updated · NVIDIA Blog · Jul 21
Summary
Cosmos 3 Edge became openly available at SIGGRAPH as NVIDIA’s new 4-billion-parameter world model for real-time, on-device physical AI across robots, vehicles and smart infrastructure.
The model is optimized for Jetson, RTX PRO, DGX and GeForce RTX systems, handles text, image, video, sound and action, and ranks No. 1 on VANTAGE-Bench in its parameter class.
NVIDIA also expanded its AI for Media push with a Synthetic Video Detector NIM microservice that scores footage frame by frame, reaching up to 92% accuracy on uncompressed video and 22-millisecond 1080p processing on RTX systems.
Wowza is embedding that detector into livestreaming workflows used in more than 35,000 deployments across 170 countries, aiming to let broadcasters and other sensitive operators flag questionable video closer to ingest.
The launches were part of a broader SIGGRAPH showcase in Los Angeles, where NVIDIA positioned neural rendering, simulation and agent-ready creative tools as the next layer of graphics and physical AI.
As AI detectors improve, are we stopping deepfakes or just creating more sophisticated, undetectable ones?
When an AI can label a video as fake, who becomes the ultimate arbiter of reality in our news feeds?
NVIDIA’s Synthetic Video Detector: Real-Time, 92% Accurate Deepfake Defense for Critical Industries
Overview
NVIDIA launched its Synthetic Video Detector (SVD) on July 20, 2026, directly addressing the growing threat of deepfakes and synthetic media. Integrated into the Wowza Video Intelligence Framework, the SVD extends enterprise-scale video authenticity and compliance features, helping organizations ensure the integrity of their digital content. By allowing customers to maintain full control over their sensitive data and AI infrastructure, the SVD empowers employees and organizations with AI-assisted detection systems and provides guidance on interpreting results. This launch marks a significant step in combating AI-generated content and supporting video authenticity in an increasingly digital world.