NVIDIA links agentic tools and physical AI at SIGGRAPH

NVIDIA presented advances in neural rendering, world models and simulation at SIGGRAPH, positioning the technologies for creative tools, industrial design, robotics and autonomous systems. Its keynote highlighted research work on preserving artistic intent, maintaining temporal stability and rendering 4K content in real time, alongside AI physics work for weather, automotive aerodynamics and thermal design through NVIDIA Earth-2.

The company introduced Cosmos 3 Edge, a 4-billion-parameter omnimodel optimized for local deployment on NVIDIA Jetson, RTX PRO, DGX and GeForce RTX systems. The model extends Cosmos 3 with text, image, video, ambient sound and action capabilities for on-device physical AI, while Cosmos-Dreams adds closed-loop simulators that can generate autonomous-vehicle worlds from a single frame.

Creative software vendors are adding Model Context Protocol connections so agents can inspect scenes, automate production tasks and work across tools while creators retain control. Adobe, Canva, Blender, Boris FX, Foundry, SideFX and Unreal Engine are among the platforms exposing MCP or MCP-ready workflows, with NVIDIA Agent Toolkit supporting local agents on RTX PRO workstations, DGX Spark and DGX Station systems.

NVIDIA also announced a Synthetic Video Detector NIM microservice that scores video frame by frame for synthetic content, reaching up to 92% accuracy on uncompressed video in NVIDIA testing and processing 1080p video in as little as 22 milliseconds on NVIDIA RTX systems. Additional research included 21 accepted technical papers and MotionBricks, a real-time motion model trained on more than 350,000 motion clips.