LOS ANGELES, July 20 — At SIGGRAPH 2026, NVIDIA has unveiled a comprehensive suite of technologies positioning artificial intelligence as the core engine for creative production, physical robotics, and media integrity. During its keynote address, NVIDIA AI research and engineering leaders emphasized how neural rendering, world models, and agentic systems are converging to transform how digital environments are built, simulated, and governed. Central to NVIDIA’s creative pipeline strategy is the widespread adoption of the Model Context Protocol. By integrating the protocol into leading professional applications, NVIDIA is enabling AI agents to operate directly within production environments while keeping creative control firmly in human hands. Major software providers including Adobe, Affinity by Canva, Blender, Boris FX, Foundry, SideFX, and Epic Games have implemented protocol connectors. These integrations allow agents to automate repetitive tasks such as asset validation, node-tree construction, procedural rigging, and scene inspection. To ensure data security and responsiveness, these local agents run on NVIDIA RTX PRO workstations and DGX systems, keeping sensitive creative assets within controlled studio environments. In the physical AI sector, NVIDIA introduced Cosmos 3 Edge, a four-billion-parameter omnimodel optimized for edge hardware including Jetson, RTX PRO, and GeForce RTX platforms. The model delivers real-time vision analytics and robot action capabilities, eliminating the traditional trade-off between model capability and edge efficiency. The technology is being evaluated by Agile Robots, Doosan Robotics, Siemens, and Skild AI for industrial automation, while also supporting autonomous vehicle policy distillation and smart infrastructure monitoring. Cosmos 3 Edge joins the broader Cosmos platform, with Nano and Super variants now publicly available for developers. Addressing synthetic media threats, NVIDIA launched a Synthetic Video Detector microservice tailored for editorial workflows. The service analyzes video frame-by-frame to generate classification scores, enabling newsrooms to prioritize review and quarantine AI-generated footage. Maintaining high accuracy even after heavy compression, the microservice integrates with infrastructure partners like Wowza, allowing real-time detection at ingest points across global deployments. This deployment model supports on-premises and air-gapped environments, ensuring strict data residency compliance. For organizations requiring on-device intelligence, NVIDIA demonstrated its Agent Toolkit running on the DGX Station. The deskside supercomputer hosts the 550-billion-parameter Nemotron 3 Ultra model alongside harness frameworks, simulation libraries, and a secure runtime. This configuration allows engineering and creative teams to deploy domain-specific agents locally. Partners including LangChain and Nous Research have already optimized frameworks for the platform, enabling persistent, tool-using agents that interact directly with Blender and physics simulators. Underpinning these commercial offerings, NVIDIA presented peer-reviewed research advancing the intersection of computer graphics and physical simulation. Highlights include MotionBricks, a real-time motion model that drives both animated characters and Unitree G1 humanoid robots. Supporting frameworks for generative motor control, photogrammetry cleanup, and a new physics solver for complex materials demonstrate a strategic shift toward localized, agentic, and physically simulated AI infrastructure, fundamentally expanding capabilities available to creators and developers.