NVIDIA announced new generative physical AI advancements at SIGGRAPH, including NVIDIA Metropolis reference workflows for building interactive visual AI agents and NVIDIA NIM microservices that help developers train physical machines to handle complex tasks more effectively.
Behind the Scenes
Three new fVDB NIM microservices support NVIDIA's deep learning framework for 3D worlds.
USD Code, USD Search, and USD Validate NIM microservices work with Universal Scene Description (OpenUSD).
NVIDIA's OpenUSD NIM microservices, combined with the first generative AI models for OpenUSD development, enable developers to incorporate generative AI copilots and agents into USD workflows.
NIM microservices tailored for physical AI support speech, translation, vision, intelligence, realistic animation, and behavior.
Generative AI-powered visual AI agents, built using vision language models (VLMs), are being deployed in hospitals, factories, warehouses, retail stores, airports, and traffic intersections.
Final Take
NVIDIA's generative physical AI advancements are transforming industries by enabling more efficient, safe, and secure operations.
As companies like Foxconn and Pegatron adopt these technologies to streamline their manufacturing processes, the gap between simulation and reality is narrowing. With the help of synthetic data generation, developers can create robust datasets for training physical AI models, leading to more adaptable and better-performing solutions across various industries and use cases.
NVIDIA used its SIGGRAPH announcements to line up a wave of creative software behind Model Context Protocol (MCP), the open standard that lets an AI agent read an application
NVIDIA's Synthetic Video Detector Puts a Deepfake Score Inside Newsroom Workflows July 20, 2026 NVIDIA introduced the Synthetic Video Detector NIM microservice at SIGGRAPH, adding an AI-assisted signal that flags whether a video clip contains synthetic content. It is part of
Alibaba's Qwen team has opened preview access to Qwen3.8-Max , a 2.4-trillion-parameter model the company describes as its most capable system yet, with open weig
