NVIDIA Metropolis
NVIDIA Metropolis is a collection of models, libraries, and blueprints that provides everything you need to build, deploy, and scale video analytics AI agents and applications, from the edge to the cloud. You can now easily transform raw video and sensor data from real-world environments into real-time, actionable insights. This helps your organization understand what’s happening in your physical spaces and respond intelligently, while delivering exceptional scale, throughput, cost-effectiveness, and faster time to production.
How Metropolis Works
Metropolis offers a cohesive, end-to-end stack of software building blocks that handle everything from video ingestion to insight generation to advanced agentic AI-powered analytics. These components can be deployed consistently across the whole compute spectrum—on the edge, in on-prem servers, or in the cloud—so the same applications can run close to where data is generated or centrally at scale.
Agentic Video Search at Scale With NVIDIA VSS Blueprint
Dive into the details of the new VSS 3.0 capabilities in agentic search, modular design, reference workflows, and more.
Create Vision AI Applications With Generative AI Coding Agents
Learn how to generate complete, GPU-accelerated NVIDIA DeepStream video analytics pipelines using simple natural language prompts.
Coming Soon
Post-Training Recipes for NVIDIA Cosmos Reason VLM
See step-by-step workflows, case studies, and technical recipes for NVIDIA Cosmos™ world foundation models.
Solve the Training Data Challenge With the Physical AI Data Factory
Learn how to build a training data pipeline with your own video or image data. Then, curate, augment, and evaluate it with Cosmos for vision AI models.
Get Started With Metropolis
Start using the latest Metropolis vision language models and vision foundation models.
NVIDIA Cosmos Reason
Industry-leading physical AI reasoning vision-language model enables video analytics AI agents to see, understand, and act in the physical world like humans.
Vision AI NIMs
Discover GPU-optimized microservices that bring ready-to-use vision and multimodal models through simple APIs.
Embeddings
Turn images, videos, and text into vector representations for physical AI and multimodal understanding using NVIDIA models like Cosmos Embed, C-RADIO, and NV-CLIP.
Post-train your vision AI models with domain-specific data to boost accuracy.
Cosmos Cookbook
Access recipes to post-train Cosmos Reason VLM and Cosmos Embed with supervised fine-tuning and reinforcement learning.
Learn More
TAO Toolkit
Explore this low-code solution that uses transfer learning to fine-tune and optimize vision AI models with your data.
Start developing vision AI applications with foundational Metropolis frameworks.
Video Search and Summarization (VSS) Blueprint
The VSS blueprint lets you build customizable video analytics AI agents to deliver powerful insights with seamless edge-to-cloud integration.
Try It Now
NVIDIA DeepStream
This is a complete streaming analytics toolkit for AI-based multi-sensor processing, video, audio, and image understanding.
Learn MoreGenerate high-quality synthetic data to safely and efficiently train your AI models.
Physical AI Dataset
Unblock data bottlenecks with an open-source, validated dataset for training vision AI in industries, cities, robotics, and autonomous systems—now free on Hugging Face.
Explore the NVIDIA Physical AI DatasetPhysical AI Data Factory Blueprint
Build a synthetic data generation pipeline with your own video or image data. Then, curate, augment, and evaluate it with Cosmos open world foundation models (WFMs) to accelerate vision AI model development.
Isaac Sim
Enable developers to create realistic synthetic data from complex 3D environments to train vision AI models.
Starter Kits
Develop Video Analytics AI Agents
Build intelligent systems that can see, understand, and interact with the world through computer vision and real-time visual reasoning.
Build a Vision Inference Pipeline
Develop a streaming pipeline with DeepStream that ingests videos, preprocesses frames, and runs optimized vision AI models.
Build Agents for Smart City and Warehouse Operation
Explore examples for optimized VSS blueprint configuration with sample data, tailored prompts, and report templates.
Post-Train Vision Language Model
Refine a vision-language model on task-specific multimodal data so it better aligns visual understanding with domain-specific concepts and instructions.
Fine-Tune Vision Foundation Models
Adapt powerful pre-trained vision backbones with targeted domain data so they specialize on your tasks while retaining their broad visual understanding.
Tech Blog : Build a Real-Time Visual Inspection Pipeline
Generate Synthetic Data
Create synthetic images and videos to expand training datasets, reduce collection costs, and improve vision model robustness across diverse scenarios.
More Resources
Ethical AI
NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their supporting model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
For more detailed information on ethical considerations for this model, please see the Model Card++ Explainability, Bias, Safety & Security, and Privacy Subcards. Please report security vulnerabilities or NVIDIA AI Concerns here.
Develop, deploy, and scale AI-enabled video analytics applications with
NVIDIA Metropolis.