NVIDIA Cosmos™ is a platform of state-of-the-art generative world foundation models (WFMs), advanced tokenizers, guardrails, and an accelerated data processing and curation pipeline. It is built to power world model training and accelerate physical AI development for autonomous vehicles (AVs) and robots.

Features

• Cosmos Predict: Generates up to 30 seconds of continuous video from multimodal inputs with superior speed, fidelity, and prompt adherence.
• Cosmos Transfer: Multicontrol model to generate world states across different environments and lighting using ground-truth and structured inputs.
• Cosmos Reason: Fully customizable, multimodal reasoning model for planning response based on spatial and temporal understanding.
• Efficient Tokenization: Use Cosmos tokenizers to generate image or video tokens at higher compression rates.
• Accelerated Data Curation: Speed up data curation by 20X with the NVIDIA NeMo™ Curator pipeline.
• Post-Training Scripts: Customize Cosmos WFMs for downstream Physical AI use cases using PyTorch scripts.

Benefits

• Accelerate Virtual World Generation: Provides developers with easy access to high-performance world foundation models, data pipelines, and tools to generate synthetic data.
• Physics First Data: Multimodal models pretrained on 20 million hours of robotics and driving data to generate world states grounded in physics.
• Open Access: Cosmos WFMs, guardrails, and tokenizers are licensed under the NVIDIA Open Model License.
• Continuous Improvement: Easily build data curation, augmentation, evaluation, and simulation flywheel with models and tools built for custom physical AI foundation model development.
• Rapid Deployment: Rapidly deploy Cosmos WFMs using NIM microservices and scale agentic AI and synthetic data generation workflows.