HomeAI NewsNVIDIA Cosmos Data Factory turns two prompts into training-ready datasets

NVIDIA Cosmos Data Factory turns two prompts into training-ready datasets

The platform ships five cloud-native blueprints and six composable skills for curation, labeling, and synthetic data.

NVIDIA released Cosmos Data Factory, a collection of five cloud-native blueprints and six composable skills that run inside the Cosmos open platform for world model development. A coding agent orchestrates the blueprints, so a pair of prompts can drive curation, augmentation, synthetic generation, and labeling without manual steering.

Cosmos is an open platform of frontier models, datasets, frameworks, and skills for world model development. The Orchestration core skill coordinates workflow steps, deploys the services each job needs, and manages execution on a cluster for every reference blueprint. Teams can run Auto Labeling, AnomalyGen, Augmentation, Curation and Retrieval, and Simulation independently or combine them into custom workflows.

The blueprints target concrete jobs: driving scenarios for autonomous vehicle training, image attribute augmentation for search and re-identification, video augmentation across weather and lighting, rare event generation, and defect images for visual inspection. NVIDIA measured throughput on a cloud server with eight RTX PRO 6000 Blackwell GPUs and reported up to 60x faster dataset assembly and 45x cheaper AV scenario generation than traditional methods.

NVIDIA frames the blueprints as a way to cut the tool sprawl that slows physical AI data work. Practitioners can start with the published blueprints and skills, then extend them into custom pipelines. NVIDIA will walk through the workflow in a livestream on building training-ready datasets in two prompts for autonomous vehicles, robotics, and vision AI agents.

What matters

  • NVIDIA Cosmos Data Factory coordinates five blueprints and six skills from a coding agent.
  • The five blueprints cover AV scenarios, weather augmentation, rare events, and defect images.
  • Watch whether the 60x assembly and 45x cost claims hold for teams outside NVIDIA’s eight-GPU cluster.

Why it matters

Watch whether the 60x assembly and 45x cost claims hold for teams outside NVIDIA’s eight-GPU cluster.

This GenAI News article was prepared in original wording using reporting and materials published by NVIDIA Developer Blog. Source reference: https://developer.nvidia.com/blog/build-training-ready-physical-ai-datasets-with-two-prompts-in-nvidia-cosmos/.

Drafted by the GenAI News review pipeline.

latest articles

explore more