NVIDIA Cosmos Curator Overview

Description:

NVIDIA Cosmos Curator is a comprehensive solution for video processing and curation using state-of-the-art AI models. It powers training data generation for the Cosmos model family at NVIDIA. The container is built on top of Cosmos-Xenna, a GPU-accelerated streaming pipeline framework that is open-sourced independently. Cosmos Curator processes, analyzes, and organizes large-scale video content using advanced AI models and distributed computing, enabling researchers and engineers to build high-quality video datasets for AI model training.

The container does not include or distribute model weights. Users are prompted to navigate to the appropriate source and download model weights separately as required.

The container components are ready for commercial or non-commercial use.

License/Terms of Use:

The container is governed by the NVIDIA Software License Agreement and the Product-Specific Terms for NVIDIA AI Products. This container includes additional third-party open source software projects. Review the license terms of these open source projects before use.

You are responsible for ensuring that your use of NVIDIA provided software complies with all applicable laws.

This container includes third-party open-source software. OSS Sources

Attributions

Open Source License Attribution Cosmos Curator uses Open Source components. You can find the details of these open-source projects along with license information here

Deployment Geography:

Global

Program Classes:

The NVIDIA Cosmos Curator container is a software tool (Program Class: Container / Software). It does not contain or distribute embedded model weights. Users must manually navigate to the appropriate source and download required model weights prior to running curation pipelines.

This container does not include models. No model card links are applicable.

Deployment Details:

Supported Hardware Microarchitecture Compatibility: GPUs: One or more NVIDIA GPUs with minimum CUDA compute capability 8.0 and minimum GPU memory of 4 GB to run the hello-world pipeline, or 48 GB to run the reference video pipelines
CPU: x86_64 host CPU compatible with NVIDIA GPU drivers and CUDA runtime
Memory: Minimum 32 GB host memory
Storage: Minimum 200 GB disk space

This container is designed and optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA’s hardware (GPU cores) and software frameworks (CUDA libraries), Cosmos Curator achieves high-throughput video processing through a GPU-accelerated streaming pipeline.

Running Cosmos Curator

Users will need to provide a compatible ffmpeg distribution based on version 8.1.1. It can be built from source or downloaded via conda

micromamba create -y     -p "${PWD}/cosmos-curator-ffmpeg"     -c conda-forge     'ffmpeg=8.1.1=lgpl_*'

or

pixi global install --environment ffmpeg --channel conda-forge --expose ffmpeg --expose ffprobe 'ffmpeg=8.1.1=lgpl_*'

And then when launching locally via the End User Guide quickstart supply the downloaded path

For example:

cosmos-curator local launch \
  --image-tag 2.0.0 \
  --image-name  nvcr.io/nvidia/cosmos/cosmos-curator \
  --extra-volumes "${PWD}/cosmos-curator-ffmpeg:/opt/ffmpeg:ro" \
  -- pixi run -e unified python -m cosmos_curator.pipelines.video.run_pipeline split\

Reference(s):

• Documentation — NGC container page and End User Guide quickstart
• Changelog
• Source / Project Repository
• Reporting Issues — NGC container page
• Cosmos-Xenna — underlying GPU-accelerated streaming pipeline framework

Container Version(s):

Security Common Vulnerabilities and Exposures (CVEs)

Please review the Security Scanning tab on NGC to view the latest security scan results. For certain open-source vulnerabilities listed in the scan results, NVIDIA provides a response in the form of a Vulnerability Exploitability eXchange (VEX) document. The VEX information can be reviewed and downloaded from the Security Scanning tab.

Ethical Considerations:

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. Developers should work with their internal developer team to ensure these software components meet requirements for the relevant industry and use case and address unforeseen product misuse.

Users are responsible for inputs and outputs. Users are responsible for ensuring safe integration, including implementing guardrails as well as other safety mechanisms, prior to deployment.

Please report quality, risk, security vulnerabilities or NVIDIA AI Concerns here.

Get Help

Getting Started with Cosmos Curator

Deploying Cosmos Curator is straightforward — with the container image you can skip straight to the local run commands, no build required. Visit the Cosmos Curator NGC page for the End User Guide quickstart, release documentation, and deployment guides.

Support

For questions and support, visit the Cosmos Curator support page.