A powerful video curation system that processes, analyzes, and organizes video content using advanced AI models and distributed computing.

Please run git submodule sync if you have cloned the repository before and just pulled the latest update. We updated the URL for cosmos-xenna submodule on 08/04/2025.

Cosmos-Curate is a comprehensive solution for video processing and curation using state-of-the-art AI models, which powers the training data generation for Cosmos at NVIDIA. It is built on top of a framework optimized for GPU-accelerated streaming pipeline, which is now open-sourced independently as Cosmos-Xenna.

• Video Processing: Efficient video splitting, annotation, filtering, deduplication, and dataset generation
• AI-Powered Analysis: Advanced video analysis using multiple model families
• Distributed Computing: Scalable processing using Cosmos-Xenna built on top of Ray
• Cloud Integration: Support for various platforms
• Pipeline System: Modular and extensible pipeline architecture

Comprehensive documentation is available under docs/ directory.

• End User Guide - For user to run video processing pipelines
• Reference Video Pipelines Guide - general video processing pipelines
• Reference AV Pipelines Guide - multi-camera video, and (upcoming) GPS & LiDAR processing pipelines for autonomous vehicle (AV)

• Developer Guide - For contributors
• Architecture Guide - For developer to understand the core architecture
• Pipeline Design Guide - For developer to add or modify pipelines

cosmos-curate/
├── cosmos_curate/         # Curate implementation
│   ├── client              # CLI to run locally
│       ├── image_cli       # Docker image management
│       ├── local_cli       # Launch pipelines by running local container
│       ├── nvcf_cli        # Launch pipelines on NVIDIA cloud function
│       ├── slurm_cli       # Launch pipelines on Slurm cluster
│       ├── utils           # Common utilities for various CLI apps
│   ├── core/               # Core functionality
│       ├── cf              # Service entry point for a cloud function deployment
│       ├── interfaces      # Core base class to integrate model and define new pipelines
│       ├── managers        # CLIs to run inside the container to manage models, databases, etc.
│       ├── utils           # Common utilities for pipelines
│   ├── models/             # AI model inference implementations
│   ├── pipelines/          # Pipeline implementations
│       ├── examples/       # Minimal example pipelines to help understand the framework
│       ├── video/          # Reference pipelines for video curation
│   ├── scripts/            # Startup scripts in various deployment environments
├── cosmos-xenna            # Git submodule for https://github.com/nvidia-cosmos/cosmos-xenna
├── packages                # Dockerfiles and scripts related to packaging
│   ├── cosmos_curate       # Dockerfile template and conda environment recipes for building cosmos_curate image
├── tests                   # Tests for testing
│   ├── cosmos_curate             
│       ├── pipelines       # Tests for models and pipeline stages for cosmos_curate
│       ├── client          # Tests for client CLIs
├── examples                # Example configuration files and scripts

Note: To initialize and update the cosmos-xenna submodule, run:

git submodule update --init --recursive

This ensures all submodule content is checked out correctly.

For support and questions:

• Check the documentation
• Open an issue on GitHub

• cosmos-xenna team for the core library
• All contributors and users of the project

Responsible Use

This project will download and install additional third-party open source software projects. Review the license terms of these open source projects before use.

NVIDIA Cosmos source code is released under the Apache 2 License.

NVIDIA Cosmos models are released under the NVIDIA Open Model License. For a custom license, please contact [email protected].