simchowitzlabpublic/nano-world-model

A Minimalist, Batteries-included Repository for Advancing World Model Science.

What it solves

Nano World Model provides a minimalist, open-source framework for training video world models. It addresses the need for a transparent and accessible codebase to predict future video frames based on diffusion-forcing, enabling the simulation of environments for robotics and gaming.

How it works

The project implements video world models using a diffusion-forcing approach. It utilizes a unified pipeline for training, validation, and evaluation managed via a Hydra-based configuration system. The model can perform autoregressive long-horizon rollouts, where it predicts subsequent frames based on previous ones and action injections. These rollouts can then be used for downstream tasks like 3D point cloud reconstruction or Model Predictive Control (MPC) planning.

Who it’s for

This tool is designed for researchers and developers in robotics, computer vision, and AI who want to implement, evaluate, or build upon video world models without the complexity of massive, opaque codebases.

Highlights

  • Minimalist Design: Inspired by NanoGPT, focusing on a clean codebase for scientific transparency.
  • Unified Pipeline: Integrated training and evaluation workflows with easy-to-use configuration overrides.
  • Diverse Dataset Support: Ready-to-use integrations for DINO-WM, RT-1, and CSGO datasets.
  • Practical Applications: Built-in support for long-horizon generation, video-to-3D mapping, and CEM-based planning.

Related

  • Project
    open-gigaai/giga-world-1

    An open-source framework for training and deploying robot world models to evaluate robot policies through controllable video generation.

  • Project
    shengshu-ai/minWM

    A full-stack open-source framework for building action-conditioned video world models, providing a complete pipeline from data preparation to 4-step real-time inference.

  • Project
    boundless-large-model/boundless-world-model

    BWM is a physically consistent, action-conditioned video world model that serves as a high-fidelity simulator for robotic manipulation tasks.

  • Project
    Tencent-Hunyuan/HY-WorldPlay

    HY-World 1.5 is a real-time interactive world modeling framework that uses streaming video diffusion to generate geometrically consistent 3D environments based on user input.

  • Project
    nvidia-cosmos/cosmos-predict1

    A suite of world foundation models for future state prediction that generates visual simulations from text or video prompts to support Physical AI development.