Skip to content

DIAMOND

Train a reinforcement learning agent inside a diffusion world model

Visit official site

What DIAMOND is

The tool is a reinforcement learning agent trained in a diffusion world model for Atari games where visual details matter. It uses autoregressive imagination on a subset of Atari games.

What you can do with it

  • Replay stored episodes in visualization and dataset modes
  • Tune sampling parameters of the diffusion world model
  • Take control with keys such as m

Getting started

  • Clone or download the repository to a local machine
  • Configure the environment with miniconda or python venv
  • Install Python 3.10 and the required libraries
  • Run the pretrained world model and watch the agent

Pros & cons

✓ What we like

  • Visualization and dataset modes replay stored episodes
  • Environment setup works with miniconda or python venv

! What to watch out for

  • Python 3.10 and extra libraries must be installed locally
  • Training covers a subset of Atari games only

FAQ

What does DIAMOND stand for?

The card expands it as DIffusion As a Model Of eNvironment Dreams.

How do I start it?

Clone the repository, set up miniconda or python venv, install dependencies, then run the pretrained world model.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review