QuantWorlds
R

OpenDreamer

by Reactor · Released 21 July 2026

Quick answers

Can I try it?
Yes — it's publicly accessible. See 1 World built with it
Is it interactive?
Yes — it supports real-time interactive control, not just a pre-generated output. Browse other interactive models
Is it real-time?
Yes.
Is it open source?
No.
Does it have an API?
No public API is currently available.
Can I use it commercially?
Commercial use not permitted — The training and inference code is publicly visible on GitHub (github.com/next-state/open-dreamer and github.com/reactor-team/open-dreamer), but both repositories currently state verbatim: "All rights reserved. See LICENSE. This is a temporary notice; a formal license is expected in a future release." No open-source or commercial license is currently granted, despite the project being widely described elsewhere as "open-sourced." Training/inference code is publicly available; trained model weights are NOT currently distributed anywhere — running OpenDreamer requires training your own checkpoint or supplying one you already have.
What license does it use?
All rights reserved (temporary notice; formal license expected in future release)

About

An open reproduction of DeepMind's Dreamer4 world-model pipeline, built by researchers Diego Marti Monso, Francesco Sacco and Edward Hu with infrastructure support from Reactor. A 1.6B-parameter dynamics model (30 block-causal transformer layers, JAX/Flax NNX) trained on Minecraft/VPT-style gameplay data, generating real-time interactive video frames from live keyboard and mouse input.

Input types

keyboardmouse

Output types

video

Worlds built with OpenDreamer