Orbis 2

University of Freiburg · Released July 2026

A hierarchical driving world model: a low-frame-rate predictor supplies long-range context and a high-rate predictor fills in detail, generating long future video from past frames. An optional steering trajectory lets you ask what happens if the car turns. Released under MIT.

Weights
Open weightsMIT
Commercial use
Permissive license
Real-time
No real-time claim
Where to try
Run it yourself

01 Controls

What you control

  • An input driving video to take context from
  • An optional trajectory file that steers the car

02 License

Can you use it commercially?

Permissive license

The weights' model card declares MIT, a permissive license that allows commercial use. Check what the model was built on, because a base model's terms can still apply.

Declared license: MIT · Model card

A reading of the published license text, not legal advice.

03 Speed

Does it run in real time?

No real-time claim. The cited sources make no real-time claim.

Figures are the authors' own, with their settings and hardware. They are not comparable across models.

04 Setup

How to run it

  1. 01Clone lmb-freiburg/orbis2 and create the conda environment from environment.yml.
  2. 02Download the tokenizer and the L1 and L2 checkpoints with hf download sud0301/orbis2.
  3. 03Run evaluate/rollout_demo_v2.py on an input video; add a trajectory file to steer the rollout.

Summarised from the project's own instructions. Follow the source for exact commands and versions.

05 Access

Where to try it

No hosted version was verified. Run the published weights yourself.

Listed alphabetically. Windflow maintains this directory and is one of the hosts; a listing does not promise capacity at any given moment.

06 Questions

Orbis 2 FAQ

Is Orbis 2 open source?

Orbis 2 publishes downloadable weights under MIT. The declared license is permissive.

Can I use Orbis 2 commercially?

The weights' model card declares MIT, a permissive license that allows commercial use. Check what the model was built on, because a base model's terms can still apply.

Does Orbis 2 run in real time?

The cited sources do not claim real-time operation.

Where can I try Orbis 2?

No hosted version was verified; run the published weights yourself.

What does Orbis 2 take as input?

You give it an input driving video to take context from and an optional trajectory file that steers the car.

How do I run Orbis 2?

Following the official instructions: Clone lmb-freiburg/orbis2 and create the conda environment from environment.yml. Download the tokenizer and the L1 and L2 checkpoints with hf download sud0301/orbis2. Run evaluate/rollout_demo_v2.py on an input video; add a trajectory file to steer the rollout.

What are alternatives to Orbis 2?

Similar models in this directory include comma.ai world model 4B (open weights, permissive license), Oasis 3 (no public weights, service terms) and Orbis (open weights, permissive license).