We built infrastructure for real-time interactive video models and today we are sharing our first public demo.

Unlike standard video generation where you submit a prompt and wait for a file, this runs continuously and responds to your inputs as it generates. The world updates in real time as you interact with it. We are also opening API access for developers who want to build on top of it.

Try it at reactor.inc. Happy to answer questions.

I tested as Super Mario Bros. Nailed it. What are some known limitations and what is your anticipated timeline for high-fidelity, high-response interactivity?