The startup Odyssey, backed by Pixar co-founder Ed Catmull, has launched a demo version of a new generative AI model that creates interactive video in real-time. Users can navigate virtual worlds using standard game keys, exploring various environments — from a forest area with a cabin to a shopping mall or parking lot. The model streams generated video frames every forty milliseconds, providing almost instant reaction to viewer actions.
Introducing AI video you can watch and interact with, in real-time!
— Odyssey (@odysseyml) May 28, 2025
Powering this is a new world model that imagines and streams video frames every 40ms(!). No game engine in sight.
We call it interactive video, and it's free for anyone to try right now (GPUs permitting)! pic.twitter.com/QtADRXCQ8z
The Odyssey demo is available in the browser and allows interaction with the world for two and a half minutes per session, after which it can be restarted. The image currently remains blurry, and the environment sometimes changes during movement, making the experience similar to a walk in an unstable dream. Some objects have collisions, but often the user can walk through walls or other obstacles.
Odyssey uses clusters of Nvidia H100 graphics processors to generate and stream video at up to thirty frames per second. To build the models, the team developed a special 360-degree camera that captures real landscapes and creates more realistic spaces compared to approaches based solely on open data.
The company emphasizes that the goal is not to replace creative professionals but to collaborate with them. Odyssey is already working on the ability to export scenes generated by the model into popular video editing tools like Unreal Engine, Blender, and Adobe After Effects, where they can be manually refined. Developers plan to gradually expand functionality, adding more interaction possibilities and enhancing the stability of created worlds.
