Building the interaction layer
for AI-generated worlds.
We train models and build experiences where people can talk, act, explore, and influence what happens next.
interactive media
From generated content
to continuous interaction.
A generated scene is a starting point. An interactive world needs to remember what happened, respond to an action, and carry that change forward.
Our work connects world models, AI agents, and generative media to explore that missing layer.
Research in practice
Pixel-Art-GWM
A compact, action-conditioned model exploring how to preserve a controllable character as its world changes.
Representation learningJEPA in Super Mario
Exploring temporal consistency, spatial memory, and action conditioning on a trained Mario model.
Our model trainingTalking avatars
Training our own avatar models to turn a single image into a speaking character with a consistent visual identity.
Real-time generative mediaR24
An autonomous newsroom that selects stories and generates anchors and correspondents seconds before broadcast.

Building with a community.
Through Inception hackathons, mini builds, and fireside chats, we bring builders together to turn world models into interactive experiences.
Next: BRANCH at USC School of Cinematic Arts and Make It Playable.
Events, people & newsletterWriting from the studio
Our editions on interactive media and world models.
Visit the newsletter
Beyond the authored world
How AI-native games could move beyond prewritten environments, and why model scale, persistent architectures, and orchestration represent three different paths to interactive worlds.

The world model moment
Why world models need memory, consistency, and consequences—not just convincing frames. An update from our builder community and the gaming, interactive-media, and robotics tracks at Inception II.

The end of static media
Our first edition explores the move from fixed recordings to media that responds in real time, with a look at Reactor, compact world models, and our early interactive experiments.
