Avatar model
Training models to turn a single image into an animated, speaking character.
Experiments in world models, character generation, and tools for working with media.
Training models to turn a single image into an animated, speaking character.
Exploring action conditioning and spatial memory with JEPA in Super Mario.
A 4.4M-parameter world model that predicts a pixel-art environment as the player acts.
A local-first CLI for indexing and searching image and video archives.
A short film I made with AI.
Also training a 21M-parameter world model. Write-up coming soon.

Experiments with worlds you can draw, characters you can talk to, and stories you can steer.
Explore the demos