Presentation
From Pixels to Physics: Building AI's Right Brain in the Open
DescriptionFoundational models aren't new, but building LTX taught us their scope is expanding faster and further than anyone predicted. LTX is our multimodal generative world model, and in this talk Ofir shows how a system trained to do one deceptively simple thing, generate pixels across time, a.k.a. video, becomes a physically grounded foundation that reasons about light, physics, and causality. If language models are the left brain of AI, LTX is a bet on the right: the intuitive, perceptual, physical side of intelligence.
We'll trace LTX's journey from a state-of-the-art open-weights model to core infrastructure across domains, from media and entertainment to physical AI. Expect real applications spanning robotics and simulation, a look under the hood at the architectural shifts that made it possible (MoE, stronger text encoders), and the conviction driving the project: when a model is open and production-ready, it compounds, for researchers, developers, enterprises, and the applications this room will build next.
We'll trace LTX's journey from a state-of-the-art open-weights model to core infrastructure across domains, from media and entertainment to physical AI. Expect real applications spanning robotics and simulation, a look under the hood at the architectural shifts that made it possible (MoE, stronger text encoders), and the conviction driving the project: when a model is open and production-ready, it compounds, for researchers, developers, enterprises, and the applications this room will build next.
Event Type
Stage Session
TimeTuesday, 21 July 20261:00pm - 1:30pm PDT
LocationWest Hall A
Gaming & Interactive
New Technologies
Production & Animation
Research & Education
Artificial Intelligence/Machine Learning
Computer Vision
Generative AI
Graphics Systems Architecture
Physical AI
Real-Time
Rendering
Robotics
Simulation
Spatial Computing
Full Conference Supporter
Full Conference
Experience
Discover
