Professor Chris Ré
Stanford CS · Foundation Models
Pioneer of weak supervision and ML systems; cofounder of Together AI and Cartesia. His first ever video-gen investment.
We are building next-generation foundational World Model.
A world model in its native composition: sky above, ground below, characters held at the focal plane.
01 · WORLDS · ARCHITECTURE
A world model that lives in 3D and video.
Geometry, characters, and memory live in a single substrate S. Four channels can ask to change it — sense, edit, imagine, act — but a learned operator κ decides what becomes real. Rendering is one-way; commitment is event-driven.
The model proposes. κ decides. The substrate remembers.
01 · WORLDS · INTERACTIVE
The same substrate that renders the world tracks your position, your gaze, what you've said, and what every character remembers. Step in, walk around, talk to anyone — every reply is grounded in the world's state.
Worlds, on demand.
01 · WALK
A scene you can walk into.
Drop in a raw white-mesh scene. The model textures, lights, and inhabits it — then hands the player WASD and lets them walk.
Live · walking1920 × 1080 · WASD
02 · TALK
And talk to anyone in it.
Every character speaks in-world, in-character, and remembers what you've said — replies are grounded in the substrate's current state.
Live · dialog1920 × 1080 · DIALOG
03 · RERENDER
Then rerender it any way.
Same scene, same physics, four different paintings. Drag the origin to mix render styles in real time — one world, any look.
JAPANESEBYZANTINEWHITE MODELMEDIEVAL
ADVISORS · BACKERSMMXXVI
BACKED · ADVISED BY
Backed and advised by the best.
Founders and operators who've shipped foundational ML, consumer platforms used by billions, and academic research that defines the field. They help us think.
Professor Tengyu Ma
Stanford AI Lab · Founding AI Advisor
Stanford AI Lab Professor; creator of one of the world's best-performing embedding models.
Dr. Andreas Weigend
Former Chief Scientist · Amazon
Advised Jeff Bezos on investment; advisor to Angela Merkel.
Professor Sriram Vishwanath
ECE Chair · Georgia Tech
Leader in distributed systems and information theory. CalTech, Stanford lineage.
Advisors · Operators · ResearchersHover to pause
Justin Kan
Co-founder · Twitch
Mercedes Bent
Partner · Lightspeed
Professor Mike Shou
NUS · AI Video Gen Advisor
Professor Matthew Rhodes-Kropf
MIT + Harvard · Tectonic Ventures
Josh Clemm
VP Engineering · Dropbox
Monisha Perkash
Fmr. Product Lead · Meta
Dr. James Koppel
MIT PhD · CS / Languages
Dr. John Whaley
Stanford CS · Inception Studio
Andrew Radin
CEO/Co-founder · Aria
Tim Shi
Co-founder · Cresta · Recursion
Henry Shi
Anthropic · Founder of Super.com
Anne Lee Skates
Founder & GP · Parable · ex-a16z Partner
Justin Kan
Co-founder · Twitch
Mercedes Bent
Partner · Lightspeed
Professor Mike Shou
NUS · AI Video Gen Advisor
Professor Matthew Rhodes-Kropf
MIT + Harvard · Tectonic Ventures
Josh Clemm
VP Engineering · Dropbox
Monisha Perkash
Fmr. Product Lead · Meta
Dr. James Koppel
MIT PhD · CS / Languages
Dr. John Whaley
Stanford CS · Inception Studio
Andrew Radin
CEO/Co-founder · Aria
Tim Shi
Co-founder · Cresta · Recursion
Henry Shi
Anthropic · Founder of Super.com
Anne Lee Skates
Founder & GP · Parable · ex-a16z Partner
RESEARCH · WORLD MODELS
One model.
Every world we build.
First Intelligenceis a single foundational World Model. Everything you'll see below — robots, generated scenes, drama, consumer apps — is an application built on top of it.
01 · Imagine
Synthesize any scene.
From a single sentence, image, or reference — photoreal, persistent, populated.
02 · Predict
Forward-tick the world.
What happens in the next moment, minute, consequence. Real physics, learned end-to-end.
03 · Embody
Live inside it.
Robots, characters, players — agents who perceive, predict, and act inside the model.
03 · Embodied AI
The world model,
inside the robot.
Real physics · Real consequences
Not a video. A world with mass.
A video model paints a convincing frame and gets the physics wrong. Ours keeps an explicit substrate — geometry, mass, contact, momentum — so it tells a robot which way the box will tip before it pushes it.
We pilot the model inside humanoid and manipulator platforms with selected robotics labs.
03 · EMBODIED AI · CHAPTER 1 OF 3
04 · ROBOT DATA · PHYSICS-BASED
Pure data-driven models hallucinate worlds that don't exist. We're building the new generation — physics-based and 3D-driven — where robots interact with the real world. When real captures aren't enough, the model produces high-quality, physically-correct synthetic data for embodied AI training.
Tired of fake world models.
Grounded in physics.
01 · CAPTURED
Real robot footage on a workstation.
02 · SUBSTRATE
The same scene as 3D, ready to re-render.
0
ZERO HALLUCINATION
Pure data-driven models invent worlds that don't exist. Ours doesn't — every frame obeys real physics.
1:1
REAL-WORLD CONTACT
Robots see, touch, and act on the world as it actually is — not on a learned guess of it.
∞
SYNTHETIC AT SCALE
Physics-grounded 3D-driven data for embodied AI training, at endless variety, when real captures aren't enough.
A new generation → of entertainment
A world you can touch.
Any story you can step into.
We're building the next generation of entertainment — worlds that aren't watched from outside, but lived from within. Tangible, dimensional, responsive.
Pick any film, any show, any scene that's ever moved you. Walk through its door. Talk to its characters. Stay as long as you want.
Not a movie you watch.
A world you arrive in.
02 · PALETTE
An agent that produces an interactive story, end to end, on a canvas you can rearrange.
Palette runs a real production pipeline — script, screenplay, casting, scene design, storyboard, edit — as agent steps on a freeform canvas. Drag any node. Re-run any step. Watch it work, or hand it the brief and walk away.
Live · auto-plays the full pipelineOpen full demo →
03 · INTERACTIVE WORLDS
Watch a story unfold. Talk to anyone in it. Step into the world.
We are building a new loop for interactive worlds — stories you can step inside, powered by a model that can see, understand, and interact with the physical world. Each scene is a doorway: hover any panel to step closer.
01 · Watch
Watch
A short story plays — production-grade output from Palette. Cinematic photography, no UI in the way.
→
02 · Talk
Talk
Pause the scene and ask the character anything. They answer in-character, in-scene, with full memory of the story so far.
→
Inside ep 03 · The Debt● Live
Talk
Look
Walk
03 · Enter
Enter
When the moment lands, step inside. The story becomes a walkable world — talk, look, interact, leave when you want.
01Watch
02Talk
03Enter