Fei-Fei Li's World Labs unveils Atlas, a multimodal world model that generates image/video frames with pixel-perfect camera control and reconstructs them in 3D

World models generate, reconstruct, and simulate any possible world. They understand how worlds appear, behave …

World Labs 2026-09-02

Context & Ripple Effects

World Labs’ spatial-AI arc began with a 2024 preview that turned an image into a game-like 3D scene, then broadened with Marble’s editable 3D environments from prompts, photos and other media in 2025. Atlas extends that work from scene generation into controlled image/video views and 3D reconstruction.

The company’s $1 billion financing was explicitly aimed at world models for robotics and scientific discovery, while Tencent had already released an open-weights model for 3D-consistent video from a single image. The competitive question is increasingly whether a model can preserve usable spatial structure, not merely produce plausible footage.

First-order effects

World Labs broadens its offering beyond Marble’s editable environments to a model that combines view-controlled image and video generation with 3D reconstruction.
Public Atlas demonstrations describe turning footage from a few phone cameras into new camera angles and generating portions of a scene that were not filmed, giving creators a spatially aware post-production workflow.

Second-order effects

Tencent’s HunyuanWorld-Voyager and other world-model developers face a sharper benchmark: 3D-consistent output alone is less differentiating when camera control and reconstruction are presented as one system.
Atlas gives World Labs a concrete capability set to pursue the robotics and scientific-discovery applications cited in its funding round, without establishing any confirmed integration with Autodesk products.

Third-order effects

World models are developing into a shared spatial layer for synthetic media and physical-AI simulation, where controllable viewpoints and recoverable geometry matter alongside visual quality.
If model builders continue combining generation with reconstruction, competition will shift from producing individual clips toward controlling reusable, navigable representations of scenes.

The trend: Generative AI is moving from making isolated images and videos toward world models that maintain and manipulate a scene’s spatial structure.

Discussion

@erynqian Eryn Qian on x
I'm behind the 3D side of this magic, turning them into Gaussian splats you can actually move through. I spend my days with my face pressed against Atlas's worlds. What looks glam from across the room has to stay sharp an inch from the wall. Still feeling surreal.
@eerac Eric Rachlin on x
One cool use for Atlas is creating “bullet time
@venturetwins Justine Moore on x
Re-shoot an existing video from a new angle 🤯 Truly insane work from @theworldlabs, this new model is a game-changer. The initial clip was filmed with three cell phones on tripods - the model reconstructed a full view of the scene to reframe the video.
@theworldlabs @theworldlabs on x
Introducing Atlas: The world's first multimodal world model that generates image and video frames with pixel-perfect camera control and reconstructs them in 3D. Model the world, move the camera, and simulate space & time.
@davidpantera_ David Pantera on x
Two of us filmed this on a couple phones, and we used Atlas to turn it into a short film. This is just a clip from the @theworldlabs tech blog Atlas is able to generate the parts of the scene that none of the cameras actually filmed. It can freeze time, allowing you to reframe sh…
@haozhang623 Hao Zhang on x
Atlas is out! What excites me most is the 3D capability, months of work with @bardienus : an AR omni model that beats every specialized model at 3D reconstruction. First clip: fly through the Porsche museum, 500+ frames streamed into a single consistent point cloud; feed-forward …
@davidpantera_ David Pantera on x
With Atlas, every motion is something you direct, not describe. I made this with a single input image. Atlas uses precise camera geometry as a native input type, going beyond prompt-based instructions for camera control. This lets you actually frame every shot and control every m…
@xrarchitect Ian Curtis on x
This scene was generated from one input image. Atlas fills the remaining gaps. → atlas (world labs) → spark.js → three.js
@mtslive @mtslive on x
World Labs co-founder @jcjohnss explains the bet behind Atlas that world models will do for visual and physical understanding what LLMs did for language:
@theworldlabs @theworldlabs on x
By positioning multiple input views within the model's spatial context, Atlas allows you to generate image and video frames with precise control. Here, we hand-designed a camera trajectory to generate a 1 minute video at 1440p resolution from seven reference images.
@scobleizer Robert Scoble on x
Another big step to getting a Holodeck. Fable 5.1 is more important for today but this is more important to the future. Big news day in AI.
@dr_singularity Dr Singularity on x
We will have Holodeck technology before 2030.
@theworldlabs @theworldlabs on x
For robotic simulation, Atlas reconstructs a space from just a few photos and generates the photorealistic RGB and depth data any robot's sensors would observe on any trajectory. Robots can now be trained and tested in far more spaces. Until now, scanning spaces like these requir…
@benmildenhall Ben Mildenhall on x
our new model Atlas also happens to be a capable text-to-image generator, providing it with a strong foundation of “world knowledge”... this generalizes across both single and multi view domains, meaning you can step into even highly stylized scenes with precise 3D control
@k_s_schwarz Katja Schwarz on x
So excited to share what we've been working on! Atlas is honestly just so fun to work with!! One of my favorite parts: Camera control is insanely good, allowing you to do crazy flythroughs. How many examples can you spot in the blog where we fly through sth weird? 🚀
@ychngji6 Chongjie Ye on x
A pretty special day for me — Atlas is out! I started my PhD fascinated by how much of a real 3D world we can recover from just a few 2D views. Years later, that became my first project at @theworldlabs. Still so much to solve. More soon 🗺️
@brittaninatali Brittani Natali on x
I play with splats all day, every day, and I've never seen anything like this. The scale and fidelity of these worlds is insane. Welcome to Earth, Atlas.
@benmildenhall Ben Mildenhall on x
I'm particularly excited by Atlas's ability to reconstruct scenes from a very small number of input image — was able to create this flythrough of London's Natural History Museum by combining 3 input images I found from completely separate sources on Google Images
@neilsonks Neilson on x
whats interesting is you can “trick” Atlas into becoming a 3D modeling tool by placing input images in 3D space the model then creates a coherent world this is how this scene was made - all the model sees is camera pose + RGB
@benmildenhall Ben Mildenhall on x
Atlas is an autoregressive diffusion model built from the ground up for the task of “next frame prediction”. It is simultaneously a world class method for camera-controlled video generation, novel view synthesis, and sparse 3D reconstruction.
@venturetwins Justine Moore on x
Incredible launch from World Labs ✨ Atlas is a new multimodal world model that generates up to 1 minute of video with precise camera control and spatial consistency. The model takes text and image as inputs, as well as camera paths. Watch it 👇
@theworldlabs @theworldlabs on x
Atlas understands both the spatial structure of the world and how it evolves over time. This lets us turn a handful of ordinary cameras into a “bullet time” multiview capture studio, capturing dynamic moments frozen in time.
@jasteinerman Jake Steinerman on x
The new Atlas world model is literally black magic. This shouldn't be possible from just 3 cameras 🤯
@davidpantera_ David Pantera on x
Make a video that uses angles you never actually filmed, thanks to Atlas. We filmed this using a couple of phones on tripods, yet Atlas can generate the angles of the scene that none of the cameras ever filmed, because it's a world model. Also, jumper tips welcome :)
@a16z @a16z on x
World Labs CEO @drfeifei on the need for world models: “There's no language out there. You don't go out in nature and there's words written in the sky for you.”
@keunhongp Keunhong Park on x
Atlas can control space and time to create “bullet time” videos. Having worked on dynamic 3D reconstruction during my Ph.D, it's shocking how easy this is with Atlas. Our interns @DrTunnels @HaoZhang623 @davidpantera_ @ZeCh452318 had a lot of fun smashing things.
@tamrrat Tamrat on x
umm wait, what in the 4D video is this??
@petergyang Peter Yang on x
Yeah Fable 5.1 is really cool but this is bonkers
@theworldlabs @theworldlabs on x
We pre-trained Atlas from scratch to take multimodal inputs, including camera movement, and turn it into 3D grounded views. Atlas puts you in control: direct the views, reconstruct real spaces with your inputs, and build explorable worlds. Read more: https://www.worldlabs.ai/...
@martin_casado @martin_casado on x
🔥An incredible accomplishment. Think of it as a video model with full camera control. And the scene remains (nearly) 3D consistent. Built on a fully internal base model. There are many use cases, from video editing, to 3D reconstruction to robotics. Great technical blog too.
@ccatalini Christian Catalini on x
The scaling economics of world models are different: everywhere else, verification is the bottleneck. Here you just simulate your way out of it with more compute. https://x.com/...
@ziqi__ma Ziqi Ma on x
Atlas can do all the really hard things with camera-controlled frame gen - orbits, large rotations, getting close to objects without degenerating etc., along with everything you get from an omni ar model. And it can give you a pikachu on the canvas after turning 180😊!So proud to …
@gowthami_s @gowthami_s on x
Happy to share what I've been working on since I joined World Labs early this year: Atlas, our next-generation world model for spatial intelligence! Atlas can generate a world, reconstruct a real one, and simulate how it changes through space and time. The range is hard to descri…
@yunzhuliyz Yunzhu Li on x
Another glimpse of Real2Sim with Atlas: capture a real environment with just a number of casual photos from a phone, turn them into a sim, then simulate diverse robots navigating different trajectories. Atlas generates the RGB and depth observations they would see along the way.
@jcjohnss Justin Johnson on x
Today we sharing Atlas, our new multimodal world model. One reason this model is so special to me is that it combines two core visual intelligence tasks I have worked on for over a decade: generation and reconstruction
@kaiwynd Kaifeng Zhang on x
As a robotics PhD, I've been asked many times why I chose to intern at World Labs. The reason is, I've always believed that learned models can fundamentally change how we build simulation, perception, and interaction systems for robots, if we use them in the right way. And now he…
@yunzhuliyz Yunzhu Li on x
One month into joining World Labs, it's been incredible to witness Atlas come to life under the leadership of @jcjohnss, @BenMildenhall, and @drfeifei. 🚀 Proud to contribute on the robotics side and show how Atlas can bridge world models and physical AI: turning a few casual real…
@davidpantera_ David Pantera on x
@drfeifei uses folding laundry as a prime example of how difficult everyday physical tasks are for AI because current AI lacks true spatial intelligence. Atlas is a world model that takes a BIG step towards solving that problem. We filmed this with her using just a couple of phon…
@sarahdingwang Sarah Wang on x
“Model the world, move the camera, and simulate space & time.” Absolutely incredible. Congrats to @drfeifei and the entire @theworldlabs team!
@drfeifei Fei-Fei Li on x
I'm so excited that our @theworldlabs team has achieved a major milestone today! Introducing Atlas - a first of its kind multimodal world model trained from scratch! 🚀 Atlas is capable of generating frames with pixel-perfect camera control, reconstructing large scenes from as few…
@theworldlabs @theworldlabs on x
Atlas also outputs explicit 3D from one to many input images, beating top open source reconstruction models. Passing more images gives Atlas more context: the more it sees, the less it imagines.
@xrarchitect Ian Curtis on x
can't be more proud of the team. I've been playing around with this model and there is something really special about moving/directing inside the world and still having persistence wherever you travel
@wendlerch Chris Wendler on x
I am so excited to share what we have been up to! TLDR; We built Atlas — the best camera conditioned frame autoregressive transformer in the world. Finally, I can tick training a frontier model off my bucket list :):) Tysm & shoutout to my incredible team at @theworldlabs
@chaoyuaw Chao-Yuan Wu on x
A world model promises one thing: predicting how world operates & changes over space and time. It's extremely hard to completely solve this (as it pretty much means modeling the whole universe and beyond), but Atlas has made a sizable step toward it. You know something is right w…
@dan_jeffries1 Daniel Jeffries on x
In a decade the combination of... 1) Much better chips and memory architectures, if not radically new ones 2) Internalized world models 3) Real continual learning that changes weights on the fly 4) Reasoning in embedded space ...will make today's models look like child's play.
@davidpantera_ David Pantera on x
Introducing Atlas, our new omni world model for spatial intelligence. If you'd like to build with it, request access below and we'll reach out! We're excited to see what you build, and to work with you to make Atlas the go-to world model for generating, reconstructing, and simula…
@jpatel41 Jeetu Patel on x
Congratulations to @drfeifei and the entire @theworldlabs team. This is super exciting and consequential to the world and AI advancement. You clearly need more than words to express the significance of this ;-).

Chronicles