World Labs Previews Real-Time World Generation With Atlas
The AI model creates new views as you move, using reference images and earlier frames to keep the scene consistent.
World Labs introduced Atlas (opens in a new tab) on September 1 as a model for creating, reconstructing, and simulating worlds. A week later, the company has previewed a version that generates those worlds in real time.
The September 8 demonstration (opens in a new tab) shows a person exploring scenes with separate Move and Look controls. Atlas creates the next view as the person changes direction, giving the experience the controls of a first-person game.
Generation
Atlas generates frames one at a time, using earlier images to guide what comes next. World Labs says it has optimized a version of the model to run fast enough for interactive exploration.
The demo moves through a plant-filled room, a kitchen, and a courtyard. The interface, labeled Frameboy Advance, shows movement and viewing inputs beside the scene as it changes.
Reconstruction
World Labs says Atlas maintains 3D consistency by building a spatial context from reference images and previously generated frames. Each view carries a position in space, helping the model keep track of where things belong.
Photos of a real location can also become part of that context, allowing a person to explore a reconstruction of the place. More reference views give Atlas more information about the scene and leave fewer unseen areas for it to imagine.
Early Access
World Labs has not announced the real-time version’s frame rate, resolution, hardware requirements, or pricing. Atlas is entering early access with select partners.
Developers can request access through World Labs (opens in a new tab). The company says Atlas will power future versions of Marble and other products, bringing its world-generation and reconstruction tools into more creative workflows.
More from Capture & Create.
Browse all stories
Blockworks · 2 min read
Blockworks Turns Your Words Into Builds with AI on Apple Vision Pro
A new AI demo makes a boat, a bicycle picture, and a statue from virtual blocks you can pick up and take apart.
Tripo AI · 4 min read
Tripo Adds 3D Animation From Text Prompts
Creators can generate short actions from prompts and combine up to five motion segments, with skeleton options for Unreal Engine, Unity, and avatar workflows.
Robot simulation · 1 min read
NVIDIA’s Robo Olympics sends simulated robots sprinting, skating, and kickboxing
Sports videos and written prompts guided OpenAI’s Astra in building the events, with NVIDIA’s simulation tools turning motion references into tested robot movements.