A world model called 'Atlas' has appeared that can generate images, videos, and 3D spaces. It boasts a wide range of functions, including 'generating a 3D space from a single image,' 'camera work that is faithful to instructions,' and '1440p video generation.'
A 'world model' refers to an AI model that can learn and reproduce the physical properties and structures of objects in the real world. AI company World Labs has recently announced a new world model called ' Atlas .'
Atlas: A World Model for Spatial Intelligence | World Labs
Introducing Atlas; A Foundation Model for Spatial Intelligence - YouTube
Atlas can generate a video that shows 'the world surrounding that image' based on a single image. The video can be up to 1 minute long, with a maximum resolution of 1440p, and is characterized by its ability to accurately reflect instructions such as camera movement and angle.
The following are the results of comparing the camera control performance of Atlas with other AI models, as judged by humans. The horizontal axis shows the percentage of times Atlas was rated as superior. It achieved an 81% win rate against Google's Gemini Omni Flash and a 75% win rate against MiniMax's MiniMax H3.
It can generate not only videos but also images. In the example below, the image of the stuffed animal in the upper left is input, and the camera angle is changed to output an image of the stuffed animal taken from behind.
You can also recreate scenery by inputting multiple reference images.
It also supports simple image generation.
In addition to images and videos, it can also output 3D spaces. Below is an example of generating a 3D space based on a single image. First, input 'an image of a corridor somewhere'.
Atlas generates scenery for parts of the image that are not included.
Atlas is currently available to a limited number of partners, with early access expected to begin within the next few weeks. Users who wish to participate in early access will need to register for the waitlist.
Atlas is our scalable foundation model that can perceive, generate, reason, and interact with virtual and physical worlds.
— World Labs (@theworldlabs) September 1, 2026
We're opening up early access in the coming weeks.
Learn more and sign up here: https://t.co/SHjFaNWedP
Related Posts:
