Li Fei-Fei's World Labs releases Frameboy, enabling real-time exploration of AI-generated 3D worlds

Li Feifei's World Labs team releases Frameboy: AI-generated worlds can be explored in real time, at up to 15 frames per second

Phoenix Tech Download the app for exclusive early access November 06, 2026 06:27:29 From Beijing

Phoenix Tech News, November 6 (Beijing Time): AI company World Labs, founded by Li Feifei, released a new demo project called Frameboy on November 5 local time, allowing users to explore AI-generated three-dimensional worlds in real time. The project is based on the autoregressive world model Atlas launched in September this year. By optimizing inference efficiency, it streams generated frames in real time from cloud GPUs to the browser, reaching up to 15 frames per second (15 fps). Li Feifei subsequently reposted the related news on social platform X and said: "Real-time Atlas! Try it now!"

Unlike traditional AI video generation tools, Frameboy does not generate a fixed video in one go; instead, it allows users to freely move the viewpoint within the scene, with the AI continuously generating new frames based on the camera's position and orientation. This means users no longer need to predefine a complete camera movement trajectory, and can instead explore virtual spaces built by AI in real time, just like operating a camera in a video game.

World Labs introduced that Frameboy uses a highly optimized Atlas model, greatly improving inference speed while retaining the ability to synthesize novel views. Atlas supports pixel-level camera control, can generate scene images from different angles based on a single image or multiple reference images, and uses spatial context to maintain the three-dimensional consistency of the scene.

In the official demonstration scenarios, users can start from a single image, have the AI complete the originally invisible space, and move and observe within the generated environment. For scenes such as buildings and landmarks in the real world, users can also input photos from multiple viewpoints to let the model generate new viewing angles. World Labs believes that this technology can be applied in the future to fields such as film and television production, game scene design, architectural visualization, and robot simulation.

At the same time, World Labs has also integrated Frameboy's real-time generation capability into its web application Viewfinder. Users can first build a scene frame by frame through Atlas, then switch to Frameboy's real-time rendering mode to explore the generated space. Currently, Viewfinder is still in closed beta, and users need to apply to join the waitlist.

However, Frameboy is still a research preview project. World Labs acknowledges that this technology relies on high-performance cloud GPU servers, and its operating cost is significantly higher than traditional local rendering methods. The officially announced 15 fps is also the maximum frame rate and does not mean that all scenes can stably reach this level.

In September this year, World Labs first released Atlas, positioning it as a multimodal world model for spatial intelligence, capable of processing text, images, video, and three-dimensional spatial information, and supporting scene generation, spatial reconstruction, and robot simulation. The launch of Frameboy this time means that Atlas is further evolving from a model that requires waiting for generation results to a real-time interactive system that can continuously respond to user operations. (Editor/Yu Lei)

For more first-hand news, please download the Phoenix News app and subscribe to Phoenix Tech. For in-depth reports, please search "Phoenix Tech" on WeChat.