World Labs, founded by Li Fei-Fei, has launched its first application, Frameboy, an optimized demo of the Atlas world model. Powered by cloud GPU streaming, it achieves a maximum frame rate of 15fps with roughly 70ms latency, delivering an experience akin to cloud gaming. Frameboy offers a curated selection of scenes for users to explore: it can generate new viewpoints from a single image to build an explorable environment, or produce new viewpoints both inside and outside photos based on multiple images of the same location. It is also integrated into the Viewfinder app, enabling users to input their own content, construct scenes frame by frame, and explore and expand virtual worlds in real time. Users can switch viewpoints as if operating a camera, generating three-dimensional, consistent world views, with each frame containing both color and depth information. At its core, Atlas is a general-purpose world model pretrained from scratch that can process text, images, video, and 3D spatial information. Its key mechanism is "spatial context," which associates images with the corresponding camera's 3D position and orientation, accepts precise camera poses, and generates images from the corresponding viewpoints—setting it apart from ordinary video generation models that control camera movement through text prompts. Atlas adopts a multimodal autoregressive diffusion Transformer architecture, can generate high-resolution long videos, and can also output 3D representations such as depth information, point clouds, and 3D Gaussian splatting, supporting applications such as achieving bullet-time effects and building robot simulation environments at low cost. Frameboy's launch also opens up new creative possibilities in fields such as film location scouting, spatial concept design, and game level ideation.