Atlas world model goes real-time: Frameboy streams cloud-GPU frames to the browser at up to 15 fps
Synopsis
Atlas is an autoregressive world model that previously took tens of seconds to minutes to produce a generated video, so users planned camera paths cautiously. The newly released Frameboy research preview serves a highly optimized Atlas variant from cloud GPUs and streams frames to the browser at up to 15 fps, updating the view continuously as the camera moves. The demo can generate scenes from a single image or text caption and retains and improves novel view synthesis from multiple images, but it must run on a powerful cloud GPU server and costs more than local rendering.
Interpretation
Frames generated by Atlas can now stream back to the browser continuously while the user drives the camera, instead of requiring a generate-then-wait step. Previously users waited tens of seconds or even minutes to see an Atlas-generated video, forcing cautious, methodical scene building and camera-path planning that limited iteration cycles. The Frameboy demo streams frames from cloud GPUs to the browser at up to 15 fps, using an Atlas variant highly optimized for the fastest, most efficient inference while maintaining novel view synthesis.
Real-time interaction turns a single generation into continuous navigation, letting users try ideas quickly. In manual iteration a user clicks four keypoint cameras into a room, presses generate, waits for the video, then drags a keypoint and waits again; in real-time interaction the camera moves around the room and the view updates continuously. The comparison is shown as two views of the same living room, illustrating that once latency and cost drop, users no longer perceive a discrete act of generation.
The same real-time capability is integrated into the web app Viewfinder, where users build a scene frame by frame and then explore it through a streaming overlay. Frameboy offers a curated list of scenes, while Viewfinder lets users build a scene from their own inputs and generate 3D-consistent views from different camera angles. Each new frame in Viewfinder is generated by Atlas with both color and depth to visualize the 3D space; the app is currently in closed beta with a waitlist.
Perspective
The demonstrated scope is a research preview in which Frameboy streams frames from a highly optimized Atlas variant on a powerful cloud GPU server to the browser at up to 15 fps; scenes can start from a single image or text caption or be driven by multiview captures of real places, and Viewfinder's real-time overlay render mode is currently in closed beta.
The material states that Frameboy currently must run on a powerful cloud GPU server and is not cheap compared with local rendering methods; the real-time variant keeps novel view synthesis while speeding up inference, but no quantitative comparison is given.