ByteDance Preparing Real-Time Spatial Video AI Model Under Zhang Yiming

ByteDance is preparing an artificial intelligence model for real-time spatial video generation, which could launch as soon as next month. [1][3] Founder Zhang Yiming is personally overseeing the development of this new system. [1][3] The upcoming model is expected to build upon ByteDance's existing Seedance video generation technology. [3] This system would allow users to generate interactive virtual worlds tailored for games, short-form dramas, and live streams. [3] The reported specifications include cloud-generated, on-demand video running at approximately 20 frames per second with a latency of around 0.05 seconds. [3]

Rendering this spatial content remotely is intended to reduce the computational requirements and cost of the user's hardware. [3] The model is reportedly designed to generate virtual environments that react to the movements and voices of users wearing headsets from Pico, ByteDance's extended reality division. [3] World models were previously reported as ByteDance's top AI priority for 2026, commanding the company's largest data budget at three to four times the spending of its rivals. [3] The company explicitly targeted shipping at least one world model by the end of the year to measure against Google's Genie. [3]

If successful, ByteDance's approach could shift the extended reality competition away from local hardware specifications and toward cloud capacity and model capabilities. By offloading rendering to the cloud, the company leverages its existing content distribution strengths rather than fighting solely on headset hardware. This development positions ByteDance in direct competition with Google's Genie in the emerging world-model space. What the sources don't address: How ByteDance plans to price or distribute the massive cloud computing power required to sustain real-time, low-latency spatial generation at consumer scale.