ByteDance Builds Real-Time Virtual Worlds with Zhang Yiming Leading AI Push

According to a Bloomberg report, ByteDance is preparing an AI model for real-time spatial video generation targeting robotics and autonomous systems, with founder Zhang Yiming personally overseeing development and a potential launch as early as next month.

The new model builds on Seedance, ByteDance's existing cinematic video generation model, but extends beyond passive video clips to enable users to create interactive virtual worlds for live streaming, short dramas, and games — shifting from "watching video" to "entering a world."

Zhang Yiming is coordinating internal departments to concentrate AI resources and compute on this project, signaling its strategic priority. ByteDance's main competitors are Meta, which has published research on world models, and Google, which is betting on spatial intelligence. By leveraging Seedance as a foundation, ByteDance aims to push video generation toward 3D interaction.

Technical details remain scarce, but the direction is clear: real-time, spatial, and interactive. Applications span ByteDance's core domains: live streaming hosts could use AI-generated virtual spaces instead of green screens; short drama characters could interact with viewers; game worlds could be generated on the fly rather than pre-designed. For robotics and autonomous systems, a spatial video generation model provides a piece of the puzzle for real-time spatial understanding and reaction.

Zhang Yiming previously stated he wants to "operate AI like operating a company"; this move shows he intends to bring AI into the physical world. Beyond efficiency-focused agents, a new competitive battlefield is opening around real-time spatial intelligence.