ByteDance Developing Real-Time Spatial Video AI Model Supervised by Zhang Yiming
ByteDance is preparing to launch a real-time spatial video generation AI model supervised directly by founder Zhang Yiming, with a release targeted as early as next month. Built on ByteDance's Seedance video model, the technology allows users to create interactive 3D virtual environments for live streaming, short dramas, and gaming. By shifting intensive 3D spatial rendering entirely to the cloud, the model significantly reduces hardware compute requirements and lowers the initial cost of virtual reality headsets like ByteDance's Pico. This initiative aims to connect ByteDance's AI models, cloud infrastructure, content platforms, and XR hardware into a unified growth flywheel. The model generates dynamic video streams on demand at 20 frames per second with an ultra-low latency of just 50 milliseconds (0.05 seconds) while responding to headset users' voice and physical movements. Similar in concept to Google's Genie, it shifts competitive focus from headset hardware specs to underlying AI capabilities and cloud infrastructure.
## BACKGROUND
Spatial computing and generative world models—such as Google's Genie—aim to dynamically generate interactive 3D virtual environments from prompts or user inputs in real time. ByteDance operates Pico, a virtual reality headset business, and previously launched Seedance, a high-quality AI video generation model.