Multimodal Digital Human Driving: Motionverse Engine and Metaverse Applications

Digital humans, also known as virtual humans, have progressed from simple CG models (L1) to fully autonomous, expressive agents (L5) that can understand intent and continuously learn. The article outlines the five maturity levels (L1‑L5) defined by the SenseTime research institute, highlighting the increasing reliance on AI, big data, and multimodal inputs.

The Motionverse engine, developed by Zhongke ShenZhi, consists of three core components: multimodal motion command collection, AI‑driven digital human modeling and rendering, and real‑time animation output. It supports language, text, sensor, video, controller, and script inputs, enabling flexible, low‑cost motion capture and AI‑based generation.

Large‑scale datasets (≈150 hours of motion video, tens of millions of frames) are used to train visual‑language models that drive realistic facial micro‑expressions across nine emotions and generate context‑aware body motions, even with sparse sensor setups.

Multimodal driving applications include sparse sensor‑based motion capture for live streaming, emotion‑driven facial micro‑expressions, and AI‑driven gesture generation for virtual customer service agents.

Four SDK integration layers are offered: data only, data + asset, data + asset + cloud rendering, and full workflow integration, enabling enterprises to embed digital human capabilities into animation, gaming, live broadcasting, and industry‑specific solutions.

In summary, the Motionverse platform provides enterprise‑grade SaaS tools for digital human creation, live streaming, and virtual客服, while open SDKs connect these capabilities to broader metaverse ecosystems, positioning Zhongke ShenZhi as a full‑stack solution provider for real‑time AI‑driven digital humans.