While large language models were still "typing" in two-dimensional chat windows, Mofa Technology directly generated its skeleton, muscles, and vocal cords at once. On May 30th, the company launched the world's first 3D digital human open platform for developers, "Mofa Star Cloud," enabling AI to evolve from "speaking" to "making expressions, gestures, and moving the body."

The platform's core is a 3D multimodal real-time generation engine: input any text, and it outputs synchronized speech, micro-expressions, and body movements in milliseconds. It is adapted for mobile phones, tablets, in-car systems, and TVs, maintaining a refresh rate of over 30fps and latency below 100 milliseconds. Developers do not need to pre-record animations or bind expensive GPUs; a low-cost entry-level chip can drive it locally, with the cloud only responsible for model inference and updates.

image.png

Li Yu, CTO of Mofa Technology, introduced that the secret lies in the "cloud-end split architecture" — separating heavy computing geometric modeling from lightweight rendering instructions. The terminal only receives compressed action bitstream, "like watching a video online, but with only 1/10 the traffic of regular videos."

The application scenarios have gone beyond the "demo" stage: A smart hotel in Shanghai Center integrated the Star Cloud SDK into the front desk tablet, allowing the digital front desk to answer check-in, invoice issuance, and nearby catering questions in real time. A provincial government hall's guidance terminal generated a sign language version of a digital human through this platform, providing barrier-free services for hearing-impaired people. In the recruitment market, Star Cloud collaborated with a leading HR SaaS company to launch an AI interviewer. After candidates provide voice answers, the digital interviewer will nod, record, and ask follow-up questions based on semantics, replicating the rhythm of real interaction.

The platform uses tiered billing: basic models are free to call, while advanced emotion control and industry knowledge packages are charged by usage. Enterprises can purchase private deployment outright. On the day of launch, more than 2,000 developers applied for the beta test, covering four fields: education, healthcare, retail, and IoT.

Chai Jinxiang, CEO of Mofa Technology, said that the next step will be to open source some motion-driven interfaces and collaborate with chip manufacturers to launch the "Star Cloud Ready" certification. "The goal is to run on 1 billion mid-range devices within one year, making embodied intelligence the default way of interaction, rather than just a display in high-end exhibitions."