sola1039/HAST-SEC

What it solves

Creating realistic, controllable virtual characters is hard. Most systems need expensive hardware or complex training. This project offers a local, non-commercial research demo that lets you talk to a 3D character in Chinese and give it simple movement commands — like raising an arm or lifting a leg — all running on your own computer without external APIs.

How it works

The system combines a small text model (about 0.8B parameters) with a lightweight motion model. You type a message or a motion instruction in the chat box. The text model understands your intent, then the motion model generates continuous character movement that plays in a web-based viewer. The viewer also shows real-time graphs for the last 15 seconds, including pose metrics and resource usage. A local service handles all inference, so no internet connection is needed after installation.

Who it’s for

This is aimed at researchers, students, and hobbyists who want to experiment with language-driven character animation. It is explicitly a non-commercial learning and research release — not a finished product. Users are expected to modify code, train further models, and adapt the system to their own needs. It’s also a demonstration of the company’s broader research direction.

Highlights

• Chinese chat interaction with a virtual character.
• Semantic motion control: you can say simple instructions like "wave your left hand" or "raise your right leg".
• Runs locally on CPU or NVIDIA GPU; no external API keys required.
• Import your own legally obtained VRM characters (full skeleton, under 100MB).
• Live 15-second motion curves, pose metrics, and resource usage display.
• Open source under a non-commercial license for learning and teaching.