jtydhr88/ComfyUI-HY-Motion1

What it solves

This project provides a ComfyUI plugin that enables the generation of 3D human motion from text descriptions. It simplifies the process of creating 3D animations by automating the motion generation, prompt optimization, and exporting to industry-standard formats like GLB and FBX.

How it works

The plugin integrates the HY-Motion 1.0 model for text-to-motion generation. It uses a pipeline consisting of a text encoder (based on Qwen3-8B LLM) and a motion diffusion network. To improve results, it includes a "Prompt Rewrite" feature that uses an LLM to optimize user input and estimate the required motion duration. The generated motion is then processed through various nodes for previewing (2D skeleton or interactive 3D viewer) and exporting.

Who it’s for

3D animators, game developers, and AI researchers who use ComfyUI and want to generate human motion sequences without manual keyframing.

Highlights

• Text-to-Motion: Generates 3D human animations directly from text prompts.
• LLM-Powered Prompting: Automatically rewrites prompts and estimates motion duration for better quality.
• Flexible Export: Supports GLB (via frontend) and FBX (with custom Mixamo character retargeting).
• Memory Optimization: Supports GGUF, AWQ, and BitsAndBytes quantization to run on lower VRAM GPUs, with options to offload LLMs to the CPU.
• Interactive Preview: Includes a Three.js-based 3D viewer with playback controls and skeleton/mesh toggles.