Hunyuan Motion 1.0 — Text-to-3D Human Motion Generation at Scale
Hunyuan Motion brings natural language to life with high-fidelity, skeleton-based 3D animation
Hunyuan Motion is a next-generation text-to-3D human motion generation model suite that transforms plain English prompts into fluid, realistic, skeleton-based animation. Powered by a billion-parameter Diffusion Transformer trained with Flow Matching, Hunyuan Motion scales instruction following, motion detail, and temporal smoothness to a new level. The project provides ready-to-use CLI and an interactive Gradio app for fast iterations, reproducible workflows, and cross-platform compatibility. With large-scale pre-training, high-quality fine-tuning, and reinforcement learning from human feedback, Hunyuan Motion achieves state-of-the-art quality and reliability for games, film, VR/AR, virtual production, and digital human pipelines. Explore the official open-source implementation, pretrained weights, and quickstart tools on github.com.
Text-to-Motion
3D Animation
What is Hunyuan Motion?
Hunyuan Motion is a series of text-to-3D human motion generation models that produce skeleton-based character animations directly from natural language prompts. Built on a Diffusion Transformer and trained with Flow Matching, Hunyuan Motion scales to the billion-parameter level to significantly improve instruction following and motion quality compared with previous open-source systems. The training pipeline combines three phases—massive pre-training on diverse human motion, fine-tuning on curated high-quality sequences, and reinforcement learning with human feedback—to deliver smooth, physically plausible motion that aligns with detailed textual directions. The result is a prompt-to-animation workflow that integrates seamlessly into modern 3D pipelines. The project ships with a standard model (≈1.0B parameters) and a Lite variant (≈0.46B parameters) for different performance envelopes, together with cross-platform support, batch-friendly CLI, and an easy Gradio UI for interactive exploration. Full details, updates, and usage instructions are available on the official repository on [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
Diffusion Transformer with Flow Matching scaled to ~1B parameters for superior instruction following and motion quality [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0)
Three-stage training: large-scale pre-training, high-quality fine-tuning, and RLHF for natural, coherent motion [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0)
Ready-to-use CLI and Gradio app for fast local inference and interactive visualization [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0)
Flow Matching
Key features
What makes Hunyuan Motion different
Diffusion Transformer + Flow Matching at billion scale
Hunyuan Motion scales a DiT backbone with Flow Matching to ~1.0B parameters, boosting instruction following and motion quality while preserving stability across frames. See the official model description and training overview on [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
Three-stage training pipeline
Large-scale pre-training on 3,000+ hours of motion data builds broad priors; 400 hours of curated fine-tuning enhances detail and smoothness; RL from human feedback refines naturalness and prompt adherence. Details are documented on [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
Model zoo: Standard and Lite
Choose HY-Motion-1.0 (~1.0B parameters) for state-of-the-art motion fidelity or HY-Motion-1.0-Lite (~0.46B) for lighter deployments. The repository provides weights and instructions for quick setup [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
Text-first motion generation
Hunyuan Motion takes plain-language prompts and outputs motion that aligns with the intended action, style, and pacing, enabling creative control without hand-animating every pose.
Prompt engineering and duration estimation
Optionally connect an LLM-based duration estimator and prompt rewriter module to improve pacing and clarity. Hunyuan Motion exposes simple flags to enable or disable these helpers as needed [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
CLI and Gradio included
Run batch jobs via the CLI for large prompt sets or spin up a local Gradio server for interactive visualization. These tools are maintained in the official repo with clear instructions and examples on [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
Cross-platform support
Hunyuan Motion runs on macOS, Windows, and Linux, helping mixed teams share workflows. The repo includes requirements and scripts for consistent setup and inference [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
Production-friendly outputs
By generating skeleton-based motion, Hunyuan Motion fits into rigged character pipelines and supports downstream retargeting, blending, and clean-up using familiar DCC toolchains.
How Hunyuan Motion works
From prompt to skeleton animation
1
Instruction parsing and duration planning
A user writes a natural-language prompt describing the desired action, style, and pacing. Hunyuan Motion can optionally use a prompt rewrite and duration estimation module to clarify intent, infer timing, and standardize phrasing prior to generation. This step increases alignment between the text description and the motion outcome, especially on complex or multi-step actions as documented in the official repository on [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
2
Flow-Matched DiT sampling and skeleton synthesis
Hunyuan Motion samples motion trajectories with a Diffusion Transformer trained via Flow Matching. The model outputs smooth, temporally coherent, skeleton-based 3D motion that adheres to the prompt. The result can be viewed interactively in the Gradio app or saved via the CLI for integration into your 3D pipeline. Full usage instructions are provided on [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
Where to use Hunyuan Motion
From previsualization to final animation
Game development and prototyping
Use Hunyuan Motion to rapidly generate run, jump, attack, emote, and traversal motions to prototype gameplay and iterate on character feel. Designers can test prompt variations and quickly export skeleton-based motion for retargeting to in-game rigs. For larger libraries, the CLI supports batch inference and consistent output naming. The official repo shows the recommended workflow and flags on [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
Film, VFX, and virtual production
In cinematic pipelines, Hunyuan Motion accelerates previz and blocking. Directors can sketch intent as text, let Hunyuan Motion generate baseline motion, and hand it to animators for refinement. The smooth, instruction-following behavior helps keep revisions tight and predictable across shots.
Digital humans and virtual influencers
Pair Hunyuan Motion with digital humans to synthesize expressive gestures, idle loops, and stylized acts. Because Hunyuan Motion is prompt-based, non-technical creators can explore motion ideas faster and collaborate with technical directors for polish and delivery.
VR/AR and immersive experiences
Hunyuan Motion supports the rapid creation of ambient crowd motion, guided interactions, and narrative beats that enhance immersion. Skeleton-based outputs make it straightforward to retarget animations to headset-optimized rigs and runtime constraints.
Education, research, and labs
Educators and researchers can use Hunyuan Motion as a reference for Diffusion Transformer and Flow Matching approaches to motion. The project’s open-source code and model zoo enable reproducible experiments and instruction-following benchmarks [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
Studios building internal motion libraries
When building a consistent library of house-style motions, Hunyuan Motion provides a coherent base. Teams can specify tone, pacing, and style in text, generate consistent takes, and standardize review via the Gradio app before ingesting into asset management.
Frequently asked questions
Everything you need to know to start generating with Hunyuan Motion
What exactly does Hunyuan Motion generate?
Hunyuan Motion generates skeleton-based 3D human motion from text prompts, designed to integrate into common 3D animation pipelines. The model focuses on realism, temporal smoothness, and instruction following so that actions, styles, and pacing reflect what you describe. See the project overview and examples on [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
How is Hunyuan Motion trained?
Hunyuan Motion uses a three-stage process: large-scale pre-training on over 3,000 hours of motion to learn broad priors, fine-tuning on 400 hours of curated high-quality data for detail and smoothness, and reinforcement learning with human feedback to further refine instruction alignment and naturalness. The technical summary and training notes are in the official repo on [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
What is the difference between HY-Motion-1.0 and HY-Motion-1.0-Lite?
Both are part of Hunyuan Motion. HY-Motion-1.0 is the standard, ~1.0B-parameter model that maximizes instruction following and motion fidelity, while HY-Motion-1.0-Lite (~0.46B) is optimized for lighter deployments and faster turnaround. Choose based on your GPU budget and motion quality needs. Model download guidance is provided on [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
How do I run Hunyuan Motion locally?
Install PyTorch and the project requirements, download the pretrained weights following the repo’s instructions, then choose your preferred interface: use the CLI for batch prompts or launch the Gradio app for an interactive UI. A step-by-step quickstart is detailed on [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
Can Hunyuan Motion estimate duration and rewrite prompts automatically?
Yes. Hunyuan Motion can optionally use a duration estimator and prompt rewrite module to improve pacing and clarity. You can enable or disable these modules via simple flags. If they’re not available, you can explicitly turn them off to avoid connection errors, as described in the repository’s usage notes on [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
Which operating systems does Hunyuan Motion support?
Hunyuan Motion supports macOS, Windows, and Linux for inference. Installation instructions, dependency lists, and tooling for both CLI and Gradio are available on [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
Is there an interactive demo?
Yes. The project includes a Gradio app you can run locally for interactive visualization. The official repo also points to a Hugging Face Space and an official site for trying the model. Find links and setup on [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
How do I integrate Hunyuan Motion outputs into my pipeline?
Because Hunyuan Motion produces skeleton-based motion, it can be retargeted to your rigs, blended with existing clips, and passed through your DCC tools for polishing. The CLI is suitable for batch jobs, while the Gradio app supports creative exploration and review before import. Learn more on [github.com](https://github.com/Tencent-Hunyuan/HY-Motion-1.0).
Start creating with Hunyuan Motion
Turn prompts into production-ready skeleton animation today. Follow the quickstart in the official repository, launch the Gradio app for instant visualization, and iterate fast with the CLI for batch motion generation. Everything you need to deploy Hunyuan Motion locally—including dependencies, checkpoints, and usage notes—is available on github.com.
Related Models
Explore more AI models from the same provider
Hunyuan 3D
Transform your ideas and images into stunning, production-ready 3D assets with Tencent's revolutionary Hunyuan 3D. Featuring advanced diffusion models, professional texture synthesis, and seamless workflow integration for game development, product design, and digital art.
Learn More
Hunyuan Image
Hunyuan Image 3.0 transforms your ideas into stunning, photorealistic images with unprecedented prompt adherence and intelligent reasoning. Powered by 80B parameters and 64 experts MoE architecture, it delivers exceptional semantic accuracy and visual excellence. Experience the future of AI image generation with native multimodal understanding.
Learn More
Hunyuan Video Generator
Hunyuan Video transforms your text descriptions into stunning, high-quality videos with exceptional physical accuracy and temporal consistency. Powered by a 13B parameter Unified Diffusion Transformer architecture, it generates up to 5-second videos at 720p resolution with superior motion dynamics and visual fidelity. Experience the future of video creation with advanced Flow Matching schedulers and parallel inference capabilities.
Learn More
Hunyuan World
Transform text & images into high-quality 3D models. Unleash your creative potential.
Learn More
Hunyuan Video Avatar
Bring portraits to life. Create expressive talking-head videos from a single image and audio.
Learn More
Hunyuan Custom
Hunyuan Custom is Tencent's state-of-the-art multimodal video generation solution that allows users to create customized, subject-consistent videos using AI. Upload an image, type a prompt, or add audio/video input to generate cinematic-quality content in seconds.
Learn More