# Shao Music

> Shao - High-fidelity music generation by unified acoustic-token pipeline. Open-source complete music works.

- Skill: `belentani7/shao-music` (Agent Skill)
- Install (CLI): `npx skillmds@latest add belentani7/shao-music`
- Raw SKILL.md: https://api.skillmd.com/api/skills/belentani7/shao-music/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: DevOps & Infra
- Author: belentani7 (https://skillmd.com/u/belentani7)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/belentani7/shao-music

---


# Shao - High-Fidelity Music Generation

Open-source system for high-fidelity music generation using unified acoustic-token pipeline. Generates complete musical works from text descriptions and lyrics.

## When to Use
- User needs the highest quality open-source music generation
- User wants complete musical works (not just loops)
- User wants text + lyric controlled music
- User has NVIDIA GPU with 24GB+ VRAM

## Key Features
- Unified acoustic-token representation
- 64-layer RVQ acoustic token hierarchy
- Two-stage generation pipeline
- Complete musical works (not short clips)
- Text and lyric control
- Coarse-to-fine acoustic detail generation

## Requirements
- NVIDIA GPU with 24GB+ VRAM (RTX 4090 or higher)
- Docker and NVIDIA Container Toolkit
- CUDA-compatible NVIDIA driver
- Python and Node.js

## Installation

```bash
git clone https://github.com/Shao-Music-AI/Shao.git
cd Shao
# Follow ENVIRONMENT_SETUP.md for Docker setup
```

## Quick Start

```bash
# Docker-based inference
docker build -t shao .
docker run --gpus all -it shao
```

## Architecture

| Stage | Description |
|-------|-------------|
| Backbone | Generates coarse acoustic tokens |
| Super-Resolution | Completes higher RVQ token layers |
| Decoder | Reconstructs audio waveform |

## Online Demo
- https://shao-music-ai.github.io/Shao-demo/

## Model Weights
- HuggingFace: liujiafeng/Shao-MusicGeneration-v1.0
- arXiv: https://arxiv.org/abs/2605.01790

## Related
- `song-generation`: Tencent SongGeneration/LeVo
- `heartmula`: HeartMuLa music models
- `rvc`: Voice conversion and singing