Practical overview
When it fits
Use YuE2 when a music-generation workflow benefits from symbolic control over melody and chords, or when adapting a transcribed song to new lyrics or a different style.
Who it is for
- Musicians and composers who want to generate songs from lyrics, style prompts or symbolic scores.
- Developers building music-generation or music-editing workflows around a locally run Python model.
- People who want to try song generation through the linked browser demo without installing the model.
Documented uses
- Generate a complete song with vocals and accompaniment from lyrics and a style prompt.
- Inspect or revise a generated melody-and-chord plan before audio rendering.
- Create a cover by transcribing a source recording, then applying new lyrics or a target style.
- Supply an ABC score as input to guide generation.
Requirements & limits
The repository links a hosted browser demo described as free and requiring no installation. Local use is documented for Linux with Python 3.12 and an NVIDIA GPU with BF16 support and 24 GB VRAM.
- The local workflow has explicit Linux, Python 3.12 and GPU requirements.
- Model files download from Hugging Face on first use.
- Generation can use full symbolic planning, melody-only planning, direct lyric-and-style generation, or a supplied ABC score.
- The repository preserves the original YuE-v1 code, documentation and license on a separate branch.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up YuE2 locally on Linux
Install the documented local Python package and generate a first song.
Before you start
- Linux
- Python 3.12
- NVIDIA GPU with BF16 support and 24 GB VRAM
- Internet access for model files downloaded from Hugging Face on first use
Steps
Clone the YuE repository and change into its directory.
Create and activate a Python 3.12 virtual environment.
Upgrade pip and install the package from the repository.
Run the provided examples/generate.py command with an output directory.
Expected outcome
The documented output includes outputs/first-song/audio.flac; the directory also retains score and generation artifacts.
Guide limitations
- The documented local quick start specifies Linux and the stated GPU requirements.
- Model files are downloaded from Hugging Face on first use.
Prepare a YuE2 song cover
Use a transcribed melody score, new lyrics or style, and YuE2's melody-planning mode to prepare a cover.
Before you start
- A source recording
- SheetSage2 for transcription
- New lyrics or a target style
- YuE2 local Python setup
Steps
Transcribe the source recording with SheetSage2.
Review the resulting melody ABC score.
Provide the score and new lyrics or target style to YuE2.
Use cot="melody" with a score that has no chord symbols.
Expected outcome
YuE2 generates a cover with accompaniment that can adapt to the target style.
Guide limitations
- The cover workflow requires a transcribed and reviewed melody score.
- The guidance specifies melody mode and a score without chord symbols.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 13 October 2026
- Original website title
- multimodal-art-projection/YuE · Official homepage title · retained in its source language
Related options
YuE-v1 — The repository points readers seeking the original YuE to a separate branch preserving its code, documentation and license.