Repolode
Submit
timoncool

timoncool/YuE2-Studio

Windows desktop studio for the open YuE2 song model: generates full songs with vocals from a style and lyrics, with an editable ABC score, on a local NVIDIA GPU.

Open-source alternative to:Suno

View on GitHubWebsite
Stars
920
Forks
113
Open issues
27
Last commit
2 hours ago

What is it?

YuE2 Studio is a desktop front end for the YuE2 song model. You describe a style and write lyrics, and the model first composes a melody with chords, shown as sheet music in ABC notation, then performs it as a full song with vocals. The score can be read, edited and re-rendered with a different sound without changing the composition. The studio runs offline on an NVIDIA card with at least 6 GB of VRAM; AMD and Intel cards go through Vulkan and are marked experimental.

The engine is yue2.cpp, a native C++/CUDA port of YuE2, and the application itself is built with Rust and React inside a Tauri window, so the runtime path does not use Python. It ships as a Windows installer with auto-update or a portable folder, and an MCP server at a local address lets an AI agent drive the studio.

Beyond generation, the project includes covers that transcribe a recording into a score, stem separation into six parts, audio-to-MIDI conversion, word-level karaoke timing, LoRA loading and training, and music video export.

Key features

  • 01Generates full songs with vocals from a style description and lyrics, up to ten minutes long.
  • 02Writes the melody as an editable ABC score and engraves it as sheet music.
  • 03Re-renders an existing composition with different sound settings without composing again.
  • 04Covers mode transcribes a recording's melody and sings it in a new style.
  • 05Separates a track into six stems with HT-Demucs on the GPU.
  • 06Runs offline on NVIDIA GPUs with CUDA; a Windows installer with auto-update is provided.
  • 07Exposes the studio to AI agents through a local MCP server.

Use cases

  • Sketching song ideas with a melody and chords before arranging them by hand.
  • Creating a cover of an existing recording in a different musical style.
  • Generating demo vocal tracks locally without sending lyrics or audio to a cloud service.
  • Converting an audio track into multi-instrument MIDI for further editing in a DAW.

Topics

#ai-music#music-generation#text-to-music#local-ai#sheet-music#windows#cuda#tauri#abc-notation#covers

FAQ

What is YuE2-Studio?
Windows desktop studio for the open YuE2 song model: generates full songs with vocals from a style and lyrics, with an editable ABC score, on a local NVIDIA GPU. YuE2 Studio is a desktop front end for the YuE2 song model. You describe a style and write lyrics, and the model first composes a melody with chords, shown as sheet music in ABC notation, then performs it as a full song with vocals. The score can be read, edited and re-rendered with a different sound without changing the composition. The studio runs offline on an NVIDIA card with at least 6 GB of VRAM; AMD and Intel cards go through Vulkan and are marked experimental.
Is YuE2-Studio free to use? What license does it have?
YuE2-Studio is open source under the MIT license. Check the license terms before using it in commercial products.
Is YuE2-Studio still maintained?
Actively maintained. The last commit to YuE2-Studio was on Oct 19, 2026.
What can I use YuE2-Studio for?
Sketching song ideas with a melody and chords before arranging them by hand. Creating a cover of an existing recording in a different musical style. Generating demo vocal tracks locally without sending lyrics or audio to a cloud service. Converting an audio track into multi-instrument MIDI for further editing in a DAW.
Where can I get YuE2-Studio?
The source code and installation instructions are on GitHub: https://github.com/timoncool/YuE2-Studio

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects