YuE2 Studio Lets Windows Users Edit and Render Full AI Songs Locally
A Rust and Tauri desktop app wraps the YuE2 song model with an editable score, covers, LoRA training, and a local MCP server.
Takeaways
- YuE2 Studio wraps the open YuE2 song model in a native Windows app with no Python.
- Built in Rust and React via Tauri, calling the yue2.cpp C++/CUDA engine locally.
- Writes editable ABC sheet music first, then renders full vocal songs up to ten minutes.
- Ships four GGUF quantisations from BF16 down to Q5_K_M for 6 GB VRAM GPUs.
- Includes LoRA training, covers via SheetSage2, stems, MIDI, karaoke and audio processing.
- Exposes a local MCP server with 162 tools for Claude Code and other agents.
YuE2 Studio is a new open-source desktop app that packages the YuE2 song-generation model in a native Windows installer. Core inference runs on a local GPU. YuE2 first creates a readable score, then synthesizes the vocals and instrumentation, allowing users to edit melody, harmony, tempo, key, and song structure before rendering again.
The installer removes the need to maintain a Python environment and includes automatic updates. Its desktop shell combines Rust, React, and Tauri, while a C++ implementation called yue2.cpp handles model inference through a loopback HTTP service on the same machine.
A score before the sound
YuE2 is a 3.58-billion-parameter open-source music model developed by the Multimodal Art Projection community with HKUST and HKGAI. Its AR-NAR Mixture-of-Transformers architecture combines autoregressive and non-autoregressive generation stages, predicting some elements sequentially and others in parallel through specialized transformer components.
ABC notation provides the model’s symbolic plan in a compact, text-based format that represents melody, harmony, meter, and other musical details. YuE2 converts that score into semantic music tokens, an intermediate representation of the composition, and then renders full-song 48 kHz stereo audio with vocals from tagged lyrics and a style description. Changed notes become part of the next render, while untouched sections retain their planned structure.
The upstream repository exposes this pipeline through command-line tools and Python packages. YuE2 Studio turns each stage into a graphical workflow for composing, editing, rendering, and exporting a track.