What it does
YuE2 Studio is a free, open-source Windows desktop app for the open YuE2 song model. You give it a style and lyrics, and it renders a full song with vocals on your own GPU, along with the sheet-music score the song was sung from. It also makes covers, converts tracks to MIDI, trains LoRAs, and can be driven by an AI agent over MCP. It is an independent project by Nerual Dreming (GitHub @timoncool), not by the YuE2 model authors.
Official source 1 ↗Features
- Full songs with an editable score. Write a style and lyrics, or load an example, and the song lands in your library together with its composition as ABC notation, engraved as sheet music. Instrumentals are one switch away. Official source 1 ↗
- Covers in a new style. SheetSage2 transcribes the melody of an existing recording into a score, and YuE2 sings it again in the style you choose. Official source 1 ↗
- Any track to MIDI. Turn a song, a stem or a processed take into multi-instrument MIDI (34 instrument groups plus drums) with MuScriptor on your GPU, then edit tracks, notes, chords and section markers in the built-in MIDI editor. Official source 1 ↗
- Train your own LoRA. Train a LoRA on songs of one artist or style on your own card using HOT-Step’s trainer, with every setting editable. The studio also includes a LoRA catalogue of styles, artists and sound, each credited to its author. Official source 1 ↗
- Audio processing tools. Clean up and finish tracks with noise reduction, the Spectral Lifter, a vocal naturaliser, VST3 plugins and mastering to a reference. Optional stem separation uses HT-Demucs. Official source 1 ↗
- Agent control over MCP. Connect Claude Code, Claude Desktop or Cursor and the agent can do what the studio does: write and make songs, build a dataset and train, draw covers, cut clips in the video editor and play songs. Official source 1 ↗
- Model sets matched to your GPU. On first run the studio preselects a model set your card can run, from full BF16 weights (12 GB+ VRAM, 9.7 GB download) down to a Q5_K_M light set (5.5 GB+ VRAM, 3.6 GB). Each file is checked by SHA-256 against a pinned Hugging Face revision, and interrupted downloads resume. Official source 1 ↗
Pricing & free limits
The full studio and model downloads at no cost. App and engine are MIT; YuE2 weights are CC BY-NC 4.0 with an individual-creator permission to monetise songs. Companies need a license from the model authors. Optional cloud features use your own API key. Official source 1 ↗
Billing details
Billing details were not established by the checked official sources.
Free access & limits
Free access and its limits were not established by the checked official sources.
Best for & limitations
Making a full song with vocals from your own lyrics, in languages such as English, Spanish, Mandarin, Japanese or Russian, without sending the music to a cloud service. Official source 1 ↗
Reimagining an existing song in another genre, such as turning a song into a jazz cover from its transcribed melody. Official source 1 ↗
Letting an AI coding agent such as Claude Code write lyrics and styles and produce songs through the studio. Official source 1 ↗
- Windows and a capable GPU only. Runs only on Windows 10/11 x64 with a GPU of at least 6 GB VRAM; support for AMD and Intel cards is experimental. Official source 1 ↗
- Non-commercial model license for companies. The YuE2 and SheetSage2 weights are CC BY-NC 4.0. Individual creators may monetise their songs, but companies need a license from the model authors. Official source 1 ↗
- Optional tools need more hardware. LoRA training needs an NVIDIA RTX 30-series or newer card with 11 GB of VRAM, and describing songs by ear needs about 12 GB of VRAM. Official source 1 ↗
Submit a tool
