YuE2 Studio
GitHubGitHub stars Download

Full songs with an editable score, on your own GPU

A desktop studio for YuE2, the open song model that writes a score before it sings.

Download for Windowsv3.0.0 · 299 MB Star on GitHubGitHub stars

Windows 10/11 x64 and a GPU with 6 GB of VRAM: NVIDIA (GTX 900 or newer) on CUDA; AMD and Intel on Vulkan are experimental. The models are downloaded from inside the studio when you decide to.

Writing a song: style, lyrics, one of the 110 examples loaded.

Songs made with it

Rendered on a clean install with the recommended set, as the studio saved them. In English, Spanish, Mandarin, Japanese and Russian.

Houseplant PartyEnglish funk disco.
My Truck Talks BackCountry with pedal steel and fiddle.
La SiestaLatin pop in Spanish.
外卖小哥Mandarin city pop.
月曜日のサムライJapanese rock in the style of an anime opening.
Дача, лето, комарыFull song from a style and lyrics.
More examples (6)
Дача, лето, комары — джазCover: SheetSage2 wrote down the melody of the song above, YuE2 sang it as jazz.
ПонедельникBlues rock, raspy voice.
Бабушка на дискотекеRe-rendered from its saved audio codes with 64 solver steps.
Борщ на орбитеSka punk at 150 BPM.
Кот-программистSynthwave pop, the second take.
Ночной вор БарсикStyle and lyrics written by the built-in assistant from a one-line idea.

What it does

An editable score

The score as sheet music and ABC text: melody with chords, melody only, or none.

The composition comes back as ABC notation, engraved as sheet music.

Covers

Cover mode: pick a recording and SheetSage2 writes its melody down.

SheetSage2 writes a recording’s melody down as a score, and YuE2 sings it in your style.

Any track to MIDI

Any track to MIDI: a piano roll of every instrument, played against the original, with mute and solo.

A song, a stem or a processed take becomes multi-instrument MIDI - 34 instrument groups and drums - with MuScriptor on your GPU.

Your own LoRA

Your own LoRA trained on the card: loss and drift as it learns, a checkpoint every 50 steps, one click into the library.

Songs of one artist or style become a LoRA on your card, with HOT-Step’s trainer and recipe.

A player of its own

The whole window as Winamp 2: equalizer, playlist and MilkDrop, skinned.

A ten-band equalizer on Winamp's frequencies with its presets and .EQF files; MilkDrop with hundreds of presets or a spectrum in ten looks; and a Winamp mode whose windows move apart, dock and resize, in the original skin or any from the Winamp Skin Museum.

Driven by your agent (MCP)

Connect an agent over MCP: the address and the lines to paste into Claude Code or any other client.

Connect Claude Code, Claude Desktop or Cursor and the agent does everything the studio does: writes and makes songs, fills a dataset and trains, draws covers, cuts clips in the video editor, plays songs - and sees the window and presses its buttons.

Screenshots

The studio itself, in your language.

A finished track with its lyrics and the score it was sung from, ready to reuse.
A finished track with its lyrics and the score it was sung from, ready to reuse.
Model sets: one quantisation per role, what is on disk, switching in one click.
Model sets: one quantisation per role, what is on disk, switching in one click.
The LoRA catalogue: styles, artists and sound, each credited to its author, downloaded in one pass.
The LoRA catalogue: styles, artists and sound, each credited to its author, downloaded in one pass.
A dataset of 61 songs, each with its lyrics and style, found and written by the studio.
A dataset of 61 songs, each with its lyrics and style, found and written by the studio.
Audio processing: noise reduction, the Spectral Lifter, a vocal naturaliser, VST3 plugins and mastering to a reference.
Audio processing: noise reduction, the Spectral Lifter, a vocal naturaliser, VST3 plugins and mastering to a reference.
Tracks made by tools stay in the library, each linked to the one it was made from.
Tracks made by tools stay in the library, each linked to the one it was made from.
The equalizer with its curve and MilkDrop over the studio, as the song plays.
The equalizer with its curve and MilkDrop over the studio, as the song plays.

Getting started

  1. 1InstallRun the installer, or unpack the portable archive anywhere.
  2. 2Choose a setThe first screen preselects what your card can run; it downloads only what is missing.
  3. 3WriteA style and lyrics, or load an example; an instrumental is one switch.
  4. 4CreateThe engine starts by itself and the song lands in your library with its score.
Download for Windowsv3.0.0 · 299 MB

Questions

What graphics card do I need?

Windows 10/11 x64 and a GPU with 6 GB of VRAM: NVIDIA (GTX 900 or newer) on CUDA; AMD and Intel on Vulkan are experimental. The models are downloaded from inside the studio when you decide to.

A runnable set is the YuE2-3B backbone and the VAE; SheetSage2 is optional and only needed for covers.

Your GPUSetDownload
12 GB+Full native — BF16, original weights9.7 GB
8 GB+Quality — Q8_0, near lossless4.9 GB
7 GB+Balanced — Q6_K4.0 GB
5.5 GB+Light — Q5_K_M3.6 GB

Sizes include SheetSage2. The studio preselects the set your card can run, but the download is always your decision. Every file is checked by SHA-256 against a pinned Hugging Face revision, and an interrupted download resumes.

Is it free?

Yes. The studio is free and open source, and the models download for free from inside it.

Does it work offline? Is my music sent anywhere?

Your music is yours, and it stays on your disk.

  • No account — There is nothing to sign up for.
  • No telemetry — The studio does not report what you generate.
  • Cloud is opt-in — Nothing leaves the machine until you add a key, and then only for that capability.
Can I use the songs I make?

The studio and yue2.cpp are MIT. The YuE2-3B and YuE2 VAE weights are CC BY-NC 4.0 with an individual-creator permission: musicians and creators working on their own may sell and monetize the songs they make; companies need a license from the authors. SheetSage2, used for covers, is CC BY-NC 4.0.

Can I train it on my own style?

Songs of one artist or style become a LoRA on your card, with HOT-Step’s trainer and recipe. Every setting is editable, and datasets move between this studio and MiniMax Music3 Studio.

Can an AI agent use it?

Connect Claude Code, Claude Desktop or Cursor and the agent does everything the studio does: writes and makes songs, fills a dataset and trains, draws covers, cuts clips in the video editor, plays songs - and sees the window and presses its buttons. The model’s writing rules come with it, so the agent writes the styles and lyrics itself. Pick it as the writing assistant and it writes instead of the local model.

What is inside the download?

Local by default. A cloud key is added by you, for the capability you choose.

PartRunsSize
YuE2 engine (yue2.cpp)Your GPU: CUDA, Vulkan or the processorbundled
YuE2 model setYour GPU3.6–9.7 GB
NVIDIA cuBLASNVIDIA cards only391 MB, once
Karaoke timings — Parakeet or WhisperYour machineoptional
Stem separation — HT-DemucsYour GPU or CPUoptional
Writing assistantLocal Gemma or OpenRouteroptional
LoRA trainer — HOT-Step ace-trainNVIDIA RTX 30 or newer, 11 GB of VRAM8 GB, optional
Describing songs by ear — MOSS-Music-8Babout 12 GB of VRAM10 GB, optional
VST3 host — HOT-StepYour machine, a process of its ownbundled
Audio to MIDI — MuScriptor (HOT-Step ace-midi)NVIDIA GTX 16 / RTX 20 or newer0.5–5.6 GB, optional
How is it built?

The service is compiled into the desktop binary and supervises the C++ engine. Finished songs are imported by the service itself, so a result survives a reload or a closed window.

React UI ─┐
          ├─ YuE2-Studio.exe   (Tauri window + Rust/Axum service)
Rust axum ┘        │
                   └─ yue2.cpp `yue-server`   (C++/CUDA, GGUF)

Who made this

Nerual Dreming — artist, founder of ArtGeneration.me and the Neuro-Cartel community, author of ACE-Step Studio and MiniMax Music3 Studio, which this studio grew out of.

GitHub · @timoncool Telegram · @nerual_dreming Telegram · @neuroport ArtGeneration.me Support the project