ACE-Step Studio
GitHubGitHub stars Download

Full songs with vocals, on your own GPU

A desktop studio for ACE-Step 1.5, the open music model that writes and sings full songs.

Windows 10/11 x64 and an NVIDIA GPU with 4 GB of VRAM or more (GTX 900 or newer). The models are downloaded from inside the studio when you decide to.

Writing a song: caption, lyrics and metadata, the model switcher above the form.

Songs made with it

Rendered on a clean install of the released build, as the studio saved them.

Neon HarborEnglish synth-pop from a caption and lyrics.
Дорога домойRussian folk rock with accordion.
夜の電車Japanese city pop.
Fuego lentoLatin pop in Spanish.
Midnight DriveAn instrumental: synthwave.
Classic lo-fi hip-hop instrumentalOne line in the simple form: ACE-Step's planner wrote the caption, tempo and key.
More examples (2)
Neon harbor, call my nameCover: the first song rendered again as an acoustic ballad.
Night RiderWith a LoRA from the catalogue: Synthwave XL by ryanontheinside.

What it does

A LoRA catalogue

The LoRA catalogue: 131 LoRA with their authors, likes and downloads, marked for 2B or XL.

131 LoRA picked from Hugging Face by likes and downloads, described and credited, and a search of everything Hugging Face has — every file checked for the model size it fits before it downloads.

A song from one line

The simple form: one line in, a whole song out.

The simple form: an idea in, a finished song out.

Your own LoRA

Your own LoRA trained on the card, the loss as it learns, one click into the library.

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

Stems and MIDI

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

Six stems with HT-Demucs; any track to multi-instrument MIDI with MuScriptor, with a piano roll played against the original.

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 — 164 tools — and sees and works its window.

Screenshots

The studio itself, in your language.

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 caption and lyrics, an example, or one line in the simple form.
  4. 4CreateThe engine starts by itself and the song lands in your library.

Questions

What graphics card do I need?

Windows 10/11 x64 and an NVIDIA GPU with 4 GB of VRAM or more (GTX 900 or newer). The models are downloaded from inside the studio when you decide to.

A runnable set is the DiT, the planner, the text encoder and the VAE, all in GGUF.

Your GPUSetDownload
24 GBFull native — XL turbo BF16, planner 4B BF1618.5 GB
13 GB+Quality — XL turbo Q8_0, planner 4B Q8_010.1 GB
10 GB+Balanced — XL turbo Q6_K, planner 1.7B6.7 GB
8 GB+Light — XL turbo Q4_K_M, planner 1.7B5.7 GB
4 GB+Minimal — 2B turbo Q4_K_M, planner 0.6B3.1 GB

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, acestep.cpp and ACE-Step 1.5 are MIT. The Regrind fine-tune is CC BY-NC-SA 4.0; each catalogue LoRA carries its author's licence.

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 its LoKR recipe. Every setting is editable, and each checkpoint goes to the LoRA library in one click.

Can an AI agent use it?

Connect Claude Code, Claude Desktop or Cursor and the agent does everything the studio does — 164 tools — and sees and works its window. ACE-Step's writing rules come with it, so the agent writes the songs itself.

What is inside the download?

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

PartRunsSize
ACE-Step engine (acestep.cpp)Your GPU: CUDA, Vulkan or the processorbundled
ACE-Step model setYour GPU3.1–18.5 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-trainAn NVIDIA RTX card0.1 GB + the BF16 DiT, optional
Describing songs by ear — MOSS-Music-8Babout 12 GB of VRAM9.8 GB, optional
Audio to MIDI — MuScriptor (HOT-Step ace-midi)NVIDIA GTX 16 / RTX 20 or newer0.5–5.6 GB, optional
VST3 host — HOT-StepYour machine, a process of its ownbundled
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 ─┐
          ├─ ACE-Step-Studio.exe   (Tauri window + Rust/Axum service)
Rust axum ┘        │
                   └─ acestep.cpp `ace-server`   (C++/GGML, GGUF)

Who made this

Nerual Dreming — artist, founder of ArtGeneration.me and the Neuro-Cartel community, author of YuE2 Studio and MiniMax Music3 Studio, the same studio for other music models.