Majik's Studio
v3.6.3

Download Majik's Music Studio

Free, open-source AI music generation. Apache 2.0 licensed.

Generate, arrange, split a song into stems and mix it — on your Mac or on GPU servers you point it at. You control where it runs.

New in v3.6.3
  • Fix — Lego mode on the Create tab actually works. Selecting Lego showed no Source Audio picker and no “what stem” field, so Generate died with `Missing instruction (what stem to generate...)`. Lego now has Source Audio (like Continue) plus a required stem name (bass, sax lead, …). Style is still the sound; it is not a substitute for the stem name.
  • Extract, Repaint, and Auto-Complete were the same missing-panel class. Each now has its Source Audio picker. Extract asks which stems to isolate. Repaint asks which time range to regenerate. Auto-Complete takes the partial mix. The app now refuses Generate in-panel instead of sending an empty RPC.
Recommended

macOS

Native Swift app

Native macOS app optimized for Apple Silicon. Signed and notarized — double-click to launch.

  • macOS 14+ (Sonoma)
  • Apple Silicon or Intel
  • 16 GB RAM minimum
Download .dmg

v3.6.3 · Universal Binary

Linux

Beta

Tauri 2

Built with Tauri 2 for native performance on Linux desktops. Beta — .deb and AppImage available now.

  • Ubuntu 24.04+ / Debian 12+ (glibc 2.36)
  • x86_64 processor
  • 16 GB RAM minimum, GPU recommended
Download AppImage
.deb (Ubuntu/Debian)|.rpm (Fedora/RHEL)

v0.2.0-beta · x86_64 · AppImage bootstraps its backend on first run (needs python3.11+)

Windows

Coming Soon

A Windows build is on the roadmap. Star the repo on GitHub to get notified when it ships.

See it running

Captured from v3.6.1 on macOS. Click any shot to enlarge.

Four lanes, four different waveforms

Every 2-channel file decodes and draws

Faders, sends and a master bus reading LUFS

Threshold, ratio, knee, attack, release — real parameters

Effects chain on the master bus, live while it plays

Your whole library, with artwork, artists and albums

New in 3.6.1

Stem separation that works

Split a song into vocals, drums, bass and other. Each stem lands on its own lane with solo, mute, level and pan — four lanes of real waveforms, not one track wearing four labels.

Waveforms everywhere

Every 2-channel file — which is every normal audio file — used to fail to decode for display, so the Stems lanes, Library rows and the player bar all drew blank. Fixed across the whole app.

Projects that survive

The app keeps a hash-chained journal of every change you make. Reopening a project whose checkpoint is missing now rebuilds it from that journal instead of discarding it. A 121-command project came back intact.

A real mixer and effects

EQ, compressor, limiter, reverb, saturator and chorus — genuine parameters on the audio graph, per track and on the master bus, with LUFS on the master meter.

Browser Studio

Use the studio in your browser — no install required.

majiks.music

Build from Source

Clone, build, contribute — Apache 2.0.

$ git clone https://github.com/Majiks-Studio/majiks-music-studio.git
View on GitHub

Sign in, no license keys

Sign in with Unicorn Commander SSO. Publishing to majiks.online is free with a sign-in.

majiks.online

13,971 free sounds

Drums and stems, free with an account — browse and pull them straight into a session.

sounds.majiks.studio

System Requirements

PlatformOSProcessorRAMStorage
macOS14.0+ (Sonoma)Apple Silicon or Intel16 GB min, 32 GB+ rec.~15 GB after setup
LinuxUbuntu 22.04+, Fedora 38+x86_6416 GB min, 32 GB+ rec.~15 GB after setup
WindowsComing soon

GPU recommended for all platforms. Required for reasonable generation speed. Internet required on first launch to download model weights.

Features

Multi-Server Generation

Connect multiple ACE-Step servers simultaneously. Round-robin or priority-based routing distributes generation across local and remote GPU backends.

Cover & Style Transfer

Upload a reference track and generate new songs that match its style, tempo, and feel. Cover mode re-imagines existing songs with AI vocals.

Audio Continuation

Extend any track by generating seamless continuations. The AI analyzes the ending and creates natural transitions into new sections.

Genre Variation

Take any generated song and create variations in different genres, tempos, or moods. Explore how your idea sounds as jazz, EDM, or country.

Batch Generation

Queue multiple songs with different prompts and settings. Generate a full album overnight. Background processing with progress notifications.

Import & Analysis

Import existing audio files. AI detects BPM, key, chord progressions, and structure. Use analysis results as starting points for new generations.

Synced Lyrics

AI-generated lyrics with word-level timing sync. Edit lyrics in the built-in editor. Export as LRC for karaoke or subtitle overlays.

Drag & Drop Workflow

Drag songs between library sections, into albums, or out to Finder and other apps. Drag audio files in to import or use as style references.

MCP Server

Full MCP integration for Claude Code. Generate music, manage the library, queue batch jobs, and automate production workflows programmatically.

3 Model Sizes

Small, medium, and large ACE-Step models. Choose based on your hardware — smaller models run on 8 GB systems, large models deliver studio-quality on 32 GB+.

Previous Versions

v1.5.7

macOS

Art Studio integration, catalogue grouping, album field editor, branding update.

Download

v1.5.4

macOS

Catalogue artwork thumbnails, pull-sync, metadata sync, artwork caching

Download

v1.4.2

macOS

Bug fixes: error messages, library filters, export reliability, stable variant IDs

Download

We recommend always using the latest version. Older versions are provided for compatibility testing.