s t u d i o///diskrot

Studio by diskrot is an offline generative music creation web app. It runs entirely on your machine — no cloud accounts, no subscriptions, no data leaves your computer.

Screenshot

Features

• Text-to-music generation — Describe a song with lyrics, genre, BPM, and key; get a full audio track back (ACE-Step 1.5)
• Speech and audio generation — Generate speech and sound effects (Bark)
• Text generation — Local LLM for lyrics writing and audio prompts (YuLan-Mini via llama.cpp)
• MIDI generation — Generate and arrange MIDI compositions (optional)
• LoRA training — Fine-tune music models on your own audio data
• Song management — Organize, tag, and iterate on generated songs
• Fully offline — Everything runs locally via Docker; no internet required after initial setup

Architecture

┌─────────────────────────────────────────────────────┐
│  Browser  →  Studio UI (Flutter web, :3000)         │
│               ↓                                     │
│            Studio Backend (Dart/Shelf, :8080)        │
│               ↓              ↓            ↓         │
│          ACE-Step(:8001)  Bark(:8002)  YuLan(:8003) │
│               ↓                                     │
│          PostgreSQL(:5432)  Redis(:6379)             │
└─────────────────────────────────────────────────────┘

Layer	Technology
Frontend	Flutter web (desktop-first)
Backend	Dart, Shelf, Drift (PostgreSQL)
Music generation	ACE-Step 1.5 (PyTorch)
Speech/audio	Bark (PyTorch)
Text/LLM	YuLan-Mini via llama.cpp (CPU or CUDA)
MIDI	Custom transformer model (PyTorch, optional)
Infrastructure	Docker Compose, PostgreSQL 16, Redis 7

System Requirements

• OS: Windows 10/11 (with WSL2) or macOS 12+
• RAM: 16 GB minimum, 32 GB recommended
• Disk: ~20 GB for Docker images and model weights
• GPU (optional): NVIDIA GPU with 8+ GB VRAM for accelerated generation. CPU-only mode works on all hardware.

Install

Windows — open PowerShell as Administrator and paste:

irm https://raw.githubusercontent.com/diskrotrepo/studio/stable/installer/install.ps1 -OutFile $env:TEMP\diskrot-install.ps1; powershell -ExecutionPolicy Bypass -File $env:TEMP\diskrot-install.ps1

macOS — open Terminal and paste:

curl -fsSL https://raw.githubusercontent.com/diskrotrepo/studio/stable/installer/install.sh -o /tmp/diskrot-install.sh && bash /tmp/diskrot-install.sh

The installer handles everything — WSL, Docker, NVIDIA drivers, GPU detection, and launching the stack. First startup downloads model weights and may take several minutes.

Update

Run the same installer again, or use the launcher scripts (start.bat / start.sh) in the project folder. Both auto-detect whether to install or update.

Uninstall

Platform	Command
Windows	Double-click installer/uninstall.bat
macOS	./installer/install.sh --uninstall

Usage

Once running, open http://localhost:3000 in your browser.

Development

See CONTRIBUTING.md for the full development guide.

Quick start

Run the setup script to install all development prerequisites (Git, Docker, Dart SDK, Flutter SDK), configure the environment, fetch dependencies, run code generation, and start supporting services:

Windows:

setup-dev.bat

macOS / Linux:

./setup-dev.sh

Pass --skip-docker to skip Docker installation and service startup.

Nightly (dev-channel) images

To test the latest changes from the main branch before they land in a stable release, use the nightly scripts. These pull pre-built -dev tagged images from Docker Hub — no local build required.

Script	Platform
start-nightly.bat	Windows
start-nightly.sh	macOS / Linux

Flag	Description
--gpu	Force GPU compose overlay
--cpu	Force CPU-only (skip GPU overlay)
--pull-only	Pull images without starting services

Or manually:

docker compose -f docker-compose.yml -f docker-compose.nightly.yml up -d

Building from source

To build from local source instead of pulling prebuilt images:

Script	What it rebuilds
dev.bat / dev.sh	Backend + UI only (fast iteration)
dev-all.bat / dev-all.sh	All services including models

Dev script flags

Flag	Description
--no-cache	Rebuild without Docker layer cache
--build-only	Build images and exit (don't start services)
--gpu	Force GPU compose overlay
--cpu	Force CPU-only (skip GPU overlay)

Services

Service	Port	Notes
studio-ui	3000	Flutter web app
studio-backend	8080	Dart API server
acestep	8001	Music generation (CPU or GPU)
yulan	8003	Text model via llama.cpp (CPU or GPU)
postgres	5432	Database
redis	6379	Cache/queue
pgadmin	5050	DB admin UI
bark	8002	Speech/audio generation (CPU or GPU)
midi	8004	MIDI generation (optional profile)
dozzle	9999	Docker log viewer
redisinsight	5540	Redis admin UI

Contributing

See CONTRIBUTING.md for guidelines on setting up the dev environment, code style, and submitting changes.

Security

See SECURITY.md for reporting vulnerabilities.

License

This project is licensed under the MIT License.

Vendored dependencies carry their own licenses:

• ACE-Step 1.5 — MIT (some model files Apache 2.0)
• Bark — MIT
• llama.cpp — MIT
• nano-vllm — MIT