SongForge  ·  macOS  ·  Release validation

Your songs,
your Mac.
AI that listens.

Generate full songs — vocals and instruments together — with local ACE-Step inference. Set lyrics, style, tempo, key, and language, then keep the work in a local Mac library.

macOS 14 Sonoma or later  ·  Apple Silicon recommended  ·  On-Device AI

SongForge macOS interface: text prompt input glowing pink, equalizer visualization, song library sidebar, Generate button

At a glance

What is SongForge?

SongForge is a native Mac music-generation workspace that uses a compatible ACE-Step model to create vocals and instruments locally. Users can set lyrics, style, tempo, key and language, then organise songs and variations in a local library.

Platform
macOS 14 Sonoma or later; Apple silicon recommended
Best for
Private text-to-song experiments with detailed musical controls
Processing
Normal standalone generation runs locally after model setup
Availability
Release validation; no public purchase endpoint yet

Product facts reviewed against the current application repository on . See the verification method and corrections policy.

100%
Local standalone generation
On-Device
ACE-Step 1.5 running locally
Local-first
Models, prompts, and outputs live on your Mac
1-tap
Full song generated in minutes

A local music-generation workspace for Mac

SongForge (bundle ID: com.hrs.songforge.app) is a native macOS application built by HighRoad Software (COBI Management SRL, Belgium). It generates vocals and instruments together with a compatible ACE-Step 1.5 model running locally on the Mac. You provide a prompt and can set lyrics, style, BPM, key, and vocal language. Models are downloaded or linked during setup, and the normal standalone generation path then works locally. Network exceptions include model downloads, purchase or direct-build licence validation, support links, and an Audio Unit remote backend only when the user explicitly configures and confirms one. The current release build does not offer model training as a supported feature. SongForge requires macOS 14 or later and is currently in release validation.

Current build

From a prompt to a library you control.

Write the brief, keep the results organised, then make a variation without leaving the app.

SongForge creation workspace with lyrics and style controls
Set lyrics, style, tempo, key, and generation options.
SongForge local song library
Keep generated songs and their settings in a local library.
SongForge remix workspace
Create a variation while keeping the source close at hand.

Everything you need to make music with AI

From a single sentence to a full produced track — SongForge handles the whole creative process on your Mac.

01 — Text-to-Song Generation

From idea to full song in minutes

Type what you hear in your head. SongForge's generation panel gives you full control: add a text prompt, paste your lyrics, pick a style preset, set the BPM and key, choose a vocal language, and hit Generate. Use Simple mode for one-click results or Custom mode to fine-tune every parameter.

  • Simple and Custom generation modes
  • Lyrics, style, BPM, key and vocal language controls
  • Variation groups — generate siblings from the same prompt
  • Re-roll individual songs without losing the rest
  • Parallel tasks — queue multiple generations at once
  • Keyboard shortcut ⌘G to generate from anywhere
02 — Local AI Engine

Your Mac is the studio

SongForge runs the ACE-Step 1.5 model entirely on your hardware using GGUF quantization and Apple Metal GPU inference. On first launch, download the 4-file model set (~6 GB) from Hugging Face — after that, every generation is completely offline.

  • ACE-Step 1.5 — state-of-the-art open music model
  • GGUF quantized for efficient on-device inference
  • Metal GPU acceleration on Apple Silicon
  • No internet required after first model download
  • No prompts or audio logged anywhere
  • GDPR by design — you are the sole data controller
03 — Library & Playback

Everything you make, beautifully organised

Every song you generate lands in your Library. Browse, search, and organise your creations with a full-featured player bar: scrubbing, shuffle, repeat, playback rate control, and keyboard shortcuts. Liked songs, variation groups, and smart search keep your library tidy as it grows.

  • Full player bar with scrubbing and playback rate
  • Shuffle and repeat modes
  • Liked songs & variation group organisation
  • Full-text search across all songs
  • Library backup and restore
  • Keyboard shortcuts: ⌘P play/pause, ⌘L library
04 — Export & Remix

Take a track further

The release build includes work toward MP3 and WAV export, stem separation, and VAE remix tools. Final paid-edition scope and pricing will be published only when checkout is live; model training is not part of the current release.

  • Local creation and project library
  • MP3 & lossless WAV export
  • Stem separation: vocals, drums, bass, melody
  • VAE remix tools — re-compose existing songs
  • Final paid-edition scope confirmed at launch

Three steps from prompt to playback

Choose your model

On first launch, SongForge downloads the ACE-Step 1.5 GGUF model set (4 files, ~6 GB) from Hugging Face. This is a one-time download — the app guides you through it automatically and all future generations run fully offline.

Describe your song

Enter a text prompt describing the song you want. Optionally add lyrics, choose a style preset, set BPM and key, and select the vocal language. Use Simple mode for instant results or Custom mode for precise control over every parameter.

Let the AI create

Press Generate (⌘G). ACE-Step 1.5 runs on your Metal GPU and produces a complete song — vocals and instruments — in minutes. Listen immediately in the built-in player, generate variations, or export to MP3/WAV with Pro.

Local by default, explicit when connected

Standalone generation and the project library run locally, with no SongForge account or analytics. Model delivery, purchase validation, support links, and an Audio Unit remote backend you explicitly configure are the documented network exceptions.

Local SQLite
All song data and settings stored in a local database on your device — no sync, no cloud backup.
No Account
No SongForge account is needed for standalone generation. A direct paid build may validate a licence key.
On-Device Inference
ACE-Step 1.5 runs entirely on your GPU. No prompt or audio fragment is sent to any server.
Hardened Engine
The release build uses the macOS sandbox and hardened runtime. Distribution details will be confirmed at launch.

Release details, without guesswork

The current test tier allows three completed creations. Paid-edition terms will appear here only after the public build and checkout have been verified.

Current test tier

3 creations

For evaluating the current build.

  • Three completed creations
  • Full Library & Playback
  • Keyboard shortcuts ⌘G ⌘P ⌘L
  • Liked songs & variation groups
  • Library backup & restore
Ask about the release
In validation

Planned paid edition

Not yet on sale

No checkout is advertised before it is live.

  • Final feature set confirmed at launch
  • Final price shown before purchase
  • Licence and refund terms shown at checkout
  • Model training is not included in this release
Check availability

This is release-status information, not a purchase offer. Availability, price, licence scope, and refunds will be stated at the point of sale.

Common questions

SongForge is a native macOS app that generates full songs — vocals, instruments, and structure — from a text prompt you provide. It uses the ACE-Step 1.5 AI model running locally on your Mac via Metal GPU acceleration. You describe the song you want (style, lyrics, BPM, key, vocal language) and the model produces a complete audio file entirely on your device. No audio or prompts are sent to any server.
Normal standalone generation runs locally with a compatible ACE-Step 1.5 model. Network access is still used for model downloads, purchase or direct-build licence validation when applicable, support links, and an Audio Unit remote backend only when you explicitly configure and confirm one.
SongForge uses ACE-Step 1.5, a state-of-the-art open-source music generation model. The model weights are distributed in GGUF quantized format (a 4-file set) for efficient on-device inference. You download the model set on first launch from Hugging Face — the app guides you through this automatically.
The GGUF model set is approximately 6–15 GB depending on the quantization level you choose. SongForge guides you through the download on first launch. The download is a one-time step — the model is stored locally and used for all future generations without re-downloading.
Standalone prompts, lyrics, projects, and generated audio stay on the Mac. SongForge has no analytics or SongForge account. Network exceptions are model delivery, Store or direct-build purchase validation, support links, and an optional Audio Unit remote backend that you configure yourself.
The current test build allows three completed creations and includes the Library and Playback experience. This describes the build under validation, not a public sale or a promise that commercial terms will remain unchanged.
MP3 and WAV export, stem separation, and VAE remix tools are part of the paid-edition work in the repository. Their final availability and licence scope will be stated when the public checkout is live.
SongForge is in release validation. A public Store or direct-download link will be published here after that endpoint and the corresponding build have been verified.
It suits people who prefer a Mac-native, local-generation workflow and accept a substantial model download plus hardware-dependent generation speed. Suno and Udio are cloud products with different workflows and terms. See the Suno comparison and Udio comparison; always verify competitors' current plans on their own sites.
After model setup, normal standalone generation can work offline. Model downloads, purchase or licence validation, support links, and any Audio Unit remote backend you choose to configure require a network connection.
Songs are generated locally on your Mac, and HighRoad Software claims no ownership of your output — there is no cloud service whose terms grant anyone else rights to your creations. For commercial use, review the underlying ACE-Step model license and the SongForge Terms of Use, which set out the conditions that apply to generated audio.
Commercial terms are not yet live. The final price, included features, device entitlement, and refund route will be shown here and at checkout before any purchase.
SongForge app icon

Follow the SongForge release

The product is in release validation. We will link the public build only after its download and commercial terms are verified.

Check release status Read availability notes

macOS 14 Sonoma or later  ·  Apple Silicon recommended  ·  On-Device AI · Bundle ID: com.hrs.songforge.app