StemForge

StemForge is a local, browser-based audio workstation for separating tracks, extracting MIDI, generating audio, and composing songs. It is best suited to technically confident musicians and audio enthusiasts with capable hardware who want several workflows in one place. Its breadth and local processing are strong draws, but setup and model requirements rule it out as a casual, low-maintenance option.

Overview

StemForge connects separate, MIDI, synth, compose, mix, and export workflows, letting users move material between stages and combine stems, MIDI-rendered tracks, synth outputs, and composed songs. That integration suits projects that need more than one kind of audio tool; it is less compelling if the goal is simply to split a song or make a quick remix.

The project says audio stays on the user's computer, and an internet connection is unnecessary after the initial model downloads. That makes local processing attractive to users who want to keep their audio on their own hardware. The privacy benefit comes with a practical cost: models must be downloaded, and some are substantial.

Key features

Stem separation and MIDI

Demucs and BS-Roformer can separate vocals, drums, bass, other, guitar, and piano. StemForge recommends engines and models using spectral analysis, which gives users a way to choose a suitable separation path without making them pick entirely blind. It also makes the tool relevant to arrangements that need individual instrumental parts, not only vocal removal.

BasicPitch extracts MIDI from instrument stems; vocal melodies use faster-whisper and pitch tracking. MIDI can also be sent to hardware synthesizers through Web MIDI. That range supports workflows from turning parts into editable notes to controlling external gear, though MIDI Out requires Chrome or Edge and a secure browser context such as HTTPS or localhost.

Generation, composition, and export

The Synth tab uses Stable Audio Open to generate text-conditioned audio up to 600 seconds, with optional audio or MIDI conditioning. Compose uses AceStep for song generation and RVC for voice conversion, with 14 built-in voices and a HuggingFace model browser. These tools give StemForge reach beyond separation, but their model requirements make it a heavier commitment than a focused online generator.

Mix brings stems, MIDI-rendered tracks, synth outputs, and composed songs together. Finished work can be exported as WAV, FLAC, MP3, or OGG, either track by track or in a ZIP archive. This is a useful endpoint for a multi-stage workflow rather than a collection of isolated experiments.

Pricing

StemForge is free and open source under Apache 2.0, with no subscriptions or per-track fees. The free plan runs locally on the user's hardware, so there is no paid tier to unlock; in exchange, users supply the computer, software prerequisites, and storage for models.

Model terms are not uniform: the project notes that Stable Audio Open is subject to Stability AI's Community License, and one six-stem BS-Roformer model has no specified license. Users should account for those separate model terms rather than treating the application's Apache 2.0 license as covering every model.

Platforms

StemForge supports Linux, macOS, and Windows, and runs locally through a browser. Setup requires Python 3.12, FFmpeg, and FluidSynth with a GM soundfont. NVIDIA GPU acceleration requires driver 580 or newer; CPU-only operation is significantly slower, so users without a compatible GPU should expect a less responsive workflow.

Models download lazily when first used. AceStep models for Compose require about 20 GB, while Stable Audio Open requires HuggingFace license acceptance and authentication. These requirements make available disk space and willingness to manage downloads part of the platform choice, not an afterthought.

Who it's for

StemForge is a strong fit for producers, musicians, and audio tinkerers who want separation, MIDI extraction, generation, composition, and mixing connected in a local workflow. It is especially suited to people comfortable installing dependencies and managing large models, and to those who value keeping audio on their own computer.

It is a poor match for users who want a lightweight web service, need reliable CPU-speed processing, or prefer not to handle local setup and model licensing. Hardware synth users should also check their browser and secure-origin requirements before relying on MIDI Out.

Pros and cons

• Pros: A connected workflow spans separation, MIDI, generation, composition, mixing, and export, reducing the need to move between specialized tools.
• Pros: Audio processing stays local after downloads, and the software has no subscription or per-track fees.
• Pros: Separation covers six stem categories, while export supports WAV, FLAC, MP3, and OGG.
• Cons: Python, FFmpeg, FluidSynth, and a GM soundfont add setup work before the workflow is ready.
• Cons: CPU-only processing is significantly slower, and AceStep's Compose models need about 20 GB.
• Cons: Some model terms are separate from the application's Apache 2.0 license, including one six-stem BS-Roformer model with no specified license.

Alternatives

For a broader comparison, browse AI Music Remixing Tools.

Choose Fadr instead if a freemium service with web and mobile platforms is a better fit: its free Basic plan includes vocals, melodies, drums, and bass, plus MP3 downloads, MIDI detection, and Remix Maker.

AceStep AI Music Remixer is another option for users who prefer a web-based freemium tool, with a free allowance of 10 credits a day and generations up to 120 seconds.

AetherWave Music Studio, MeloLab AI Remix, MusicWave.ai, AI Remix Music Generator, Mubert AI Music Remixer, and RiffGen Remixer AI are also alternatives to consider.

Verdict

Choose StemForge if you want a free, locally run workstation that links stem separation, MIDI, generation, composition, mixing, and export—and are prepared to manage its technical setup and models. Its main advantage is that range in a single local workflow; look elsewhere if you need a simple service, modest hardware demands, or fewer model and licensing considerations.