// AI TOOLS · STEMDECK

StemDeck

A free, open-source desktop app that runs Meta's Demucs locally to split any song into up to six stems — vocals, drums, bass, guitar, piano, and other — with no account and no upload.

Last verified · 2026-08-29 · by Moe Ameen

What StemDeck is

StemDeck is a free, open-source stem separator that runs entirely on your own machine. Instead of uploading a track to a cloud service, you install it on Windows, macOS, or Linux and the model executes locally, auto-detecting the best compute available — NVIDIA CUDA, Apple Silicon MPS, or a CPU fallback. Built by an independent developer (Thales Claro) and released under the Apache 2.0 license, it drew coverage across producer and music-tech press in August 2026 as a no-account, no-upload alternative to cloud splitters like Moises and LALAL.AI.

Under the hood it uses Demucs, Meta's open-source separation model — the six-stem `htdemucs_6s` — to split a song into vocals, drums, bass, guitar, piano, and other, with an optional UVR-MDX-NET karaoke model for a lead-versus-backing vocal split. It accepts MP3, WAV, FLAC, OGG/Opus, MP4, and M4A, plus YouTube URLs for material you have the rights to use. Once a track is processed, StemDeck opens a DAW-style waveform mixer where you can solo, mute, loop, and balance each stem, then export individual stems or a custom mix (via FFmpeg). It also runs light analysis: BPM, key/scale, and LUFS loudness.

Its pitch is privacy and price: everything runs locally, audio never leaves your machine, and it's free with no subscription. The trade-off is that "free" spends your hardware — separation is fast on a supported GPU and slow on a CPU-only machine — and it's a single-purpose audio utility, not a content or publishing tool.

What you can make with it

  • Isolated vocal stems — pull an a cappella out of a full mix for edits, sampling, or a vocal-only clip
  • Instrumental / karaoke tracks — remove the vocals to get a clean backing bed
  • Up to six separated stems (vocals, drums, bass, guitar, piano, other) for remixing and mashups
  • A lead-versus-backing vocal split via the optional karaoke model
  • Custom stem mixes exported to WAV for practice, transcription, or production
  • Quick BPM, key, and loudness readouts to prep a track for a remix or sync

How Kompozy turns StemDeck output into content

StemDeck's output is raw audio — an instrumental, an isolated vocal, a custom mix, sitting as WAV files on your disk. That audio is the ingredient, not the post, and this is exactly where [Kompozy](/) turns it into content you can publish. The pairing is compositional: Kompozy pulls a StemDeck stem *into* its own video formats. Drop your separated instrumental in as the music bed under a [Listicle Video](/glossary/output-buckets) ("5 songs that use this exact chord loop") or a [Marketing Short](/glossary/marketing-shorts), or lay an isolated vocal under a [Persona Frames](/glossary/persona-frames) HyperFrames template so a remix reveal arrives fully branded. A karaoke instrumental becomes the backing track for a cover clip; a soloed drum stem becomes the hook of a "how I made this beat" short.

From there Kompozy builds the whole release around the audio, which is the part a separator can't touch. Talk through your remix once and it generates [Persona Shorts](/glossary/persona-shorts) and HeyGen avatar video with a face-locked, consistent persona, a [Carousel](/glossary/hyperframes) breaking the arrangement down stem by stem, quote graphics, a Blog Article on the production, and an Email Newsletter — all governed by one [Persona Brief](/glossary/persona-brief) so every piece reads as one artist. Then [Autopilot](/glossary/autopilot) captions each cut, reframes it to 9:16, 1:1, and 16:9, and publishes across the eight social platforms plus blog and email behind a per-post review. StemDeck separates the audio on your machine; Kompozy wraps it in video and ships the whole drop.

  1. Separate your track in StemDeck — pull the instrumental, the isolated vocal, or a custom mix, and export the WAV.
  2. Bring the stem into Kompozy: use the instrumental as the music bed of a Listicle Video or Marketing Short, or lay a vocal under a Persona Frames HyperFrames template.
  3. Record or type the story behind the track once, and generate the surrounding set under your Persona Brief — Persona Shorts, a carousel, quote graphics, a blog, and a newsletter.
  4. Let Autopilot caption every cut and reframe it to 9:16, 1:1, and 16:9 so one release fits every platform.
  5. Review the batch in the per-post pipeline, then schedule and publish across the eight social platforms plus blog and email.

Frequently asked questions

What is StemDeck?

StemDeck is a free, open-source desktop app that separates a song into stems locally on your own machine. It runs Meta's Demucs model (the six-stem htdemucs_6s) to split a track into vocals, drums, bass, guitar, piano, and other, and gives you a DAW-style mixer to solo, mute, balance, and export. There's no account, no upload, and no subscription — audio never leaves your computer.

Is StemDeck free, and does it run offline?

Yes. StemDeck is free and open source under the Apache 2.0 license, and it runs entirely locally — processing happens on your machine with no upload. It auto-detects the best compute (NVIDIA CUDA, Apple Silicon MPS, or CPU), so it's fast on a supported GPU and slower on CPU-only hardware. There's no mobile app.

How many stems can StemDeck separate, and from what files?

Up to six stems — vocals, drums, bass, guitar, piano, and other — plus an optional lead/backing vocal split via a UVR karaoke model. It accepts MP3, WAV, FLAC, OGG/Opus, MP4, and M4A files, and can also pull audio from a YouTube URL for material you have the rights to use.

Can StemDeck turn my stems into videos or social posts?

No. StemDeck separates and exports audio; it doesn't make video, captions, or posts, or publish anywhere. Pair it with a content engine like Kompozy: use a StemDeck instrumental as the music bed of a branded short, generate the surrounding posts (Persona Shorts, carousels, a blog, a newsletter) under one Persona Brief, then caption, reframe, and publish across nine platforms.

Related tools

  • SunoThe consumer AI music generator that writes a full song — lyrics, vocals, instrumentation, and mix — from a text prompt, now under a copyright cloud over how it was trained.
  • DescriptAn AI audio and video editor you drive by editing the transcript — with the Underlord AI assistant, Studio Sound, AI avatars and voices, and translation, built for podcasts and video.
  • FFmpeg 9.0The August 2026 major release of the open-source media engine, "Lei" — native animated WebP decoding, more GPU-accelerated processing, and wider HDR handling.

← All AI tools · Get started →