How to extract stems from a song
- Add a song. Drop in an MP3, WAV, FLAC, M4A or OGG file.
- Choose the engine. AI – HTDemucs v4 gives studio-quality stems. Instant is a quick, lower-quality option that needs no download.
- Extract. The first time, the AI model downloads once (about 180 MB). A progress bar shows how much time is left.
- Download. Preview each stem, then download them one at a time or all together as a ZIP file.
What stems do you get?
| Stem | What it contains |
|---|---|
| Vocals | Lead and backing vocals, including their reverb |
| Drums | Kick, snare, hi-hats, cymbals and percussion |
| Bass | Bass guitar, synth bass and 808s |
| Other | Everything else: guitars, piano, synths, strings, pads, brass and effects |
The stems are designed to add back up to the original song. Import them into your DAW, play them together at the same level, and you'll hear the full mix.
Powered by HTDemucs v4
This stem splitter uses HTDemucs v4 (Hybrid Transformer Demucs) from Meta AI Research, one of the most widely used open-source music separation models. It processes the song in two parallel branches, one working on the raw waveform and one on the spectrogram. A cross-domain transformer lets them share information, so the model understands both the precise timing of drum hits and the harmonic structure of instruments.
Everything runs on your device using WebGPU (your graphics card) when available, or the CPU otherwise. The song is analysed in overlapping 7.8-second windows that are blended together, so there are no audible seams.
Ideas for using stems
- Practice your instrument. Mute the bass or drum stem and play that part yourself, or use the "other" stem to practise guitar and keys parts.
- Learn songs by ear. Solo the bass or "other" stem to hear parts that are buried in the mix.
- Remix and mash-ups. Load the stems into GarageBand, FL Studio, Ableton Live, Logic or Audacity.
- DJ edits. Make acapella intros, drum-only breaks and instrumental outros.
- Karaoke in your key. Mix the non-vocal stems and transpose them with the Audio Scale Changer.
Please respect copyright: only publish or share stems of music you own, or with permission from the rights holder.
Frequently asked questions
Is the AI stem splitter free?
Yes. It's free with no account or limits. Split as many songs as you like.
Are my songs uploaded?
No. The AI model runs inside your browser on your own device, and nothing is sent to a server.
How long does stem separation take?
With a recent graphics card (WebGPU), usually 1–3 minutes for a 4-minute song. Older integrated graphics can take 5–10 minutes, and CPU-only processing takes longer. The first run also includes the one-time model download.
Which format should I download?
Choose WAV if you'll edit the stems in music software, because it's lossless. Choose MP3 for smaller files you just want to listen to.
Why do I hear a little bleed between stems?
Even the best AI models leave small traces of other instruments, especially where sounds overlap in pitch and time. HTDemucs keeps this to a minimum, and it's usually inaudible in a mix.