Full-quality stems. No account, no per-track fee.Same model and same 192 kbps (or lossless WAV) as the paid tier — tracks up to 15 minutes.
One person and one graphics card. I built StemGrab, and it runs on a desktop machine in the Netherlands that I own and keep switched on; the card doing the separating is the same one I use for everything else I make. There is no office behind the word "we" you normally read on a page like this, so I have stopped writing it.
The hard numbers, so you know before you upload. Every value here is what the server actually enforces.
| Free | With a pass | |
|---|---|---|
| Price | Free. No account, no sign-up, no card. | €5 for 30 days, or €3 for 24 hours |
| Maximum track length | 15 minutes | 60 minutes |
| Maximum file size | 2 GB | |
| Files per upload | 1 | 10 |
| Jobs at once | 2 | 10 |
| Queue | Normal | Priority when the cards are busy |
| Results kept for | 45 minutes | 24 hours |
| Play-along mixes | No | Yes — no-drums, no-bass, no-guitar, no-piano, backing, rhythm |
| Stems you get | 7 files: vocals, drums, bass, guitar, piano, other, plus a clean instrumental | |
| Output formats | 192 kbps MP3 or lossless WAV — both free, plus a zip of all stems | |
| Input formats | 20 audio and video formats: MP3, WAV, FLAC, M4A, AAC, OGG, OPUS, WMA, AIFF, AIF, MP4, MOV, MKV, WEBM, AVI, M4V, MPEG, MPG, WMV, 3GP | |
| Key, scale and BPM | Detected on every upload, including the Camelot code — free | |
| The separate audio tools | 120 MB, up to 1 hour | 300 MB, up to 3 hours |
| Where it runs | On our own graphics cards in the Netherlands. Your audio is never sent to a third-party service. | |
| Your upload | Deleted the moment processing finishes, success or failure — on both tiers. | |
| Watermark | None, on any output, on either tier. | |
The free tier is not a trial and does not run out. What the pass buys is length, volume and speed — the separation model is the same either way.
Splitters normally bill per minute of audio because their GPUs are rented per second: LALAL.AI, Moises and the rest are paying a cloud provider for every second your track is being pulled apart. I paid for the card once. A four-minute song costs me a few cents of electricity, so billing you per minute would be theatre. The ads cover the power and the domain. There is something you can buy — a pass of €5 for 30 days or €3 for 24 hours — which raises the track limit from 15 to 60 minutes, lets you queue ten files at once instead of one, adds the play-along mixes and keeps your results for a day instead of 45 minutes. It is a one-off, it changes nothing about the separation itself, and the free tool beside it is not a trial and does not run out. Lossless WAV is not part of it: that is free for everyone.
The privacy setup is deliberately boring:
StemGrab runs one of the best open source-separation models available, in its 6-stem version. One upload gives you seven files: vocals, drums, bass, guitar, piano, other, and a clean instrumental. If you are curious how a neural network manages to un-mix a finished song, I wrote an honest explainer: how stem separation works, including what it still can't do.
The separation is genuinely good on vocals, drums and bass, and rougher on guitar and piano: I say so on the front page instead of pretending otherwise. When a track comes out nearly silent, which happens if you upload a podcast, the site tells you rather than handing you empty files. If something is broken or you have an idea, send me a message: it lands in my inbox, not a ticket queue.
Split a track, free →Why a pen name. I sign these pages Sam Ridder. It is a pen name: one person running a music tool out of a spare room does not need their legal name indexed for the rest of their life. Nothing else here is invented — same room, same machine, same queue.