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.
Free, no sign-up, no watermark. Drop a song, get the tracks back in about six seconds.
Drop a song — then pick what you want: all 7 tracks, vocals + instrumental, the key & BPM, or a pitch change.
What's on this page is the whole tool, not a trial version. A pass doesn't unlock the separation and doesn't improve the audio — it adds a set of extras next to it, for people who want to practise with the stems.
This is the whole tool, not a trial. Same model and the same audio quality as with a pass — nothing here is switched off.
Handled by Polar, our payment provider — they take care of the payment and the VAT. No account here: your pass is stored in this browser.
Any common audio format. Separation starts immediately: no account, no e-mail, no trial.
A music source-separation model pulls the mix apart on our GPU: vocals, drums, bass and everything else, usually in under a minute.
Preview every stem in the browser, download the ones you need as MP3, or grab the whole set as a zip. Your upload is wiped automatically.
A clean instrumental for karaoke or covers, or the isolated vocal for a remix; both come out of every split.
Mute the guitar, solo the bass, loop the drums, and practise along with the real track, one instrument at a time.
DJs and producers: grab clean source material for edits, mashups and transitions instead of EQ-hacking the full mix.
Over the 40 days from 13 July to 21 August, StemGrab logged 26,930 page views from 25,611 real visitors — bots filtered out — arriving from 155 countries, most from the US, India and Germany. Of everything on offer, the plain vocal remover is what people actually reach for: 1,776 uploads went through it, more than six times the next most-used option (the version that keeps the lyrics on screen, 273 uploads), with the full seven-stem splitter behind that at 112. Not every upload became a track: 1,336 were declined before a file ever got processed. The usual reasons are duller than a broken tool — the wrong file type, no extension on the upload, or a track that ran past the 15-minute limit. Even the optional €5 pass has its own friction: the single most common error logged anywhere on the site in 40 days was a checkout that didn't accept the e-mail address typed in, 82 times. None of this is marketing copy — it's what our own logs show once the bots are filtered out, not a guess at what a "typical" visitor does.
Yes: no account, no watermark, no daily cap tricks, and the free tool is not a trial — it doesn't expire and it uses the same model as everything else here. Ads keep the GPU running. There is one optional extra: a €5 month pass with practice tools, play-along mixes and higher limits. It buys you more to do with your stems, not better stems.
It runs for 30 days from your first track, not from the moment you pay, so buying one at night and starting the next morning costs you nothing. It is a one-off €5 payment, not a subscription. The pass lives in the browser you bought it in — there are no accounts — so it won't follow you to another device.
Your upload is deleted the moment it has been processed, and the stems are wiped from the server within 45 minutes — 24 hours if you have a month pass. Nothing is stored, shared or used to train anything — see the Privacy Policy.
We use one of the best separation models available. Vocals, drums and bass come out remarkably clean on most modern mixes. Guitar and piano are newer, experimental stems; expect them to be rougher, especially in dense mixes. Lo-fi recordings are harder for any tool.
Seven files per track: vocals, drums, bass, guitar, piano, other and a mixed instrumental (everything except vocals), plus a zip with the lot, as 192 kbps MP3 — or uncompressed WAV with a month pass. Want the notes instead of the audio? Get MIDI turns vocals, bass, guitar, piano and other into their own MIDI track each — drums excluded, since they have no pitch to convert.
Yes, on every upload: above your stems you get the key with its scale, the Camelot code for mixing in key, and the BPM. It is an estimate read from the audio — steady-tempo studio tracks are reliable, a half-time feel can read as half or double the tempo, and when there is too little to go on the badge is left off instead of guessing. More: song key & BPM finder.
It's built for music. A podcast or voice memo has no instruments to separate (you'll get a near-silent instrumental), and an instrumental track has no vocals to isolate; the site will tell you when that happens instead of leaving you guessing.
The tool is yours to use; the rights to the music are not ours to give. For covers, remixes or anything you publish, make sure you have permission from whoever owns the recording.
Same house, different tool: PicReviver restores old photos and BGNinja removes photo backgrounds — both free. The whole family lives at ShinobiTools.
A finished record is one flat file: the voice, the drums and the bass were mixed into a single waveform years ago and there is no hidden track underneath waiting to be switched off — that information is genuinely gone. A separation model reconstructs it. Trained on thousands of songs where it saw both the mix and the original parts, it has learned what a human voice looks like in a spectrogram and what a snare does, and it rebuilds each part on top of what it recognises. That is why the result is very good but never perfect, and why some stems are cleaner than others: vocals, drums and bass have distinct fingerprints, while guitar and piano overlap with everything around them. A quiet cymbal buried under a guitar in the original sounds thin when you pull it out, because the model has to guess at what it could not hear.
Separating a three-minute song takes a GPU about six seconds. Most free stem splitters rent that GPU time from a cloud provider by the second, which is why they cap you at ten minutes of audio and then ask for a card — the meter runs on every upload. We own the hardware: the same graphics cards run our other tools in the Netherlands, so an extra song costs electricity rather than an invoice, and the ads around the page cover the rest. That is the whole reason there is no account, no e-mail, no watermark and no counter that runs out.