Frequently asked questions
Everything people ask about Stemina, answered plainly. Short version: it is free, it runs on your device, and your audio is never uploaded.
Is Stemina really free?
Yes. There is no account, no credits, no trial and no limit on how many tracks you separate. A single track can be up to 30 minutes long, because the decoded audio has to fit in the browser tab’s memory. Separation runs on your own hardware, so there is no server bill to pass on. If you find it useful, there may be a “Support Stemina” link in the footer.
Is my audio uploaded anywhere?
No. Decoding, separation, playback and export all happen inside your browser tab. Nothing leaves your device, but it is kept on it: every finished separation, with a copy of the original file, is saved automatically in your browser's private storage for this site, so you can reopen it later. Remove one with Delete under Stored projects on the start screen, or clear this site's data in your browser settings to remove everything. A run that is interrupted leaves partial files, which are removed automatically after six hours.
What network requests does the app make?
One: it downloads the separation model from Hugging Face the first time, then keeps it on your device. The app ships no analytics and no third-party scripts. A link you click, such as the support link, opens another site in a new tab like any link.
What is the script on these guide pages?
The guide pages, not the app, include agentprose, an open-source script that gives AI assistants in your browser read-only tools to search and quote these guides. It is served from this site, sets no cookies and makes no request when the page loads. Only when an AI agent uses one of its tools does it contact agentprose.cloud, with the search query and the tool’s name — never your audio, and never anything from the app.
Which stems do I get?
Six: vocals, drums, bass, guitar, piano and other. “Other” is everything that is not one of the first five, such as synths, strings and pads. The model always produces all six in one pass.
Which browsers does it work in?
Desktop Chrome or Edge 121 or later is the fast path, because they support WebGPU. Without WebGPU the app falls back to WebAssembly on the processor, which works but is roughly ten times slower.
How long does a separation take?
It depends on your graphics chip. The slowest GPU profiled so far, an integrated Intel Iris Xe, takes about 17.6 seconds per four-second segment. At the default 25 % overlap a four-minute song is 80 segments, so it takes roughly 23½ minutes there, plus loading the model; a dedicated graphics card is much faster. On the WebAssembly fallback, the 15-second sample took just under two minutes.
Why is there a large download the first time?
That is the separation model: 353 MB on most machines, or 701 MB on a GPU without half-precision support. It is stored in your browser's private file system so later runs start immediately and work offline. An interrupted download resumes where it stopped.
Why does it download the model again on every visit?
Browsers may clear site storage when the disk is nearly full, and iPhone and iPad clear it after seven days without a visit. Stemina asks the browser to keep its storage, but the browser decides.
Does it work on a phone?
A recent phone may start a separation, but finishing a full song is unlikely: phones throttle under sustained load, and the work stops when the screen locks or you switch apps. A computer is the reliable choice.
Which file formats can I use?
Input: MP3, WAV, FLAC, M4A and OGG. Output: WAV at 16-bit or 32-bit float, a ZIP of every stem, or an Ableton Live project with each stem on its own track.
How do I make an instrumental or karaoke version?
Mute the vocal lane in the Mix bench to listen. Muting affects playback only: the ZIP and the Ableton project always contain all six stems. Export them, then delete or mute the vocal track in your audio editor and combine the other five. Stemina exports stems rather than a single bounced mix.
Why does a stem tick or hiss?
A short tick is usually a drum hit the model placed in the wrong stem. Hiss is broadband residue; raising the overlap on the Configure screen lowers it by about a decibel at the cost of time. Stemina deliberately applies no clean-up filters after the model.
Can I use the stems commercially?
The stems are still the original recording, so the song's rights holders decide. Separating a track does not grant any rights to it. Clear the recording before you release anything built on it.
Can I use it offline?
Yes, after the first run. Your first separation leaves everything Stemina needs on your device: the app, these guides, the separation engine and the model, which is checked against its published SHA-256 hash. After that you can open stemina.bpmtech.no with no connection at all and separate tracks. Only the first model download needs the network.
Can I install Stemina as an app?
Yes. In Chrome or Edge, use the install icon in the address bar (or Install Stemina in the menu); on Android, Add to Home screen; on iPhone and iPad, Share and then Add to Home Screen. It opens in its own window and works offline after the first run. The model stays on the device until you delete it under Model on the configure screen.
Who makes Stemina?
Maxim Salnikov, a developer in Oslo, Norway. The separation model is the work of the open-source music-separation community; the About page credits each part of it.
Last updated