StemGrab is a small, independent project — not a startup, not a subscription funnel, not a "free trial" with a countdown hidden in the fine print. One person, one machine, one job: you upload a song, the AI pulls it apart, you download the stems. That's the whole product.
Most stem-separation sites charge per minute of audio because they rent GPU time from a cloud provider — every song you upload costs them real money, so it has to cost you real money too. StemGrab runs on our own GPU, sitting in a machine we already own. Once the hardware is paid for, splitting one more track costs roughly the electricity of a light bulb for a minute. The few ads on the site cover the power bill, and that's the entire business model. No accounts, no credits, no watermarks, no daily caps designed to push you toward a paid tier — because there is no paid tier.
The privacy setup is deliberately boring:
StemGrab runs Demucs, 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're curious how a neural network manages to un-mix a finished song, we wrote up an honest explainer: how AI 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 — we say so on the front page instead of pretending otherwise. When a track comes out nearly silent (say, you upload a podcast), the site tells you rather than handing you empty files. If something's broken or you have an idea, send us a message — it lands with the person who built the thing.
Split a track — free →