Split a song into 4 AI stems — vocals, drums, bass, and other — then solo, mute, and mix them. Every stem downloads in full stereo.
The stem splitter breaks a song into four separate tracks — vocals, drums, bass, and everything else — and gives you a mixer to solo, mute, and balance them before you export. It's the same neural separation engine behind the vocal remover, run to its full four-way output rather than collapsed into a simple vocal/instrumental pair. Each stem comes out in full stereo, so you can drop them straight into a DAW and rebuild an arrangement piece by piece.
Remixers use it to isolate a drum groove worth sampling, producers to study how a record was put together, and musicians to mute one part and play along with the rest. Worth knowing: the two-stem split is derived from this same four-stem output, so if all you want is the instrumental, the vocal remover gives it to you as one piece with fewer seams to hear. More on the tradeoff in how AI vocal removal actually works.
Four: vocals, drums, bass, and everything else (guitars, keys, synths, and other instruments, grouped as 'other'). This is the standard four-stem breakdown from the Demucs model, the same AI engine behind the Vocal Remover — but instead of collapsing everything except the vocals into one instrumental track, the Stem Splitter keeps all four separate so you can work with each independently.
The Vocal Remover gives you two tracks — the vocals and one combined instrumental. The Stem Splitter runs the same separation but keeps all four stems apart, and adds a mixer: you can solo or mute each stem, adjust their levels, play them back together in sync, and download any of them individually. It's the tool for remixing, making custom backing tracks, isolating a bassline or drum groove, or practicing one instrument.
Yes. Each stem has a solo button (hear only the soloed stems), a mute button (silence that stem), and a volume slider, all applied live while the track plays in sync. Solo and mute work like a real mixing console — solo any combination to focus on just those parts, or mute a stem to hear the song without it. You can then download each stem separately in WAV, MP3, or Opus.
It uses HTDemucs, one of the strongest openly-available separation models, and quality is generally very good on well-produced music — clean vocals, tight drums, and a usable bass stem. As with any AI separation, densely-layered or heavily-distorted mixes can leave some bleed between stems (a little vocal in the 'other' stem, for example). Results run on your own device's hardware, so a modern laptop with a GPU will process faster than an older phone.
No. Decoding, AI separation, mixing, and export all happen locally in your browser — your audio never leaves your device. The only network request is a one-time download of the AI model from a CDN on first use (a download to your device, covered in our Privacy Policy); your audio is never sent anywhere. This makes it safe for unreleased tracks and stems you're not authorized to redistribute.