One song in, four studio stems out - drums, bass, vocals and everything else - separated by the htdemucs AI model on your own graphics card. Nothing uploads.
100% in your browser - your files never leave your device.
Any MP3, WAV, OGG, FLAC or M4A, decoded locally in the page. No account, no upload queue, no per-track pricing.
The open htdemucs model downloads once (~120 MB, cached) and runs through WebGPU. It reconstructs each part from the mix - the same model family studios use for official stem releases.
Drums, bass, vocals and "other" (keys, guitars, synths) as separate studio-ready WAV files - or send them straight into the TuttiDAW editor as a grouped multitrack.
Stems turn a finished mix back into building blocks. DJs and producers drop the isolated drums under another track, borrow a bassline, or rebuild an arrangement around the vocal - the raw material for remixes, edits and mashups that a stereo file simply does not offer.
For practice, stems are the ultimate play-along: mute the bass and BE the bass player, or solo the drums to study the groove. Educators use them to demonstrate arrangement; audio engineers use them to rebalance live or archival recordings where the multitracks are long gone.
This page runs htdemucs (Demucs v4), the strongest open-source separation model, split into 21 GPU-friendly pieces so your browser stays responsive while it works. Because inference happens on your own machine, there are no upload limits and no privacy questions - split an unreleased master with a clear conscience.
Drums, bass, vocals, and "other" - everything that is not drums, bass or vocals (guitars, keys, synths, strings). Summing the four back together reconstructs the original mix.
No. The model downloads to your browser once and runs on your graphics card through WebGPU. Audio never leaves your device, so unreleased or client material stays private.
A WebGPU-capable browser - current desktop Chrome or Edge (Safari 18+ too) - and a GPU. Discrete cards separate faster than the song plays; integrated graphics work but take longer.
Uncompressed 16-bit stereo WAV at 44.1 kHz, aligned sample-for-sample with each other, so they drop straight onto four DAW tracks and stay in sync.
Yes - the vocal remover page runs the same model but gives you exactly two files: the acapella and the summed instrumental.
Hit 'Open in TuttiDAW studio' after splitting and the four stems land as a grouped multitrack - mix, mute and rearrange them right away.
Explore TuttiDAW - the free browser DAW