English中文日本語한국어EspañolPortuguêsFrançaisDeutschالعربيةहिन्दीไทยTiếng ViệtBahasaРусский

🎚️ Stem Splitter

AI Stem Separation in Your Browser

This tool uses Demucs v4 (htdemucs), the open-source AI model from Meta, running entirely in your browser. Upload a song and get four isolated stems: vocals, drums, bass, and other instruments. Nothing is uploaded to any server.

The model runs through a WebAssembly build of the ggml inference engine, with the 80 MB model file cached in your browser after the first load. Separation speed depends on your device: a 3-minute song typically takes 1-3 minutes on a desktop CPU, faster on machines with more cores.

Pro Tips

  • For cleanest results, use a high-quality MP3 (256 kbps or higher) or WAV file.
  • Songs with heavy effects on vocals may leave artifacts in the stems - this is a known limitation of all separation AI.
  • Close other browser tabs during separation to free up memory - the model needs several GB of RAM.

Frequently Asked Questions

Is stem separation free?

Yes - completely free with no limits. The AI model runs on your own device, so there are no server costs to pass on.

Does my audio get uploaded to a server?

No. The file never leaves your browser. Decoding, AI inference, and stem rendering all happen locally on your device.

What can I use the separated stems for?

Remix stems in a DAW, make karaoke tracks by removing vocals, isolate a drum loop for sampling, or practice your instrument against a backing track.

How long does separation take?

A 3-minute song takes roughly 1-3 minutes on a typical desktop. The first run is slower because the 80 MB model needs to download once.

Which browsers are supported?

Chrome and Edge are recommended for best performance. Other modern browsers work but may be slower. Mobile devices can work for short clips but memory is limited.