ArtistAssets · Free music tools
Pull the song apart.
Split a mix into vocals and instrumental, or vocals, drums, bass and everything else. The AI runs on your own computer, so your song never gets uploaded.
- One file
- WAV or MP3
- Up to 100 MB
- 5:00 max
Your audio stays on your device
Settings
Vocals · Drums · Bass · Other
AI model
The first run downloads the model, 172.2 MB. It's saved in your browser, so it only downloads once.
Your audio stays on your device.
How it works
- 01
Pick 2 or 4 stems
Vocals and instrumental for acapellas and karaoke versions. Vocals, drums, bass and other when you want every part.
- 02
Let it run
The first time, the AI model downloads once (about 180 MB). Then the song is processed in short sections on your GPU or CPU.
- 03
Listen and download
Solo and mute each stem to check it, then grab 24-bit WAVs one at a time or all together as a ZIP. They line up exactly.
Worth knowing
AI separation is good, not perfect.
Expect some bleed between stems and the odd watery artifact, especially on dense mixes, heavy reverb, or instruments that share a range with the voice. "Other" holds everything that isn't vocals, drums or bass: guitars, keys, synths, pads. Only separate songs you have the rights to use.
Separation by HT-Demucs (Rouard, Massa & Défossez, ICASSP 2023; MIT), using the browser export and STFT code from demucs-web (MIT), running on ONNX Runtime Web (MIT). ZIP files by client-zip (MIT).
Need it mixed?
Want this mixed properly?
Stems are a great start. A real mix gets the vocal, low end and space sitting right, ready for release.