Vocals
Instrumental
2 stems
For acapellas, karaoke and instrumentals to rap or sing over. The cleanest split.
Free AI tool
AI Stem Splitter
Split any song into vocals, drums, bass and more.
Drop an MP3 or WAVor · MP3, WAV, FLAC or M4A, up to 10 minutes
Your song is used only to make the stems and is deleted after 24 hours.
How it works
Drop a song. Get the parts.
01
Drop a song
MP3, WAV, FLAC or M4A. A finished track, a demo or a sample you want to flip.
02
Pick 2, 4 or 6 stems
The AI listens to the whole song and rebuilds each part on its own track.
03
Preview and download
Solo or mute any stem before you download. Get each one on its own, or all of them in one zip.
Choose your split
Two, four or six stems.
Pick as many parts as you need. Fewer stems come out cleaner; more stems give you more control.
Vocals and instrumental
Vocals
Drums
Bass
Other
4 stems
Most popularFor remixes, flips and sampling. Mute the drums, keep the groove, swap the bass.
Vocals
Drums
Bass
Guitar
Piano
Other
6 stems
For learning parts, transcribing to MIDI and detailed remixes, with guitar and piano on their own.
From the blog
How we teach a computer to hear every instrument.
Our own training songs, checked stems and an honest list of what is still hard.
Vocals
Drums
Bass
Other
Questions
Before you split.
2 stems gives you vocals and an instrumental. 4 stems adds separate drums and bass, with everything else in Other. 6 stems also pulls guitar and piano out of Other. Fewer stems usually sound cleaner, more stems give you more control.
MP3, WAV, FLAC and M4A, up to 10 minutes long. Stems come back in your song's format: MP3 as MP3, M4A as M4A, and WAV or FLAC as lossless FLAC (16-bit, up to 44.1 kHz).
Clean, well-mixed songs split best. Heavy reverb, distorted guitars and instruments playing in the same range can leave a little bleed between stems.
You need the rights to the music you split. Stems from your own songs, and from Beatloo sounds you own, are yours to use.
Your upload is used only to make your stems and is deleted after 24 hours. We never use your music to train our models.