Which stems do I get?

Four: vocals, drums, bass and other. Guitars, keys, synths, strings and any percussion outside the kit land in Other, because that is how the model was trained. Anyone promising eight or twelve clean stems from a stereo master is promising more than current separation delivers.

No. Stem separation runs inside this browser tab with an open-source model your browser downloads once and caches. Your song is never uploaded, never stored, and never used to train ANYANO models.

Can I use these stems in a DAW?

Yes. Export 24-bit WAV and every stem comes back at the original length and sample position, so dropping all four onto separate tracks lines them up exactly with each other.

Why do the stems sound thin on their own?

Because separation moves energy that was shared between parts into one of them. Played together the stems sum back to something very close to your original; solo, each one is missing the masking that made it feel full in the mix.

What size and length can I use?

Up to 100 MB and 12 minutes per file, in MP3, WAV, FLAC, M4A, AAC or OGG. Six separations an hour per visitor.

Free tool

AI Stem Splitter

Turn a finished stereo mix into four separate stems — vocals, drums, bass and everything else — aligned and ready to drop into a DAW.

Four-stem separation runs inside your browser, not on our servers. Your song never leaves your machine and is never used to train ANYANO models.

Choose an audio file to separate

MP3, WAV, FLAC, M4A, AAC or OGG — up to 100.0 MB and 5 minutes. Processed on your own machine.

Four stems is what this model separates. Guitars, keys, synths and percussion that is not a kit all land in Other, and dense mixes bleed more than sparse ones.

Detail

How this works

Stem separation reconstructs parts rather than recovering them. The model predicts what each of the four sources would look like alone, and renders them as new audio at the original length and timing. Everything stays sample-aligned, which is what makes the results usable for remixing, practice, transcription and reference work.

Judge the split in context, not in solo. Drums and bass usually come out convincingly. Vocals depend on how much reverb and doubling is in the mix. Other is a catch-all, so expect guitars, keys and synths to share it.

Stems are also the best raw material for a private model: ANYANO trains on the stems and songs you own, keeps that dataset to your account, and generates in your direction.

FAQ

Frequently asked questions

Toolbox

See all free ANYANO tools

Train your sound. Generate your song.

ANYANO is the bigger version of this: prepare music you own, train a Sound that stays private to your account, then generate from it.