A free alternative to paid stem separators

Same open models, no credit system, no premium tier holding back the good output.

Drop your song here

or click to choose · mp3, wav, flac, m4a · up to 50 MB

What do you want?

VocalsTake the singing out, or keep only the singing.You get 2 files:VocalsSong without vocalsDrumsGet the drums alone, or play along without them.You get 2 files:DrumsSong without drumsBassGet the bassline alone, or play along without it.You get 2 files:BassSong without bassEverythingEvery part on its own, and the song minus each part.You get 8 files:VocalsDrumsBassOther (guitars, keys, synths)Song without vocalsSong without drumsSong without bassSong without the rest

Most people arrive at a page like this after running out of credits somewhere else, usually mid-project. The paid separation services are generally competent products — the friction is the pricing model, where a handful of free songs is followed by a monthly subscription for something you needed twice.

This is the same job without that structure. It is worth being clear about what is genuinely different and what is only marketing, because both exist in this category.

What is actually different

The underlying technology is largely shared. Most separation services, paid and free, are running open models — MDX-Net, Demucs and their derivatives — because that is what the research produced and it is licensed for use. When a paid tool is genuinely better, it is usually because they have fine-tuned a model or built an ensemble, and that difference is real but smaller than the pricing implies.

The meaningful difference here is that nothing is withheld. There is no reduced-quality free tier, no watermark, no shortened preview and no better model behind a paywall. What runs for you is what runs for everyone.

The honest cost of that is a daily cap: five songs per device. It exists because GPU time is a real expense, and a cap is a more sustainable way to handle that than a subscription.

Where a paid tool may still be the better choice

High volume. Five a day is a limit; if you are processing an album or a library, a paid service with bulk allowances will be less painful.
Model choice and fine control. Some paid tools expose multiple models, ensembles or per-stem settings that are not offered here.
Features beyond separation — pitch shifting, tempo change, chord detection, click tracks. This does one thing.
A guaranteed service level. This is run at cost, with no uptime promise attached to it.

Questions

›What is the catch?

The daily cap, and an email address. The email is how the limit is tracked without payment details, and it means occasional updates unless you unsubscribe. That is the whole arrangement — no card, no trial converting into a charge.

›Is the quality really the same as a paid service?

For the same underlying models, yes. Some paid services do run fine-tuned or ensemble models that can edge ahead on difficult material — that is a genuine advantage and worth paying for if you need it. On typical songs the difference is hard to hear.

›Will this stay free?

That is the intention, and the cost structure supports it: separation runs on GPUs that only spin up while a song is processing, and no audio is stored afterwards. The daily cap exists precisely so it does not need a paywall.

›Do you keep my files?

No. Uploads are wiped the moment separation completes, and the stems disappear once their links time out. Nothing is retained afterwards, for training or anything else.

Where it struggles

If you need to separate large volumes, want to choose between models, or need the surrounding features that paid platforms bundle in, a subscription is a legitimate purchase and this will frustrate you. It does one job, five times a day, at full quality.