In-browser AI stem separation + mixing

Take your track apart. Put it back together your way.

Latent Mixer splits any AI-generated track into clean stems — drums, bass, other, vocals — right in your browser on your own GPU. Then mix them. No uploads, no minute quotas, no per-song credits. Your audio never leaves your machine.

Currently in free open testing. No account, no credit card, no catch.

What it does

Six things Latent Mixer does that the others don't — all in your browser, all on your GPU.

In-browser stem separation

htdemucs AI separation runs on your own GPU via WebGPU. No server uploads, no per-minute charges, no song credits. Your audio stays on your machine.

Pre-split or whole-track

Drop one stereo file and Latent splits it into drums, bass, other, vocals — or load 2–8 stems you already have. No separation step if you don't need it.

Per-stem mixing

Fader, pan, mute, solo, and a per-track insert chain — EQ5, glue comp, limiter — on up to 8 tracks. De-harsh on vocals when you need it.

Honesty features

The null-check badge proves your stems reconstruct the original (e.g. "−38 dB residual"). Level-matched A/B against your source, so you mix against the truth, not loudness bias.

One-click derived exports

Instrumental and acapella are exact, because mixing is linear. Export the mix, all stems, instrumental, or acapella — 24-bit WAV, fades, loudness targets included.

Built for AI music, not audio engineers

Sensible defaults on every module. Per-stem-type presets coming soon. If you've never mixed before, this is the on-ramp.

How it works

Drop a track

Or pre-split stems. MP3, WAV, FLAC, OGG — anything your browser decodes. One file and Latent splits it; multiple files and you're already mixing.

Separate and mix

Latent separates on your GPU and lays out a mixer. Tweak faders, EQ, and inserts — or trust the defaults. Up to 8 tracks, every module toggle-first.

Export

The mix, all stems, instrumental, or acapella. 24-bit WAV, fades, loudness targets. Nothing leaves your browser unless you export it.

Why Latent Mixer is different

Most stem tools either separate or mix — and almost all of them want to upload your audio to a server and charge you by the minute. Latent Mixer does both, in your browser, on your GPU, free during testing.

Private & local

Separation runs client-side on WebGPU. A server only serves static files. No uploads, no minute quotas, no per-song credits.

Separate → mix in one place

Drop a track, split it, and mix it in the same app. No exporting to a second tool just to rebalance a vocal.

Built for AI-music people

The Suno and Udio audience has largely never mixed before. Sensible defaults first, deep controls second.

Free during testing

Same zero-gating model as Latent Mastering. A URL, not an installer. No account, no credit card.

Latent Mixer is in free, open testing right now.

Try it, break it, tell us what's wrong. We read every report.