How unmix. works

Three steps. No account. No upload. Your music stays on your device.

The AI model runs entirely in your browser using WebAssembly. Unlike other tools, you can start listening immediately as stems are separated — each segment becomes playable as soon as it's ready. The first time takes slightly longer as the model is downloaded and cached locally for future use.

AI ModelMeta Demucs v4 Hybrid Transformer

Processing100% in-browser (WASM)

Speed~30s for a 3min song

QualityStandard or HQ mode

PlaybackProgressive — listen while processing

Vocals

Drums

Bass

Other

What you can do

Remove Vocals

Get clean instrumentals from any song. Perfect for karaoke, remixing, or content creation.

Isolate Drums

Extract drum tracks for sampling, practice, or transcription. Hear every ghost note clearly.

Extract Bass

Isolate bass lines to learn, transcribe, or sample. Remove bass to practice your own parts.

Get Instruments

Separate guitars, keys, synths, and everything else. Solo any part of a mix.

Make Karaoke

Mute vocals, keep the instrumental. Instant karaoke from any song, any genre.

Remix & Mashup

Extract acapellas, swap drums, layer stems from different songs. Full creative control.

Under the hood

Demucs v4

Meta's Hybrid Transformer model — state-of-the-art music source separation. Trained on thousands of multi-track recordings.

WebAssembly

The AI model runs compiled to WASM in your browser at near-native speed. No server required, no data transmitted.

Web Audio API

Real-time audio playback with per-stem effects chain. Interactive latency of 5.8ms for instant response.

Offline-Ready

Install as a PWA. After the first model download, everything works offline. Your songs never touch a network.

FAQ

Is it really free?

Yes. unmix. is free with no limits, no watermarks, and no account required. The AI model runs in your browser.

Is my audio uploaded anywhere?

No. Your audio file never leaves your device. All processing happens locally using WebAssembly. We don't have servers to receive audio even if we wanted to.

What quality can I expect?

unmix. uses Demucs v4, the same model used by professional studios. HQ mode provides the best results at the cost of slightly longer processing time.

How long does processing take?

About 30 seconds for a 3-minute song on a modern computer. The first run downloads the AI model (~80MB), which is then cached for future use.

Does it work on mobile?

Yes, but processing is slower on mobile devices. For best results, use a desktop browser. Chrome is recommended for optimal performance.

What audio formats are supported?

MP3, WAV, FLAC, OGG, and M4A. The output is always lossless WAV for maximum quality.

Try it now

Drop a file and hear the difference.

▶ Open the App