Audio workspace
Split a song into stems.
Drop a track in, get clean instrument layers out - no upload, no install.
Offline
Ready to split
Drop your song here
or
WAV, MP3, FLAC, M4A, OGG - up to ~12 min
Choose stems
Quality
Drop a track in, get clean instrument layers out - no upload, no install.
or
WAV, MP3, FLAC, M4A, OGG - up to ~12 min
Everything happens inside this browser tab. Your song is never uploaded, and no server is involved at any point.
This is classic signal processing rather than an AI model, so it stays fast and fully offline. Results are good for remixing, practice and karaoke, but less surgical than a heavyweight server-side model.
The local engine runs the real Demucs AI model. It needs a one-time setup on your own computer - after that it is unlimited, private and works offline. Your audio is never uploaded anywhere. Use the one-click installer, or set it up by hand from the repository.
install-xvviix.bat. A black window opens and does the rest.Prefer to see exactly what you are running? Clone the repository and start the engine yourself. Nothing is downloaded from this page, and you can read engine/xvviix_engine.py before running it - it is plain Python with no third-party imports.
You need: Git and Python 3.9+. The pip install step is the big one - it pulls PyTorch and Demucs. If you already have PyTorch 2.0+ in that environment, only Demucs is fetched.
When the terminal prints READY, come back to this tab. The page checks 127.0.0.1:8765 every few seconds and unlocks the local engine on its own. To stop the engine, press Ctrl+C in the terminal.
Everything installs inside a single folder - nothing is written elsewhere on your system, and no admin rights are needed. To uninstall, just delete the folder. Requires Python; the installer tells you if it is missing.
XVVIIX is analyzing the waveform and isolating each layer.