Four clean stems
Vocals, drums, bass and other, each as its own lossless WAV, lined up sample for sample so they drop straight into a DAW.
Drop in a song and get vocals, drums, bass and everything else as separate files. Made for remixes, acapellas, karaoke and sampling, and it runs on your own computer, so nothing is uploaded and there are no per-song credits.
Drop a song. Get four files back.
The same kind of model the paid stem sites run, on your own machine and your own GPU.
Vocals, drums, bass and other, each as its own lossless WAV, lined up sample for sample so they drop straight into a DAW.
Demucs (htdemucs) runs through ONNX Runtime, on the Neural Engine and GPU with CoreML on a Mac and DirectML on Windows. No Python to install, no cloud queue.
Long mixes and DJ sets are split in overlapping chunks and blended back together, so memory stays flat and there are no seams.
MP3, WAV, FLAC, M4A and the audio from video files. Drop a whole folder to split an album.
Script it, batch it, put it in a pipeline.
The defaults are the good ones. Open the advanced panel only if you want to.
Drag a file or folder onto the window, or use the CLI.
The model runs locally, chunk by chunk, using your GPU when there is one.
Four WAVs land in a folder next to the song, named and ready for your DAW.
Stm never uploads your music. The model ships with the app and runs offline, so unreleased tracks stay unreleased.
No sign-up, no email. Open it and go.
The model runs on your machine. Your songs never leave it.
No analytics, no crash reporting, no ads. Ever.
The latest version. It checks for updates and installs them with one click.The first release is on its way. Watch the GitHub repo or get it through PXL when it lands.
Unzip and move Stm to Applications. The first time, right-click it and choose Open.
Run it. If SmartScreen appears, click More info, then Run anyway.