[ RUNS LOCALLY ]

The separation pipeline runs on local hardware.

StemFoundry splits a track with Demucs inside a Celery worker, on Apple Silicon. A separation holds roughly two gigabytes of memory and takes one to two minutes per track, so the worker, the Redis queue and the API run on a machine rather than in a serverless function. This deployment is the marketing surface only, which is why the upload and dashboard screens are not reachable here.

Everything on the landing page is real: each waveform is computed from decoded audio by the same code that runs in the pipeline, not drawn for effect.