1. Upload audio
No audio selected yet.
2. Choose what to separate
Select one or more outputs. You can change this selection anytime.
AI model not loaded yet.
Tap “Download / load AI model” in the Free AI section below. The first download needs internet and may take several minutes.
Tap “Download / load AI model” in the Free AI section below. The first download needs internet and may take several minutes.
Important: detailed parts such as rhythm guitar, acoustic guitar, and piano/keyboard need a suitable multi-instrument model. Free open-source models may not isolate every part cleanly.
Free AI backend connection
Experimental on-device AI: loads an open-source Demucs ONNX model into your browser. Your song is processed on your phone; it is not uploaded to an AI server. The first model download is about 136 MB and needs internet.
First tap “Check browser & model access.” Open this page from HTTPS hosting, not from Downloads/content://. The model is large and may exceed available phone memory even if network checks pass.
Not connected.
Choose “Test connection” after you have started a compatible local backend.
Choose “Test connection” after you have started a compatible local backend.