Private by design

In In-Browser mode, your audio is decoded and separated entirely on your own device — it is never uploaded anywhere. The in-browser AI model and code libraries are downloaded once from public CDNs (jsDelivr, esm.sh) and Hugging Face, then cached.

1. Copyright & your responsibility. Most recorded music is protected by copyright. You are solely responsible for ensuring you own the audio you load, or otherwise have the legal right to separate, store and use it (for example: your own recordings, properly licensed material, or use permitted by law such as fair use / fair dealing). Separating, sharing or publishing stems of copyrighted works without permission may be unlawful. Use this tool only for content you are permitted to process, and comply with all applicable laws and the rights of creators.

2. Your files stay yours. No file you load is sent to any third-party server. Closing the tab clears everything from memory.

3. No warranty. This software is provided "as is", without warranty of any kind, express or implied. Separation quality is not guaranteed. The developers assume no liability for any damages arising from its use, including loss of data. Keep your originals.

4. Indemnification. You agree to indemnify and hold harmless the developers of this tool from any claims arising from your use of it, including any infringement of third-party rights.

5. No affiliation. JB's Stem Splitter is an independent tool. It is not affiliated with, sponsored by, or endorsed by any artist, label, or the open-source projects listed below.

Credits & open-source acknowledgements

This tool is built on excellent open-source work, with thanks:

• Demucs — Meta AI Research (MIT). The htdemucs separation model used by both engines.
• demucs-web & the ONNX model export — timcsy (MIT). Runs Demucs in the browser.
• ONNX Runtime Web — Microsoft (MIT). In-browser model inference (WebGPU).
• JSZip (MIT) — “Download all stems” packaging.
• lamejs (@breezystack fork, LGPL-3.0) — MP3 export.
• Local server: PyTorch, FastAPI, Uvicorn & FFmpeg.

Full licenses and notices: see CREDITS.md in the project folder.