AI Source Separation
Split a mixed audio track into vocals and instruments. Demucs Small, ~5–15 s per 30 s clip.
| Mode | Speed | Quality | Best for |
|---|---|---|---|
| Demucs | Slower (heavier model) | Splits vocals / instruments | Karaoke, remixing, isolation |
AI Source Separation needs the standard AI tier (0.2 GB model). You’re currently opted into none.
Quick start
See all 2 presets
Drop audio files here — add as many as you like
or click to browse · processed one at a time · output audio
Tier opt-in required before this capability runs.
Want this capability inside the main editor with layers, history, and the full AI panel?
How it works
- 1
Add your audio
Drop or select the audio you want to process — it stays on your device.
- 2
Run the model in-browser
AI Source Separation loads its model (~200 MB) once, caches it, then runs locally in a worker. No upload.
- 3
Download the audio
Preview the result and download the audio. Re-run with different settings anytime.
Common use cases
Why it’s different
100% Private
Every model runs in your browser. Your files never leave your device — nothing is uploaded to a server.
True Alpha Channel
Exports preserve a real straight-alpha transparency channel (PNG / WebP / AVIF), not a baked-on background.
Free Forever
No account, no watermark, no credits. Open the tool and use it.
Works Offline
After the model downloads once it is cached, so the tool keeps working with no connection.
FAQ
How clean is the split?
Demucs gives strong separation; some bleed can remain on complex mixes.
Is it free to use?
Yes — AI Source Separation is completely free. No account, no watermark, no credits, and no usage limits.
Do my files or prompts ever leave my device?
No. Everything runs locally in your browser via WebAssembly/WebGPU — there is no server that receives your files, prompts, or results.
Which browser and hardware do I need?
A modern browser. Chrome and Edge get WebGPU acceleration for the fastest results; Firefox and Safari run via WebAssembly. The model (~200 MB) downloads once, then is cached for offline use.
Can I use the results commercially?
Yes. You own everything you create — NSS makes no claim to the images, videos, or text you process or export.
Does it work on mobile?
Lightweight tools run on phones; heavier models prefer a desktop with a GPU. The tool picks the best path for your device and falls back gracefully where needed.
Where can I see a step-by-step guide?
Yes — there is a full walkthrough at /how-it-works/ai-source-separate.
Ready to try AI Source Separation?
Free, private, no signup — runs right in your browser.