New Script: Demucs Stem Splitter for Vegas Pro (AI Audio Separation)

kjetil wrote on 8/17/2025, 11:57 AM

Hi everyone,

I’ve been working on a script for Vegas Pro that integrates Demucs to separate audio tracks into stems (vocals, drums, bass, guitar, piano, etc.) and then automatically imports them as separate audio tracks inside Vegas.

I’m now sharing my current build in case anyone wants to test it, play around with it.

What’s included

demucs_cli.exe (required AI engine, ~5.6 GB unpacked – zipped download provided)
Vegas Pro .dll script file
Configuration .ini file
Demo screenshot (JPG)
Demo video (MP4)
install.txt with simple installation instructions

How to use

Download and unzip the package (link below).
Follow the steps in install.txt to copy the files into your Vegas Pro script folder.
Run the script on an audio event – Demucs will process it and bring the stems back into your Vegas project as new tracks.

Download

👉 Google Drive link to package: https://drive.google.com/drive/folders/1Iw3w6y_2MVOLo5zN6YYIYvdL6W2xq_GX?usp=sharing

(Please note: the AI engine is large – around 5.6 GB when unpacked.)

Demo

Here’s a short preview video showing how it works:

Here is a screenshot of the menu showing options

Important

This is an early build and provided as-is.
Tested in Vegas Pro 21/22 (should also work in other recent versions).
Performance depends on your system (GPU recommended).
No warranties – use at your own risk.

Thanks,
Kjetil

Back to post

Comments

JimmyH wrote on 10/16/2025, 8:55 AM

@kjetil Just an update on my experience using this one

The paper from INRIA says best results are had when the audio file is up-sampled prior to Demucs and then the output down-sampled to match the original. Is that correct? I cannot tell with the stuff I've tested although maybe it is better and I'm just not able to hear. Also, it says that best results are had when no audio fx are applied prior to Demucs although I think this is not always true as I have had some improved results particularly with bass guitar stems. Sometimes I render after applying Resonant Filter and then do the splitting on that new .wav. Final note is that the 4-stem mode is usually the best for the music I've tested but the 1-stem-plus-accompaniment always matches or exceeds the splitting quality. 6-stem mode is great but output sometimes includes sharp peaks which I try to compress.