Extracting vocal and instrumental stems from a Suno-generated track

[momen3azaab-ux]

I'm trying to extract vocal and instrumental stems from a track generated with Suno, but the results have been pretty disappointing so far.

I've already tried an ensemble using:

• UVR-MDX-NET: INST3
• UVR-MDX-NET: INST HQ1
• 7-HP2-UVR
• Demucs v4 FT

Unfortunately, I'm still getting noticeable artifacts and poor separation quality.

Has anyone had better results with other models or model combinations, particularly when working with AI-generated tracks like those from Suno? I'd be interested to hear what settings, workflows, or ensembles have worked well for you.

[GianniBK]

In case you didn't know, Suno now offers an official stem separation option that costs 50 credits.

[momen3azaab-ux]

The stems splitting method in suno has a lot of artifacts and its not practical for production thats why am trying to find a strong alternative

[GianniBK]

Got it. In that case, you should step away from UVR5 for a moment and try a professional restoration approach.I highly recommend testing the new iZotope RX 12 (Standalone version) using its Music Rebalance module. It handles digital artifacts and AI-generated stems much better than standard open-source models.Alternatively, you can achieve excellent results by hosting the Music Rebalance plugin directly inside a DAW like Mixcraft. Processing it this way allows you to tweak the separation in real-time and apply instant EQ or gating to clean up the leftovers. Give this workflow a shot!