Youka ships with several AI models that separate vocals from instrumentals. Each model has trade-offs between quality, speed, and what it keeps in the backing track.
Quick comparison
| Model | Output stems | Best for |
|---|
| MDX-23C (default) | Instrumental, Lead vocals | Most karaoke projects |
| AudioShakeAI | Instrumental, Lead vocals | Highest-quality separation |
| MusicAI (Instrumental Only) | Instrumental | When you only need the backing track |
| MusicAI (Instrumental / Backing / Lead) | Instrumental, Backing vocals, Lead vocals | Full mixing control over harmonies |
| MusicAI (With Backing Vocals) | Instrumental + backing, Lead vocals | Karaoke where choirs should stay audible |
| MusicAI (Without Backing Vocals) | Instrumental, Lead vocals | Clean instrumental, no harmonies |
| MDX-Net (With Backing Vocals) | Instrumental + backing, Lead vocals | Alternative to MusicAI backing-vocal model |
| RoFormer (and variants) | Instrumental, Lead vocals | Experimental — try if defaults don’t suit the song |
| Demucs | Instrumental, Lead vocals | Open-source fallback |
Start with MDX-23C or AudioShakeAI. Switch to a With Backing
Vocals model if the song has strong harmonies you want to keep.
Pick a model when creating karaoke
Start a new project
Click Create Karaoke.
Open advanced options
Expand the advanced panel in the create dialog.
Pick a split model
Select your preferred model from the dropdown.
Create
Finish the creation flow as usual.
Try a different model on an existing project
Open the project
Open the karaoke project from the Library.
Open the Audio menu
Go to Audio > Create.
Pick the model
Click the split model you want to try.
Wait for processing
New stems appear in the Audio menu once the job completes.
