Guide to understand the difference between different models

[damiensellier]

Hi,
I really want to dive deeper into UVR but I struggle to understand the different algorithm models.
I followed different tutorials I found on youtube or reddit and on this discussion board and got good or so-so results depending on the options. But there's no such thing as a comprehensive guide about the different models and what to use for particular purpose.

Obviously models have been designed with a particular task in mind but if you read : UVR-MDX-NET Inst HQ3 that doesn't really help. Especially,what's the difference between HQ3 vs HQ4.

A wiki section on this github would be great to provide those details and what user could expect from a certain process type and model.

I'd be glad to participate in that if I could find the details somewhere.

Thanks

Damien

[MonolithFoundation]

Same question, not very clear.

[xzh2004]

Same question

[Liubomyrka]

[scryptkyddy]

UVR should have a model wiki or something.

[IC-Alchemy]

I'd recommend asking any AI about the different models and settings.

But a few quick tips:

• If a set of models has sequential numbers like HQ1, HQ2, HQ3 You want to use the highest number. These are version numbers in all cases as far as I can tell.
• VR models are the oldest models. Lower quality but run much faster on older computers.
• MDX models score the highest on tests and have the most training. Many claim MDX models are the best.
• MUCS models are the best for splitting a song into individual parts

htdemucs_6s = Vocals, Drums, Bass, Guitar, Piano, Other