Neural Amp models vs Neural Captures

Are the Neural Amp models a capture of the particular amp or are they essentially a digital recreation of the circuit of that amp?
I was trying the AC 30 model from the QC and noticed the Bass/Treble controls do not function like an AC 30 amp where the Bass cuts Bass clockwise and the Treble cuts clockwise.

They sound nothing like it even if reversing the rotation.

Any feedback would be appreciated

I may be totally wrong, so maybe someone else can confirm, but my basic understanding is:

A single capture represents an amp with one specific setting, gain, treble, bass, master, channel, etc.
The controls you have on the capture do not represent the controls on the real amp, they are only there to adjust the input gain (e.g. for different guitars) or do a very basic EQ change.

If you need an AC30 amp sound with very specific treble and bass settings, you would need a separate capture. There are commercial packs that cover a wide range of settings for one single amp with 30+ captures.

There has been some information given by NDSP in different interviews and I think it’s quite safe to say that their models are not modeled in a bottom-up approach, i.e. not based on the circuit design but rather an AI model is trained to provide a similar tonal response to the physical amp. It was also said, that they’re using robots to automate the turning of knobs on the physical amps while capturing their tonal response. Some people came up with the inofficial term of ‘deep captures’ when refering to QC’s stock amp (and drive) models.

Long story short: the models generally have similar controls to the amps they’re based on and ideally the virtual controls should respond like the amp’s physical controls. That’s just my understanding, however.


@Reverb Sorry, ignore my response. I thought you were asking for the captures, not the models…