The Neural Amp Modeler - NAM - has arrived!

What about the DuoX ? Any plans to release a Dwarf 2.0 with DuoX’s CPU power anytime soon? Or simply to re-release the DuoX?


The company needs to be in a completely different situation to even consider developing any new products.

If you try the commercial AIDA-X plugins it might change your opinion on the quality of LSTM models.
Or the newly released “The Rude”.

The WaveNet approach gets better quality by bruteforce, which needs a lot of CPU power to run.


That is highly debatable :slight_smile:

you can test the models I have published at ToneHunt against their AIDA-X counterpart. It is really tough to spot differences.


Really sad to hear. More computing power could be really worthy considering neural plugins.
Even with MDX I cannot afford using neural plugins with RMPro in the same time.

Actually, I should have said ‘good’ at least (‘ok’ is unfair). I find AidaX’s free captures to be a step lower compared to the (great) captures desktop NAM (there are also plenty of bad ones on NAM), but definitely a step up compared to the amp modellers I have tested.
That said, I only judge what one gets with the hardware purchase or for free, as I am not, personally, willing to pay for pre-amp captures, IRs or plug-ins. (just my choice not to enter that rabbit hole).
I’m not a pro musician, and I don’t play live. So my Dwarf would be plugged to the computer most of the time anyways. Given that, and the considerations about Dwarf’s limited CPU power, I’m not sure what would be the point for me to use NAM’s plugin on Dwarf anyways (but I appreciate a lot the efforts put to develop the plugin)… I mean, yes, there would be: to put effects plugins in between the amp and the cab in the virtual pedalboard… but does it matter so much, in an digital signal processing chain to stick the time-based effect after the pre-amp instead of before? If any difference, I can barely hear it in a simple chain but as soon as it becomes slightly more complex… well, I think the impact is negligible.
If so, the DuoX with NAM plugin and full sized Wavenet tone capture files could become perfection… But I understand how much you have on your plate right now, and why it’s not the priority.
In any cases, thank you for your work, Mod really rocks.

1 Like

A re-issue of the discontinued DuoX could do as well, as it would not require any new development investment, allowing for it to run smoothly with a decent sized effect chain on top of a NAM nanocapture (or even maybe a full sized NAM capture, allowing full access to the large amount of existing capture n ToneHub, without the need for retraining…)
The Midi port already allows to add an extra control stomp for players who’d like to, so it’s all good, there would not even be a need to repackage the DuoX guts into a Dwarf body…

Edit: or obviously, one could also put 2 Dwarf one after the other… the first dealing with dynamics/filter/gain/mod effects + NAM tone capture, and the second dealing with (CPU hungry) time-based effect and cab IR… But it has a cost, and again, the DuoX would solve the need for two units, since it’s roughly twice as powerful as a Dwarf. (I might go for a used DuoX at this point… :thinking:)

Yes indeed, I should have said ‘good’, as ‘ok’ is unfair.
But the fact that one needs to spend extra money to find great AidaX profiles (except for yours) makes a strong point for NAM anyways (the full desktop app).
In order to increase AidaX’s value (and Dwarf’s, since AidaX is less ressource-hungry compared to nano-NAM), it might be good to provide a larger panel of free, high quality trained LSTM models, out-of-the-box. Just my 2 cents.

In my brief experience with NAM, Aida etc is that sure, you need a nicely captured model (@lordsilly kraken vx captures are reaaally good and I do use the same amp miced so I can compare), but the thing that matters most is the IR, I’ve spent 15eur on a really good IR and I’m set forever.



I could not agree more.

I’ve recently acquired some IRs from chopTones and I’ve been impressed with how much it can improve the sound.


Nice, I bought York Audio IRs, there’s so many combinations and everything to choose from but it’s not strange at all that I like SM57+R121 the most, same as in real life!


I’m busy reworking a plugin from PiPedal that allows you to use 3 IR cab files and blend them together. Is this something that’s of interest?


Short answer: yes.

Longer answer: it depends. If it is more resource friendly than running multiple instances of the IR convolver and a mixer after them, then absolutely!

It utilises about the same as the IR Cab loader🙌🏼
I might call on some of you MODGUI wizards to help in that regard…my html/css skills are not the greatest…


have you tried the MOD SDK for a starting point?

I have, but the screenshots refuse to generate…seems that the bug is still persistent in that regard

I’have Mod Duo…How can i upload the .nam/ir? From file manager? Why don’t have a NAM folder?

I talked to Steve Ack and asked him, how the implementation of NAM in the Dwarf could be in the future. Here is his answer:"Hi, Michael. Glad to hear it!

I’m not working on MOD’s stuff in particular, so I can’t really say, unfortunately. I’m prioritizing my attention on my own repos since that’s the best way for NAM to be able to have a positive impact more broadly.

I am willing to do some targeted help for specific companies if we can find terms that work, but that hasn’t happened with MOD I’m afraid."

1 Like

Yep. We had a chat but I think we did not end it in the same page :frowning:

It is complex because when most people ask “Does the MOD Dwarf runs NAM?”, what they are actually asking is “Can I run those thousands of models from ToneHunt in the MOD Dwarf?”

But the fact is that the desktop CPUs to which most, if not all of these models, are meant to run are dozens of times more powerful than embedded CPUs like the Dwarf or the RPis. Even in the mighty Duo X the standard NAM models will not run properly.

That is not an issue with the embedded CPUs that are deffective or the engine of NAM that is not efficient enough, but the bare fact that the models are just too heavy.

There is no magic. It we want to run NAM models in the Dwarf, they need to be more light and efficient, so the effort here is not in making NAM itself better, but changing the training AND making it possible to find such models.

We have run multiple attempts already - and even worked with other people from the NAM project - to get to help getting something that is at least acceptable, and I believe that somehow led to the creation of the “nano” model weight in their training system. Later we put a bit more of effort and added a “buffered” switch to the NAM plugin that adds the functionality of the Portal Plugin inside NAM, allowing it to run using less than 20% of the CPU with nano models.

Once the “nano” model was available, we promply trained our “House of Music” collection - about 20 models from a Fender Deluxe, Marshall JVM410, Orange Rockerverb 50 and Fender Twin - using NAM nano model weights and posted them in Tonehunt. That was about 6 months ago. At the time the tag system from Tonehunt was broken and thus, the models were not easy to find. At the time I posted the issue to the ToneHunt team. Six months later, the issue persists…

So, from my side, there is not a reason to write a single line of code before speaking about model management. After so many years doing this, it raises me multiple red flags when a developer jumps straight to coding solutions and forecasts of billable hours without even bothering to understand what the issue really is.

I offered to send a unit for free, no strings attached, just so that, at least, the desired experience could be understood. But I’ve got a negative. So that’s where we’re at :slight_smile:


Thanks for your honest thoughts. At the moment I am happy with AIDA and hoping to get a Silver Jubilee model.

1 Like