Fractal vs Capture

NOTE: This post isn't meant to speak negatively of anyone's product. It's simply meant to show different attributes found in two types of modeling.



All this talk about amp captures lately has made me curious about a couple things. Even though amps, models, and captures are all supposed to sound the same, Fractal has always had a "feel" that I haven't been able to get out of other modelers. People have said the Quad Cortex models aren't great, but the captures are amazing... so I (literally) dusted off my Quad Cortex and decided to make a capture of my '77 Marshall JMP to see how it stacked up.

Tonally, it's way better than the Kemper, but still has some excited frequencies. Not exactly apples to apples, but more like an apple vs a different kind of apple if that makes sense. It's not bad by any means... just a little different. I remember Cliff saying something a while back about hearing the quality of the modeling by rolling off the guitar's volume knob and comparing it to what a real amp does, so that's what I did.

It seems like captures can "sound" close when the guitar is on 10, but they still don't "feel" close in my opinion, so I played something on my looper pedal with the guitar's volume really low with a swell to 10 at the end and recorded it through my Marshall JMP with the LB-2 load box, the Fractal Brit 800 model with the LB-2 impedance curve, and a QC capture I made of the JMP through the LB-2. Disclaimer... I'm recovering from Covid, so my ears are still plugged up which made it a little tricky to match the Fractal with my amp by ear, but today it isn't about a direct "tone match," it's about what makes a digital recreation "feel" right or not. The capture was not changed or altered in any way, and I even captured it twice to make sure my results were consistent.

I'm going to do this in two parts. Part 1 is "Full Volume" where we hear all three examples with the guitar's volume all the way up. There's similar tonality and gain structure. They aren't identical, but that's what we're listening for here. The capture is a bit of an outlier, but it's still in the ballpark and would probably work well for some players. I notice the chord bloom with the amp and Fractal, but it's much more linear with the QC. This is when things should sound the closest.


Order: Marshall, Fractal, Capture



Part 2 is "Volume Rolloff" where we hear all three examples with the guitar's volume knob rolled off and turned up at the end. This is where the magic happens in my opinion. Even lightly picked with the volume rolled off, you hear this angry volatility in the amp like it can't wait to explode. The Fractal does a great job of replicating it; like you can hear the components in the amp working with and against each other as if it's trying to figure out what going to happen next. When we hear the capture, all of that chaos is gone. It's soft, polite, and doesn't really have any of those swirling elements we hear in the other two. I think this is why models and captures feel so different.


Order: Marshall, Fractal, Capture



All that to say, I think the crazy nonlinear behavior we hear in the Volume Rolloff clip ends up being the stuff that makes a model respond more accurately to our playing and what gives us an authentic playing experience. We may not hear it with our guitars on 10, but we can definitely feel it.


Try ToneX, its has the feel the QC can't seem to capture well.
 
So I made a similar comparison between the FM9, NAM and ToneX. Recorded a DI track with some random chugging and guitar volume roll-offs, then reamped it thru all three and here are the results:

FM9


NAM


TONEX


FM9-NAM null test


FM9-TONEX null test


It seems both NAM and ToneX can nail the dynamic behaviour of the FM9, but while NAM nails everything pretty much perfectly, ToneX still sounds a bit different especially on the low and high ends.
Also, it's been a nightmare to precisely match the gain amount on ToneX, for some reason it's not able to match it by itself while NAM does it with no effort.
The accuracy of NAM is quite evident even looking at the waveforms, it looks almost identical to the FM9 clip while ToneX shows a lot of differences.

FM9-NAM-TONEX.jpg
 
Last edited:
So I made a similar comparison between the FM9, NAM and ToneX. Recorded a DI track with some random chugging and guitar volume roll-offs, then reamped it thru all three and here are the results:

FM9


NAM


TONEX


FM9-NAM null test


FM9-TONEX null test


It seems both NAM and ToneX can nail the dynamic behaviour of the FM9, but while NAM nails everything pretty much perfectly, ToneX still sounds a bit different especially on the low and high ends.
Also, it's been a nightmare to precisely match the gain amount on ToneX, for some reason it's not able to match it by itself while NAM does it with no effort.
The accuracy of NAM is quite evident even looking at the waveforms, it looks pretty much identical to the FM9 clip while ToneX shows a lot of differences.

View attachment 117043


Nice work! This mirrors my findings - and I agreee - getting the tonex gain right is somewhat of a mystery. Almost none of my captures come out with enough gain. Accuracy wise I've been able to get pretty good with tonex too, but NAM is still better. The other challenge naturally, is to compare all 3 to a real reference amp (eq matching FM9 for fairness if needed).
 
So I made a similar comparison between the FM9, NAM and ToneX. Recorded a DI track with some random chugging and guitar volume roll-offs, then reamped it thru all three and here are the results:

FM9


NAM


TONEX


FM9-NAM null test


FM9-TONEX null test


It seems both NAM and ToneX can nail the dynamic behaviour of the FM9, but while NAM nails everything pretty much perfectly, ToneX still sounds a bit different especially on the low and high ends.
Also, it's been a nightmare to precisely match the gain amount on ToneX, for some reason it's not able to match it by itself while NAM does it with no effort.
The accuracy of NAM is quite evident even looking at the waveforms, it looks almost identical to the FM9 clip while ToneX shows a lot of differences.

View attachment 117043

Pay attention starting at 0:34 in those clips to hear how they respond to dynamics. In this case, you want that portion to “pulse” with dissonant notes since that’s what real amps do when the gain is high and you roll off your volume knob. You can really hear it in the FM9-NAM Null test. It would be great to try this comparison with a real amp in the mix as a baseline to know what we’re listening for.
 
Pay attention starting at 0:34 in those clips to hear how they respond to dynamics. In this case, you want that portion to “pulse” with dissonant notes since that’s what real amps do when the gain is high and you roll off your volume knob. You can really hear it in the FM9-NAM Null test. It would be great to try this comparison with a real amp in the mix as a baseline to know what we’re listening for.
Good ear! Actually even looking at the waveforms that part is where there are more differences between the fm9 and nam (nothing as staggering as those of tonex though). So not 100% accurate but still pretty impressive imho.
Imagine what an hybrid approach could yield: a base model based on one of Fractal amp models, but with a neural network able to tweak all parameters of the model during the training to match as close as possible a real amp, after the user chooses the same or the closest amp model available. (I think Cliff has a patent about something similar but based on classic non-AI profiling)

PS: here's the patent I was talking about: https://patents.google.com/patent/US20200279546A1/en?inventor=clifford+chase
 
Last edited:
I've read ToneX is not very good with a power amp and real cab? Any experiences?

Fractal would seem to be way better in that regard, too.

ToneX has not cracked the code on removing cabinets that are baked into captures. DI captures without cab show some promise. It is cheap enough that it is a good extra thing to have around- amp-in-a-box, distortion pedal, EQ pedal, IR pedal, DSP Expander, etc...

I will never be able to trade my Fractal for one though- Axe-FX3 is like a guitar tone workstation wonderland. I go to the Axe-FX 3 to make real tones and enjoy a complete rig.
 
Last edited:
I’m convinced enough with the ToneX pedal I’m selling my FM3 (already sold AxeIII) and also some louder tube amps such as the SV20 (awesome amp I just bought recently ironically enough) and various attenuators and load boxes.

ToneX (with the right capture) gives me the amp sound I’m after and does it in a small and cheap form factor. Just can’t really justify playing a loud amp into a load box and an IR box just to use it at home.

More so, it works with pedals better than anything I’ve owned as far as modelers go. Big reason I went back to some real amps is so I can hit them with pedals. Dime a fuzz face, super hot boost pedal etc sound great hitting the front end of a tube amp, but don’t work so well into the Axe. Some pedals can still sound good, don’t get me wrong, but often I’m really running next to no output level or the input clips.

I have no idea why it’s so good, or what’s different, but I can crank a drive pedal, fuzz etc into the ToneX pedal and it doesn’t clip, doesn’t turn to muddy farts, and has all the cleanup you’d expect. Take a plexi model, back the gain off some, hit it a cranked Fulltone ranger treble booster and that amp tone roars, back off the volume pot and it cleans up proper. Do the same on the FM3 and you need to turn the level way down on the pedal. For some reason the ToneX pedal itself, or the capture, I don’t know really, responds fantastic to pedals, and for that reason above anything else, it’s what I’m moving forward with.

All the options these days can sound great, are dynamic, et al., but for how good the ToneX can sound at its price point is pretty awesome that it can even be in the discussion compared to FAS gear.

Now for people who need tons of options, complex and complete rigs, it’s not going to cut it, but for me, I’m pretty simple. I’ve got a matchless capture, a plexi and a mt15 so clean, crunch rock and high gain and I’m honestly set. Run into a UA golden reverb, stick a few fuzz and ODs in front, and I’m really happy when my tones.
 
ToneX has not cracked the code on removing cabinets that are baked into captures. DI captures without cab show some promise. It is cheap enough that it is a good extra thing to have around- amp-in-a-box, distortion pedal, EQ pedal, IR pedal, DSP Expander, etc...

I will never be able to trade my Fractal for one though- Axe-FX3 is like a guitar tone workstation wonderland. I go to the Axe-FX 3 to make real tones and enjoy a complete rig.

I don't actually believe there is a way to do it accurately - kemper had the same problem. There's simply no way to infer which parts of the sound come from where. However, if you just make amp only captures, it doesn't really matter. Kemper had an annoying issue where simply adding an "ir" to any profile would change the DI tone of the profile to your cab - but I don't think tonex couples them like this.
 
Good ear! Actually even looking at the waveforms that part is where there are more differences between the fm9 and nam (nothing as staggering as those of tonex though). So not 100% accurate but still pretty impressive imho.
Imagine what an hybrid approach could yield: a base model based on one of Fractal amp models, but with a neural network able to tweak all parameters of the model during the training to match as close as possible a real amp, after the user chooses the same or the closest amp model available. (I think Cliff has a patent about something similar but based on classic non-AI profiling)

PS: here's the patent I was talking about: https://patents.google.com/patent/US20200279546A1/en?inventor=clifford+chase
The hybrid idea sounds awesome! Obviously I have no idea how to do it or if it could even be done, but choosing an amp model that’s close to your own amp, having the Fractal send a signal through the real amp, and have you repeatedly sweep each amp knob from zero-10 as it learns the range and effect of each control. It’s probably not possible at the moment, but it would be cool.
 
Small update for those who would actually care, but I think I’m going to keep the FM3, adds a lot of utility for the price, and having the ToneX pedal and the FM3 gives the ability to run dual amp rigs, the FM3 can control the ToneX pedal via midi, you get lots of cool FAS reverbs and plex block, rotary etc, and can of course do cool things like have a foot controller assign rate and mix to both increase on the rotary block for example.

Running the ToneX in the loop and changing screens between loop and amp block they are both exceptional sounding, the FM3 is just like a Swiss Army knife of routing options, amp and effects and cabs that it doesn’t make sense to sell it and net like $800-850 or whatever they go for these days. Just buying an active bypass loop box and a rotary pedal would run be nearly the same price. On top of that, the ToneX takes pedals great, can serve as a second amp block or work as an IR loader box, or give models the FM3 doesn’t have yet, so I think I’ll be enjoying the ride with both units for a while, so you FAS forum guys haven’t seen the last of me yet lol.
 
Small update for those who would actually care, but I think I’m going to keep the FM3, adds a lot of utility for the price, and having the ToneX pedal and the FM3 gives the ability to run dual amp rigs, the FM3 can control the ToneX pedal via midi, you get lots of cool FAS reverbs and plex block, rotary etc, and can of course do cool things like have a foot controller assign rate and mix to both increase on the rotary block for example.

Running the ToneX in the loop and changing screens between loop and amp block they are both exceptional sounding, the FM3 is just like a Swiss Army knife of routing options, amp and effects and cabs that it doesn’t make sense to sell it and net like $800-850 or whatever they go for these days. Just buying an active bypass loop box and a rotary pedal would run be nearly the same price. On top of that, the ToneX takes pedals great, can serve as a second amp block or work as an IR loader box, or give models the FM3 doesn’t have yet, so I think I’ll be enjoying the ride with both units for a while, so you FAS forum guys haven’t seen the last of me yet lol.
Have you tried running dual amps simultaneously with both units? I'd expect latency would cause phase issues...
 
Have you tried running dual amps simultaneously with both units? I'd expect latency would cause phase issues...

I have and haven’t noticed significant issues, sometimes run two two different external amps as well, so the physical placement of the two amps can also have some various delay and phasing based upon the listening position

I personally don’t think some phasing is that bad of thing in many cases with regards to guitar tones.

I love the phasing/comb filtering you can get when your using multiple mics on a cab, it can sound worse, but sometimes it adds some cool character to the sound and some added realism as we can eliminate it in the digital world but with real cabs often experience it to some degree.

Could probably account for it with a delay block set super short if really an issue, though no idea what actual added latency is produced and if it’s static or not.
 
Have you tried running dual amps simultaneously with both units? I'd expect latency would cause phase issues...

It caused a lot of phase issues for me, but you can micro adjust with a delay if you really want to do it. It's a bout 3-4 total extra ms using it in a loop (2-3 for TX and 1 for loop ada)
 
Wow, this experiment with the volume rolled down is revealing. I couldn’t quite articulate why the fractal always felt better than the other modelers, but it’s very clear how the response differs here.
Yes, the older technology modelers just don't react correctly when compared to a good tube amp, heck bad tube amps don't react correctly either but at least the tubes make up for some of it.

I'll spend all night on one "crunch" scene using the guitar's volume control to go from clean to mean, and being able to do that with the Fractals is what sold me on them. After the updates to the firmware since the FX3 was released we've seen that behavior improve even more.
 
It would be great to try this comparison with a real amp in the mix as a baseline to know what we’re listening for.
Agreed, but it needs to be one of Fractal's reference amps so there's a one-to-one comparison between the amp and the Fractal, then profile that same amp for the other two.
 
Agreed, but it needs to be one of Fractal's reference amps so there's a one-to-one comparison between the amp and the Fractal, then profile that same amp for the other two.

If you mean the actual reference amp, I disagree - that doesn't really present an end user scenario. If you mean the same make/model/wattage/etc then yes, absolutely - and EQ matched / tweaked as close as possible.
 
If you mean the actual reference amp, I disagree - that doesn't really present an end user scenario. If you mean the same make/model/wattage/etc then yes, absolutely - and EQ matched / tweaked as close as possible.
That leaves the door open for all sorts of variance.
 
Someone on TGF in the Helix 3.6 thread wrote - “Add a new block to play Steve Ack’s NAM Neural Amp Modeler captures! Is it even possible?” NAM is under the MIT licence so it poses no problem in terms of code exposure. And the cpu demands would be light so having multiple instances in a preset should be no problem.
 
Someone on TGF in the Helix 3.6 thread wrote - “Add a new block to play Steve Ack’s NAM Neural Amp Modeler captures! Is it even possible?” NAM is under the MIT licence so it poses no problem in terms of code exposure. And the cpu demands would be light so having multiple instances in a preset should be no problem.

Would be amazing but I don't see it happening. I'm not sure how light the CPU demands would be on an embedded platform though
 
Would be amazing but I don't see it happening. I'm not sure how light the CPU demands would be on an embedded platform though
The NAM plugin just got a performance improvement so I’d think there’s generally room for optimization. And the open source community is now looking at running the plugin on a Raspberry Pi. It’s still early days for the technology. I guess the real question in terms of performance in an embedded environment is whether the chipset architecture is appropriate for running the neural network.

Anyways, it’s an interesting proposition and if I was the technical lead I’d have a good long hard look at it because the implications are pretty far reaching.
 
Last edited:
Back
Top Bottom