Amp Modeling, Profiling and Neural Capture Explained

Digital guitar rigs now use a vocabulary that often sounds more divided than the technology really is. One platform is called a modeler, another creates profiles, another makes captures or Tone Models, and newer systems add terms such as neural capture, machine learning and cloning.

The useful distinction is between methods, not brands. A traditional amp model is designed as an adjustable representation of amplifier behaviour. A profile or capture starts with a real device or rig in a particular configuration and derives a digital representation from what that setup does to an incoming signal. Some capture systems use neural networks to learn that input-to-output relationship.

Those methods increasingly overlap. Line 6 Helix Stadium, for example, contains conventional amp models and also supports PROXY Cloning of physical amps, preamps, amp-and-cab rigs and drive pedals. ^S2S3 The old shorthand — “modelers model amps, profilers take snapshots” — is therefore too simple for the current landscape.

Traditional modeling starts with the idea that an amplifier can be represented as an adjustable system whose behaviour changes as you alter its controls and feed different signals into it.

The exact implementation is proprietary and differs between companies, but the important point for the player is that the controls belong to the model rather than being added only after a captured tone. Line 6 describes its Amp Models as including the preamp, tone stack and power-amp components. ^S2 Fractal Audio describes its Amp block in still more component-level terms, including preamp tubes, tone stack, cathode follower, power supply, power amp, phase inverter and amp/speaker interaction. ^S5

That makes modeling useful when you want to move around a virtual amplifier: change gain, master level or tone controls and keep shaping the amp without making a new capture of physical hardware. It does not mean every model reproduces every possible state of the original amplifier equally; it means the platform was designed around an adjustable amp representation rather than a user-created capture session.

For current coverage of this part of the guitar world, Guitar Radar follows developments on its Amp Modeling and Guitar Modeling topic pages.

Profiling and capture reverse the workflow. Instead of beginning with a pre-built virtual amp, the system begins with a real target — an amplifier, pedal, preamp, cabinet or complete rig — and derives a representation from its measured or recorded response.

The phrase that matters is “the rig as configured during capture”. Kemper says traditional PROFILING analyzes the complete signal chain of a real amplifier setup; a Studio PROFILE can include the amp, speaker and microphone chain. Kemper’s Liquid Profiling then adds modeled gain structure and tone-stack behaviour, and Kemper says its Gain and EQ controls respond like those on the original amplifier. That is Kemper’s claim about its own system, not a universal property of profiles. ^S1

What a capture contains also depends on how it is connected. Line 6 PROXY can create an Amp+Cab Clone with the cabinet, speaker and microphone in the captured chain, or an Amp Clone through a load box with no cabinet included. ^S3 So “capture” does not automatically mean either amp-only or amp-plus-cab.

Line 6 is a useful example of why brand-based categories no longer work. Helix Stadium contains adjustable Amp Models, but its PROXY engine also sends test signals through a physical target, uploads the collected data to Line 6 servers and returns a Clone. The Clone then has its own Gain and EQ controls. ^S3

One device can therefore be both a modeler and a capture platform. The method matters more than the logo on the enclosure.

Machine-learning capture is another way to derive that digital representation. The system is given an input signal and the processed output from the real equipment, then trained to approximate the relationship between the two.

Neural DSP’s Neural Capture™ is one proprietary implementation. The Quad Cortex manual says Version 1 sends recorded test signals through the target and trains a neural network on the device to reproduce its tone and dynamic response; it also documents captures of a miked cabinet or an amplifier connected through a reactive load box. Version 2 moves the training to Cortex Cloud. ^S6 The same manual recommends making a new capture with more gain on the original device when a substantially higher-gain result is needed, rather than relying on an extreme post-capture gain adjustment. That is a useful reminder that editability is not the same thing as reconstructing every position of the original amp’s controls. Guitar Radar covered the platform’s latest software in its CorOS 4.1.0 story.

IK Multimedia’s TONEX uses another neural-network workflow. IK says a prepared dry guitar or bass capture track and the processed signal are compared by its deep neural network to create a Tone Model. IK also says TONEX can model an amp and cabinet together and then virtually separate the cabinet afterwards. ^S7S8 That separation is a TONEX-specific capability claim, not a rule for neural captures in general. The newest hardware built around Tone Models is covered in Guitar Radar’s TONEX Board story.

The open-source Neural Amp Modeler (NAM) makes the training relationship especially visible: a standard workflow uses paired input and reamped output recordings to train a .nam model, while its full trainer can use different model architectures such as WaveNet. ^S9S10S11 The official NAM plugin also implements a separate IR loader and a generic Bass/Middle/Treble tone stack around the loaded model. ^S12 For how community captures are now shared and used, see Guitar Radar’s TONE3000 NAM plugin story.

“Neural” therefore describes a family of methods, not one universal file format or one fixed set of controls.

Impulse responses are often discussed alongside modelers and captures, but they do a different job.

In guitar rigs, IRs are commonly used to reproduce the measured response of a speaker, cabinet and microphone chain. Line 6 places Cab/IR processing after the amp stage and adds a matching Cab block by default with an Amp block, while a real guitar cabinet makes a virtual Cab block redundant. ^S4 Fractal Audio likewise separates its Amp and Cab blocks; its Cab block uses IRs for speaker and cabinet reproduction and can add mic, room and related processing. ^S5

An IR should therefore not be treated as the same thing as modeling the amplifier’s nonlinear behaviour. Some captures already include a miked cabinet; others capture the amp directly and use a separate cab model or IR afterwards. The next guide, Cab Sims, IRs and FRFR, follows the signal from the cabinet stage to the speaker.

The technical distinction becomes practical in four places.

Editability: a traditional amp model is usually the more natural starting point when you want to treat the virtual amp like an amp and move its controls over a wide range. Captures can also be editable, but their added controls should not automatically be read as exact reconstructions of every original knob position.

Portability: profiling and capture are useful when the important thing is a particular physical setup — your amplifier, pedal chain or a carefully dialed-in studio tone — that you want to store and recall.

Cabinet choice: if a capture already contains a miked cabinet, adding another cab IR can duplicate that part of the chain. With a direct amp capture, a cab model or IR can complete the rig.

Playback: FRFR belongs at the output end. It is an amplification and monitoring approach intended to reproduce an already-shaped modeler signal over a wide frequency range. It is not itself a modeling or capture technology, and “flat response” is a design goal rather than a promise of perfect neutrality in every room. Examples of the kind of products involved include HeadRush’s FRFR CAB 8 and 12 and Avantone Pro’s GTR-100 power amp for modelers.

There is no single winner because the methods solve different problems.

If you want a broad library of adjustable virtual amplifiers, modeling is the straightforward fit. If you want to preserve a particular real rig, profiling or capture becomes more relevant. If community-made captures matter to you, the platform’s sharing ecosystem can be as important as the learning method. For studio recall, either approach can work: a model stores an adjustable parameter state, while a capture preserves the response of a particular configured setup.

The useful question is not “Which technology is best?” but: do you want to build a tone by adjusting a virtual amplifier, preserve and reuse the behaviour of a particular real setup, or use a platform that can do both?