NAMFORUM Sysop: Created page

2026-09-01T06:05:49Z

Created page

New page

= Neural Capture =

'''Neural Capture''' is Neural DSP's technology for creating digital models of physical amplifiers, cabinets, pedals, and other compatible audio equipment.

Neural Captures are used within Neural DSP's Cortex hardware ecosystem and can be shared through Cortex Cloud.

There are currently two generations of the technology:

* '''Neural Capture Version 1'''
* '''Neural Capture Version 2'''

Both create playable Capture blocks, but they differ substantially in training method, computational requirements, and the types of equipment for which they are optimized.

== Overview ==

A Neural Capture measures the response of physical equipment to test signals generated by the Cortex system.

Depending on the target equipment, a Capture can represent:

* Amplifier
* Amplifier and cabinet
* Combo amplifier
* Cabinet
* Overdrive or distortion device
* Fuzz
* Compressor
* Larger signal chain

Once created, a Neural Capture can be inserted as a block in a Cortex signal chain.

== Basic capture process ==

A simplified capture path is:

'''Cortex capture signal → device under test → Cortex return'''

For a pedal:

'''Cortex → pedal → Cortex'''

For a direct amplifier:

'''Cortex → amplifier → suitable load/direct-output device → Cortex'''

For a miked amplifier and cabinet:

'''Cortex → amplifier → cabinet → microphone → Cortex'''

The Cortex system sends test signals through the equipment and records its response.

The captured data is then used to create the Neural Capture model.

== Neural Capture Version 1 ==

'''Neural Capture Version 1''' is the original Neural DSP capture algorithm.

V1 training occurs directly on compatible Cortex hardware.

It is:

* Fast
* Computationally efficient
* Fully offline
* Suitable for most amplifiers
* Suitable for cabinets
* Suitable for many overdrive pedals

A typical V1 Capture takes only a few minutes to create.

Because training happens locally, no cloud connection is required for the modeling process.

V1 remains fully supported alongside Version 2.

== Neural Capture Version 2 ==

'''Neural Capture Version 2''' was introduced in November 2025.

V2 uses a newer, higher-resolution capture algorithm designed to reproduce more complex dynamic behavior.

Unlike V1, the training process does not occur entirely on the Cortex hardware.

Instead:

# Cortex hardware performs the physical measurement.
# Cortex Control manages the capture session.
# Captured data is sent to Cortex Cloud.
# Cortex Cloud performs the higher-resolution training.
# The completed Neural Capture is returned to the Cortex ecosystem.

An internet connection and Neural DSP account are therefore required when creating V2 Captures.

== Why V2 was developed ==

Some analog devices are particularly difficult to reproduce using capture systems.

Their behavior may depend strongly on:

* Input amplitude
* Playing dynamics
* Transients
* Previous signal history
* Compression
* Power-supply behavior
* Interaction between controls

Neural DSP developed Capture V2 specifically to improve reproduction of these characteristics.

Examples include:

* Vintage fuzz pedals
* Compressors
* Highly dynamic tube amplifiers
* Amplifiers exhibiting sag and bloom
* Devices with blend controls

V2 is not intended merely as a replacement with a higher version number.

It represents a more computationally demanding option for cases where additional dynamic accuracy is useful.

== V1 versus V2 ==

Both capture generations remain useful.

'''Version 1''' is generally appropriate when:

* Offline capture is required
* Fast capture is important
* DSP resources are limited
* The target is a conventional amplifier
* The target is a cabinet
* The target is a relatively straightforward overdrive

'''Version 2''' is particularly useful when:

* Maximum dynamic detail is desired
* Capturing fuzz
* Capturing compression
* Capturing a highly touch-sensitive amplifier
* Capturing power-amp sag or bloom
* Capturing complex dynamic behavior

V2 models use more processing resources during playback than V1 models.

The choice therefore involves both modeling requirements and available DSP.

== V2 capture time ==

Neural DSP states that a V2 Capture session typically takes approximately '''10 minutes'''.

This includes the cloud-based training process.

Actual completion time can depend on the capture session and network/cloud conditions.

V1 is substantially faster because its lighter training algorithm runs locally.

== Creating V2 Captures ==

The current V2 workflow uses '''Cortex Control'''.

A typical procedure is:

# Connect the target equipment according to the displayed connection diagram.
# Open Cortex Control.
# Log into the Neural DSP account.
# Select '''New Neural Capture'''.
# Select '''Neural Capture Version 2'''.
# Configure the capture connections.
# Calibrate levels.
# Enter the required metadata.
# Start the capture.
# Wait for cloud processing.
# A/B the resulting Capture against the original equipment.
# Save the Capture.

Cortex Control acts as the connection between the Cortex device and Cortex Cloud during V2 creation.

== V2 metadata ==

Metadata has an unusually important role in Neural Capture V2.

Before V2 training begins, the creator identifies the type of equipment being captured.

Current V2 device categories include:

* Default
* Amp
* Amp + Cab
* Combo Amp
* Cab
* Overdrive
* Fuzz
* Compressor

Neural DSP states that Cortex Cloud uses this information during processing.

Selecting the correct target type is therefore part of the modeling process rather than merely library organization.

== Capture calibration ==

The Neural Capture workflow includes a calibration stage.

Correct send and return levels are important because the capture system must measure the device over an appropriate operating range.

For nonlinear equipment, changing the signal level entering the device can change:

* Distortion
* Compression
* Dynamic response
* Attack
* Sustain
* Cleanup behavior

Do not treat calibration as merely a way to prevent clipping.

It determines the conditions under which the device is measured.

See [[Gain Staging and Calibration]].

== A/B testing ==

After the capture is completed, the Cortex workflow provides an A/B comparison between the original equipment and the resulting Neural Capture.

This allows the creator to evaluate the model before saving it.

Listen for differences in:

* Gain
* Frequency response
* Pick attack
* Dynamics
* Compression
* Sustain
* Cleanup
* Low-frequency behavior
* High-frequency behavior

Comparisons should be made at closely matched levels.

A louder signal can easily be perceived as fuller or better even when the underlying model is not more accurate.

== Capturing amplifiers ==

An amplifier can be captured directly without its physical cabinet response.

This requires a safe and appropriate method of obtaining the amplifier output.

A typical path is:

'''Cortex → amplifier → reactive load/direct-output device → Cortex'''

The resulting Capture can then be combined with:

* Cortex cabinet processing
* An impulse response
* A physical guitar cabinet

See [[Cabinets and IRs]].

== Tube amplifier safety ==

'''Never connect the speaker output of a tube amplifier directly to an ordinary Cortex input.'''

Doing so can damage both the amplifier and the Cortex device.

When capturing a tube amplifier, use:

* A suitable D.I. output while the amplifier remains connected to the required speaker load, or
* An appropriate reactive load box between the amplifier speaker output and the Cortex input

The amplifier must always see the load required for safe operation.

== Capturing amplifier and cabinet ==

A complete amplifier/cabinet signal chain can be captured using a microphone.

For example:

'''Cortex → amplifier → cabinet → microphone → Cortex'''

The resulting Neural Capture can then include the combined behavior of:

* Amplifier
* Speaker
* Cabinet
* Microphone
* Microphone position
* Return path

Additional cabinet processing may not be required when the cabinet and microphone are already part of the Capture.

See [[Capture Types]].

== Cabinet captures ==

Neural Capture can also capture a cabinet independently.

A cabinet Capture represents the measured speaker/cabinet/microphone path used during creation.

This provides an alternative to conventional static cabinet impulse responses within the Cortex environment.

Cabinet Captures and conventional IRs should not automatically be assumed to be technically identical simply because they can serve a similar role in a guitar signal chain.

== Pedal captures ==

Neural Capture can reproduce compatible nonlinear pedals.

V1 has traditionally been used for devices such as:

* Overdrive
* Distortion
* Boost

V2 extends the practical range toward more dynamically difficult devices such as:

* Fuzz
* Compressor

The physical pedal settings used during capture form part of the operating point represented by the resulting model.

For substantially different settings, separate Captures may be useful.

== Fuzz capture ==

Fuzz circuits can present a particularly difficult modeling problem.

Some fuzz pedals interact strongly with:

* Source impedance
* Guitar volume control
* Pickup
* Input level
* Circuit bias
* Playing dynamics

Neural DSP identifies fuzz as one of the primary applications for Capture V2.

V2 is specifically intended to improve behaviors such as guitar-volume cleanup that can be difficult for simpler capture systems to reproduce.

== Compressor capture ==

Capture V2 adds explicit support for compressors.

Compression is inherently dynamic because the processor's behavior changes according to the level and timing of the incoming signal.

Neural DSP notes that V2 can capture compressors with high accuracy, although extreme threshold settings and long release times may produce greater differences from the original hardware.

This illustrates an important principle:

No capture architecture should automatically be assumed to reproduce every possible device and setting perfectly.

The result should always be evaluated against the original.

== Neural Capture blocks ==

Once created or downloaded, a Neural Capture can be placed on '''The Grid''' as a processing block.

V1 and V2 Captures are visually distinguishable.

V2 Capture blocks use a different double-layered outline.

Both versions provide the same general parameter-editor layout.

== Cortex Cloud ==

'''Cortex Cloud''' is Neural DSP's online system for storing, sharing, and discovering Cortex content.

Users can make Captures available to other Cortex users.

V2 relies more directly on Cortex Cloud because its training process occurs there.

V2 Captures are automatically stored privately in the creator's Cortex Cloud profile during creation.

They can subsequently be managed and shared according to the available Cortex Cloud controls.

== Sharing Captures ==

Community Neural Captures can be downloaded from Cortex Cloud and used in compatible Cortex hardware.

Useful capture information includes:

* Creator
* Device manufacturer
* Device model
* Capture type
* Device settings
* Cabinet
* Microphone
* Gain
* Instrument
* Notes

Good metadata is particularly valuable when a Capture becomes separated from the discussion or session in which it was originally created.

== Quad Cortex ==

'''Quad Cortex''' is Neural DSP's flagship Cortex hardware platform.

It combines:

* Neural Capture playback
* Neural Capture creation
* Algorithmic amplifier models
* Cabinet modeling
* Effects
* Routing
* MIDI
* USB audio
* Preset management

Quad Cortex can create and play both Neural Capture V1 and V2.

V1 can be created directly on the hardware.

V2 creation uses Quad Cortex together with Cortex Control and Cortex Cloud.

== Cortex Control ==

'''Cortex Control''' is Neural DSP's desktop application for managing Cortex hardware.

It provides computer-based control over functions including:

* Preset editing
* Device management
* Cortex Cloud interaction
* Neural Capture creation

The complete V2 capture process can be initiated and managed through Cortex Control.

V2 creation requires Cortex Control because the application connects the hardware capture process with cloud training.

== Nano Cortex ==

'''Nano Cortex''' is Neural DSP's compact Cortex hardware platform.

Nano Cortex supports Neural Capture playback and V1 capture functionality.

With NanOS 2.2.0, Neural DSP added the '''Capture 2 Player''', allowing Nano Cortex to load and play V2 Captures created using Quad Cortex.

At the introduction of Capture V2, direct creation of V2 Captures on Nano Cortex was not yet available.

Because this capability is under active development, consult current Neural DSP documentation for its latest status.

== Quad Cortex mini ==

The '''Quad Cortex mini''' extends Neural DSP's Cortex hardware family with a smaller form factor while retaining the broader Cortex ecosystem.

Its exact capture and playback capabilities should be checked against current Neural DSP documentation because Cortex hardware and software features continue to evolve.

== Factory Captures ==

Neural DSP supplies factory Neural Captures in addition to community-created content.

When Capture V2 was introduced, Neural DSP added '''669 V2 Captures across 41 devices''' to its library and official Cortex Cloud profile.

These included examples of:

* Guitar amplifiers
* Combo amplifiers
* Overdrives
* Fuzz pedals
* Compressors

Factory libraries continue to evolve, so exact counts should be treated as historical information rather than permanent specifications.

== Neural Capture and conventional Neural DSP models ==

Neural Capture should not be confused with Neural DSP's conventional algorithmic device models.

Quad Cortex contains both.

An algorithmic amplifier model is developed by Neural DSP to represent an amplifier and normally exposes modeled controls.

A Neural Capture is created by measuring a particular physical device or rig at a particular configuration.

They are different technologies that can coexist in the same Cortex preset.

== Neural Capture and NAM ==

Neural Capture and [[Neural Amp Modeler]] are both data-driven approaches to reproducing physical audio equipment, but they are separate technologies.

A Neural Capture is not a ''.nam'' file.

The ecosystems differ in:

* Capture process
* Training
* Model representation
* Playback engine
* Distribution
* Hardware integration

NAM provides an open-source model and playback ecosystem.

Neural Capture is Neural DSP's proprietary technology integrated with Cortex hardware and Cortex Cloud.

== Neural Capture and TONEX ==

Neural Capture and [[TONEX]] can both create models of amplifiers, pedals, cabinets, and complete rigs.

However, their model formats and playback systems are not interchangeable.

Both systems also now distinguish the physical capture process from computationally intensive training in some workflows:

* TONEX Modeler can preserve capture data and train it later.
* Neural Capture V2 sends captured data to Cortex Cloud for training.

The engineering implementations remain different.

== V1 and V2 compatibility ==

V1 remains supported after the introduction of V2.

This is important because V2 is more computationally expensive and requires cloud processing during creation.

The two systems therefore serve different practical purposes rather than V2 simply making V1 obsolete.

A user may deliberately choose V1 for:

* Faster capture
* Offline operation
* Lower DSP usage

and V2 for:

* Greater dynamic detail
* Fuzz
* Compression
* Highly responsive amplifiers

== Troubleshooting ==

If a Neural Capture differs substantially from the original equipment, check:

* Connections
* Send level
* Return level
* Calibration
* Clipping
* Load-box configuration
* Microphone path
* Capture type
* Metadata selection for V2
* Additional cabinet processing
* A/B comparison level

For V2, also check:

* Internet connection
* Cortex Control
* Neural DSP account login
* Audio-driver configuration
* Software permissions
* Other applications using the audio device

A DAW or other program accessing the same audio system can interfere with a V2 capture session.

See [[Troubleshooting Captures]].

== Official resources ==

* [https://neuraldsp.com/manual/quad-cortex Quad Cortex User Manual]
* [https://neuraldsp.com/news/introducing-neural-capture-version-2 Introducing Neural Capture Version 2]
* [https://neuraldsp.com/quad-cortex Quad Cortex]
* [https://neuraldsp.com/cortex-control Cortex Control]

== See also ==

* [[Capture Technologies]]
* [[Neural Amp Modeler]]
* [[TONEX]]
* [[Kemper Profiling]]
* [[Capture Types]]
* [[Creating a Capture]]
* [[Capture Signal Chain]]
* [[Gain Staging and Calibration]]
* [[Reamping for Capture]]
* [[Cabinets and IRs]]
* [[Troubleshooting Captures]]

'''[https://namforum.com/ Discuss Neural Capture, Quad Cortex, and Cortex workflows on NAMFORUM]'''

[[Category:Neural Capture]]
[[Category:Capture technology]]
[[Category:Commercial capture systems]]

Neural Capture - Revision history

NAMFORUM Sysop: Created page