What Neural Amp Modeler does
Neural Amp Modeler (NAM) trains a neural network to imitate a specific guitar amplifier or pedal. You play a standard test signal (the reamp files NAM provides) through your gear, record the output, and NAM's trainer learns the difference and exports a .nam model file. Training runs in the browser through Google Colab with nothing installed, or locally with the Python trainer from PyPI (neural-amp-modeler, version 0.13.0, Python 3.10 or later), with step-by-step videos for Windows and Mac. Most players never train a model themselves: they download captures other people have made from TONE3000, a community library NAM points to.
To play models, NAM offers Gateway, a free VST3/AU plugin and standalone desktop app for Windows 10 or later and macOS 10.15 or later, plus ParametricOD, a concept plugin that models an overdrive across the full range of its knobs and switches. The code lives in three MIT-licensed repositories: the trainer, the core DSP library and the iPlug2-based plugin; a community LV2 build covers Linux. The project is maintained by Steven Atkinson, funded by donations through Ko-fi, and its core is used in commercial products: NAM's site lists builders including Darkglass, Two Notes, Hotone, Valeton and MOD Audio.
