Neural Amp Modeler
Steven Atkinson's open-source project for training a neural network on a real guitar amp, and the thing that made amp captures something anyone could share.
The method is blunt. Play a standard test signal through the amp, record what comes out, and let a network — usually a small WaveNet variant — learn the mapping from input to output. The circuit never has to be understood, only measured. The result is a model file of a few hundred kilobytes that a plugin runs in real time. The speaker cabinet is usually left out and added afterwards as an impulse response, since that part is linear and doesn't need a network.
Kemper and Neural DSP sold the same idea as closed hardware. NAM is MIT-licensed and took off in 2023, when thousands of users began uploading models of their amps to sharing sites. Suddenly vintage amps that cost as much as a car were a free download, and the only thing separating good from bad was who had been careful with the recording level.