NAM file inspector

Nothing you choose here leaves your browser. A Neural Amp Modeler capture is a JSON file, so this page opens it and reads out what is actually in there — which architecture, what sample rate it expects, how heavy it will be, and who captured what. No plugin, no DAW, no upload.

…or drop files anywhere on this page.

The sample rate is the one that catches people

A capture is trained at a particular sample rate and expects to be played at the same one. The field is optional, and when it is missing the specification says to assume 48 kHz — which is what every player does, silently. A great many shared captures leave it out.

So if your session runs at 44.1 kHz and the capture was trained at 48, the amp is being fed samples at a rate it was not modelled for. It still makes a sound. It is not the sound that was captured. This page tells you which case you are in, which nothing else does without opening the file by hand.

What the architecture tells you

The architecture field says which shape of neural network is inside, and that decides what can load it. WaveNet and LSTM are the original designs, now called A1. A2 is the newer architecture, and software or hardware has to be updated to understand it — an older player will not simply make do.

And one thing this page will not pretend to do. It cannot convert an A1 capture into an A2 one. The weights in the file are the trained parameters of one particular network; there is no arithmetic that turns one architecture's weights into another's. Making an A2 version means training again from the original recorded audio, which is not in the file and never was.

Parameters, and what they cost you

The weights list is every number the network learned. More parameters generally means a more faithful capture and more processor time per sample — which is why the lighter variants exist, and why a pedalboard unit cares about this number in a way your desktop does not.

Where to go next

amp dating — the serial number on the real thing
pedal dating — the same for pedals
JSON to YAML — a .nam is JSON, so this reads one too

What this does not do

It reads the file and reports what it says. It does not play the capture, judge whether it sounds good, convert it to another architecture, or turn it into an impulse response — an IR is a linear filter and an amp capture is deliberately not linear, so the one cannot represent the other.

Nothing is uploaded, so it cannot read a file that is not on the device you are using, and it does not remember what you opened. It reports the file's own claims: if a capture says it was made from a Fender Deluxe, that is what the person who made it typed.