Make new Neural Amp Modeler (NAM) models from the captures you already have.
Blend two amp captures into one amp that changes character as you dig in, or
turn a pack of fixed-gain captures of one amp into a single model you sweep
with the ordinary Input knob. NAM Mixer lets you hear the result first, then
trains it into one standard .nam file you can load anywhere NAM runs. The
original captures aren't needed once it's trained.
Status: Beta. It runs on your own computer. Your captures and audio stay there unless you choose to train on your own free Kaggle GPU account.
⬇ Download · What you can make · How it works · Good to know · Troubleshooting · User guide · Developer guide
| 🎚️ Dynamic Hybrid | A clean amp that opens up into a crunch amp the harder you play. The change is driven by your picking, not a footswitch. |
| 🔀 Parallel Blend | Both amps together at one fixed ratio, like running two amps at once, whatever your playing level. |
| 🎨 Character Blend | One amp's tone with another amp's feel and drive, built from measured differences rather than a simple mix. |
| 🎛️ Continuous Gain | Several captures of one amp (Gain 1 … 10, say) collapsed into one model you sweep with the player's Input gain. |
| 🔈 Cabinet | Audition any cabinet IR, and optionally train it into the model as a full-rig capture. |
| 🧰 NAM Tools | Change a model's output volume or its descriptive metadata, add a cabinet, or inspect what's inside a .nam. Tools never overwrite your original file. |
Every trained model is checked against what you auditioned before you download it, and you can compare the two by ear in the app.
The desktop app needs no Python or other setup:
| Platform | Download | Notes |
|---|---|---|
| macOS (Apple Silicon) | NAM-Mixer-macOS-arm64.dmg |
Not notarised: the first time, right-click the app and choose Open (or allow it in Privacy & Security). |
| Windows (x64) | setup.exe · .msi |
Installer or MSI package. |
| Linux (x64) | AppImage · .deb |
Portable AppImage, or a Debian/Ubuntu package. |
Older versions and release notes are on the Releases page. Prefer to run it from the source code? See Run from source.
The Builder takes you through five steps, shown along the top of the app:
- Choose amps. Pick an Amp A and an Amp B
.nam, a test performance (a recorded DI riff is included), and your instrument and pickups. Then click Prepare amps for comparison. This is the only slow step. - Compare amps. Listen to Amp A, Amp B and the combined result. Test gain lets you play the clip harder or softer.
- Shape the sound. Choose Dynamic Hybrid, Parallel Blend or Character Blend and adjust it. Changes are instant, and the levels of the two amps are matched for you.
- Finish & polish. Optionally add a cabinet IR and set the output level.
- Make a model. Create the training files, then train on a free Kaggle GPU (recommended) or on your own computer. When training finishes, the model is checked, and you can listen to it side by side with what you designed.
A Tone Wizard can suggest settings from a plain-English description ("glassy clean that breaks up into a British crunch"). It works on built-in rules, or you can connect an AI model in Settings.
Continuous Gain has its own tab: add your captures, let it analyse and choose which ones to train on, train, then test and export. Save your work at any time in Sessions; unsaved Builder work is also autosaved, so you can restore it after closing the app.
The user guide covers every step and control in detail.
Does it merge the models' neural networks? No. Combining the weights of two separately trained networks doesn't produce anything musical. NAM Mixer plays the same audio through each amp with the real NAM engine, combines the audio, and trains a fresh model on that result. More
Will the trained model sound exactly like what I auditioned? It will be close, but it's a new model learning to imitate the result, so always listen to it. The app reports how closely the model matches (for its full and lite versions, and at quiet playing levels), and lets you compare the two by ear.
Do I need a powerful computer or a GPU? No. Auditioning and designing run on an ordinary computer. For training, the free Kaggle GPU option needs only a Kaggle account with a verified phone number (setup). Local training works too, but it's slower without a GPU, and the first time it needs a one-time download of about 1.4 GB.
Where can I use the finished model? Anywhere that loads standard NAM models. Trained models are standard NAM A2 files. The optional experimental "Create both" cabinet export is different: some players can't load it, so a standard head-only model always comes with it.
Will it line up my two amps' timing automatically? No. It never blindly phase-aligns amps. Different amps naturally delay different frequencies by different amounts, and that's part of how they sound. NAM Mixer only offers a correction when it finds the same fixed delay on several different recordings, which is what real recording latency looks like. Even then, correcting is your choice, and Original is the default. How the Timing check works
Is anything sent to the internet? Your captures and audio are never uploaded unless you choose Kaggle training, which uses your own account. AI suggestions can run on a model on your own computer. When it starts, NAM Mixer asks GitHub whether a newer version exists. That's an anonymous check that sends nothing about you or your files, and you can turn it off in Settings → Advanced. There's no telemetry and no account.
How do I update? When a new version is out, NAM Mixer tells you at startup and offers the right installer for your computer. Choose Not now to carry on; you can check again any time in Settings → Updates.
Why does the result change when I pick a different pickup? The pickup (input profile) changes how hard the virtual guitar drives both amps, so the amps are rendered again. It only affects the preview. Training always uses the official NAM training signal. More
- macOS says the app can't be opened. Releases aren't notarised yet. Right-click the app, choose Open, or allow it under System Settings → Privacy & Security.
- "Train A2 (Kaggle)" is unavailable or has no GPU quota. Kaggle only gives GPU time to accounts with a verified phone number (kaggle.com/settings). The Settings → Getting started checklist shows what's still missing.
- The renderer is missing (running from source). Use Download
nam_render automatically in Settings, or run
scripts/download_nam_render.sh(.ps1on Windows). - Windows:
nam_render.exeflashes and closes. That's expected. It's a helper the app calls, not the app itself. Start NAM Mixer withscripts/run.ps1. - The browser shows a 403 error on macOS (running from source). macOS
uses port 5000 for AirPlay. NAM Mixer uses port 5001 by default; to pick
another, run
PORT=5003 scripts/run.sh. - A loaded session says a file is missing. Loading prepares the amps straight away; if an amp or cabinet file it used has been deleted, choose it again and prepare the amps.
Requires Python 3.10+. On macOS or Linux:
git clone https://github.com/daverage/nam-mixer.git && cd nam-mixer
python3 -m venv .venv && source .venv/bin/activate
python3 -m pip install -r requirements.txt
scripts/download_nam_render.sh # prebuilt native NAM renderer
scripts/run.sh # opens http://127.0.0.1:5001/For Windows, building the renderer yourself, local training, tests and the code layout, see the developer guide.
For users
- User guide: every step, design mode, cabinet option, Continuous Gain, the Wizard, AI Assistant and TONE3000 tabs, Sessions, NAM Tools and Settings, plus how the blending works
- Release notes: what changed in recent releases, and upcoming changes under "Unreleased"
For contributors
- Developer guide: running from source on every OS, building
nam_render, tests, project layout, current status and limitations
Technical notes
- Continuous Gain: the design and the frozen configurations behind the Continuous Gain tab
- A/B timing: how the Timing check and the optional fixed-offset correction work, with measurements
Component references
- Desktop app: building and bundling the desktop app
- Renderer: building the native
nam_rendertool and its NAMCore pin - DI clips: provenance, credits and limitations of the bundled DI recordings
Beta. NAM Mixer renders source models through real NAMCore inference and generates trainable A2 bundles end to end. Listen to and validate each generated model before relying on it in a performance or production setup. See current status and limitations.
NAM Mixer is copyright © 2026 Andrzej Marczewski and is released under the
MIT License. The bundled genre/style DI recordings are credited and
documented separately in assets/di/README.md; their
upstream terms continue to apply. Neural Amp Modeler, NAMCore, and other
third-party components retain their own copyrights and licenses. The DI
clips come from the NAMtoClo
project; NAM Mixer shares no code with it.