AI Preset aoutomatzation QC ready

I’ve been working on a project that automate preset creation for Neural DSP. The idea is simple: you provide an artist, a song, or a description of the tone you’re after — the system researches the rig, maps it to plugin parameters and generates a ready-to-use .xml file you can drop straight into Cortex Control.

Already working with Plini X and Gojira X. Tested on dozens of presets, generated tones load and work in the plugin right away.

The goal is “describe it → get it → fine-tune” — not a perfect clone, but a solid starting point in seconds instead of hours at the knobs. Especially useful for people just getting started with Neural DSP who don’t want to get lost in a sea of parameters.

Technical side in short: the system reverse-engineers the binary JUCE ValueTree format used by Neural DSP presets and modifies parameters programmatically via Python. The project runs on Claude (version 4.6 minimum — other models aren’t recommended: Gemini can’t handle binary file output, GPT requires constant debugging and fighting with the format).

But I have a question for you — because I genuinely wonder about this myself: does a project like this even make sense, or is tweaking the knobs exactly where the magic happens and we shouldn’t take that away? One thing I know — the research the system does along the way (artist rig, mics, signal chain) is often genuinely eye-opening and teaches you more than an hour at the controls ever would.

What do you think?

3 Likes

I’ve been using Gemini to help me get some tones so I’m all in for this! I’m still new to guitar and have limited time so anything that can help me get at least close to a start point is great. And I’ve found it’s actually helped me learn about what tweaks lead to what as it often explains why it has made the suggestions.

A very impressive and admirable undertaking, Cebulus! For me personally, the journey is half the fun but, I’m sure there are a lot of folks who would love this kind of app. Especially for someone who is new to modelers or even just new to the QC, this kind of resources could be invaluable. I have seen requests on this forum for this kind of capability. Thanks for your endeavors.

GojiraXCustomGPT.zip (2.5 MB)

LiveCebik01_GojiraX_MIDIPRO.zip (5.9 KB)

Important: For scenes triggered via MIDI, CC Value commands are used. The MIDI controller should send MIDI Channel 1, CC#1 to CC#8, value 1, depending on the number of scenes.

Yo también estoy usando Gemini, pero más por desconocimiento que otra cosa.

Me aporta parámetros, sobre todo de ecualización, de los que tengo muy poco conocimiento.

Pero a la vez, es una forma de aprender a perillear los bloques, como en el mundo analógico.

Para quien ya sabe y controla cualquier elemento del Quad Cortex, el proyecto tiene que ser una pasada, pero para los novatos, el ir ajustando según te indica la IA, es la única forma de ir enterándote a base de probar, de porqué los valores indicados, se aproximan al sonido que buscas y te dan una satisfacción enorme cuando éste se consigue.

Eso si, hay que dedicarle mucho tiempo.

I think it is a great idea!

If it works well, wouldn’t mind paying a few $$ for it. So many songs I’d like to emulate, never enought time…

I’d check that out as well! Am considering playing around with Claude CoWork and my QC (and also Ableton), I’m all for finding efficient ways to get great results!

I had the same idea recently and was doing some research before starting to implement it.

For what it’s worth, AI thinks that reverse-engineering and building NDSP binary XML presets in their proprietary format is shady from a legal perspective and warns that they might get after you if you make this service public. I didn’t go as far as checking it with real lawyers though, so maybe it’s wrong.

But yeah, I think it’s a super nice idea, and I’d love to have something like this

I’m not planning to build this as a commercial service or anything profit-oriented — for me it’s more of a research/hobby project and a tool for my own workflow.

Regarding the binary NDSP preset format: yes, AI also points out that reverse engineering and publicly offering generated preset files could be legally questionable. I haven’t verified that with an actual lawyer, so I’d treat it more as a cautious warning than a definitive statement.

The interesting part, though, is that the binary preset file can largely be bypassed with a MIDI Mapping XML file. You can create a mapping that sets plugin parameters with a single CC message, or even split it into scenes — something like MIDI BAKE / MIDIPRO. In practice, you load a neutral preset, import the MIDI mapping, trigger a scene or CC, the plugin sets the parameters, and then you simply save the result as a normal preset.

In some ways, I actually find this more convenient for testing. Generating this kind of mapping with AI takes a few dozen seconds, and locally it would probably take only seconds. The MIDI Mapping file itself is not binary or specially protected — it mainly requires the correct structure and accurate target mapping.

In the ZIP I shared, there are ready-made instructions for a Custom GPT — both for the classic preset-file approach and for the MIDI-based workflow. Turning this into an actual app is a separate topic: API, UX, compatibility, distribution, and potential legal considerations. For now, I have the Plini X instructions ready, and a few more plugins are in progress.

I still think the idea is really good — especially as a tool for quickly prototyping presets and scenes.

2 Likes

This is a FANTASTIC idea. Specially the MIDI tone modification. Love it.

I would like do something similar but for Windows petrucci plugin. Great inspiration.

Petrucci.zip (23.4 KB)

Thanks! Attached are the instructions for Petrucci’s Archetype Plugin. It should work in Antigravity, CustomGPT, and Claude Projects. It includes an extended MIDI version for scenes and stomps, a MIDI one-shot, native preset instructions, analysis and web search instructions, and an agent manual.

1 Like

The journey is indeed half the fun but most of us use something aside from our feet to get to most places.

wow! Thanks. Can I use it on claude free? How?

Hey! To answer your question about using this on Claude free.

Short answer: It works on the free plan, but with some inconvenience. On the paid plan (Pro) it’s much smoother.

On the free plan you can still upload the files manually at the start of each conversation. On the Pro plan you can create a Claude Project where the files are loaded once and always available — no re-uploading every time.

What is this package?

It’s a set of 5 instruction files that turn Claude into an expert on the Archetype Petrucci X plugin. Once loaded, you can ask things like “make me a preset like the intro of Pull Me Under” or “live setup: clean, rhythm, lead on 3 footswitches” — and the AI will generate a ready-to-use preset file or MIDI configuration.

How the AI mode actually works — important to understand

The AI doesn’t hear your tone. It searches the web for information about a given song or artist’s rig, then translates what it finds into plugin parameters using the mapping instructions included in the package. It tries to reconstruct the tone based on that research.

If it finds limited information online, it falls back on the tonal patterns built into the instructions — and I’ll be honest, there aren’t that many of them yet, simply because there wasn’t enough time to build more. But that’s something anyone can supplement over time.

You can also work the other way entirely — if you already know what you want, just tell the AI directly, either by typing or using voice input: “set the lead gain to this, rhythm EQ like this, add a delay with these values.” No web search needed, just direct parameter control.

Option A — Claude Free (manual, no Project)

At the start of each conversation:

Open a new chat on claude.ai

Upload all 5 files from the package (use the attachment/paperclip button)

After uploading, write what you need

Downside: you need to do this every new conversation.

Option B — Claude Pro + Project (recommended)

Log in to claude.ai → click “New Project” in the left panel

Name it e.g. Petrucci X AI

In the “Project Knowledge” section, upload all 5 files from the package

From now on, every conversation inside that project automatically has access to the instructions — no re-uploading needed

MIDI configuration — important note

The package supports two modes:

Classic mode — the AI generates an .xml preset file, you drag and drop it into the plugin. No MIDI needed, zero configuration.

MIDIPRO mode — the AI generates a MIDI Mapping file for scenes/stomps. This requires a MIDI controller.

If you have a Quad Cortex:

You’re almost ready. QC footswitches in MIDI mode send CC#1, CC#2, CC#3 etc. — exactly what the instructions expect. You just need to enable MIDI Out in QC settings and configure the knobs/footswitches according to the CC values in the generated file.

If you have a different MIDI device (MIDI footswitch, keyboard, pedalboard etc.):

Check your device’s manual to find which CC numbers each button/pedal sends, then tell the AI when requesting a setup — it will adjust the mapping to your hardware.

One more thing about MIDI: the plugin has over 300 mappable parameters. Asking the AI to map everything at once — especially in complex setups like 4 scenes plus stomps — can overload it and produce errors. Keep it focused: only request mappings for the parameters you actually need to change. Less is more here.

How to talk to the AI in the project?

In your own language — you can write in Arabic, English, or anything else. The AI understands requests like:

“Make me a preset for the intro of Another Day”

“Live setup: 3 scenes — clean, crunch, lead solo”

“MIDIPRO — PRO MIDI SETUP for Pull Me Under”

If anything is unclear, just ask inside the project — the AI will explain every step.

Hope this helps! If you have a different MIDI controller than QC, let us know which one and we can help with the configuration :+1:

1 Like