How AI plugin makers build a plugin: new code or tested blocks
Answer in brief
A maker either asks an AI model to write new DSP code for each plugin or has it pick and connect blocks that already work. New code can go anywhere but has to prove itself every time. Tested blocks limit the palette but start from parts that are known to run.
Writing new code for each plugin
Most AI makers ask a language model to write signal-processing code and then compile it. Noizefield Audio Plugin Coder (APC) guides coding agents through a C++ project built on JUCE. Amorph has the AI write Cmajor, a language made for audio, and compiles it inside your DAW. pluginmaker.ai generates Faust code and compiles it into the plugin, and ChatDSP writes code for an Ableton device that you can also take elsewhere.
The appeal is range, since a model can try to write any filter or oscillator it can describe. The cost is that every plugin is new software that nobody has run before you play it.
Assembling tested blocks
The other approach keeps the code fixed and lets the AI make the choices. Studio 56 picks from a library of about a hundred verified DSP blocks, such as wavetable, FM and granular oscillators, drive stages, a tape machine and a character compressor, then sets and wires them to match your description. Chatmu says its agent never writes code and instead writes a patch from 31 DSP modules.
Imagine Plugins takes a block approach without any AI. You place blocks that its DSP developers built or licensed and connect them yourself, and Romplur works the same way with your own samples.
What this means for crashes
Code a model just wrote can contain mistakes that compile fine, such as a value that keeps growing until the output blows up. Each maker handles that risk its own way. APC lists pluginval, a plugin validation tool, among its requirements. Chatmu says it rejects a broken patch and returns to the last good version. ChatDSP calls itself experimental and warns that smaller local models may need more prompts to fix code errors. With tested blocks the parts have already run many times, so the remaining risk sits in how they are set and connected.
What this means for CPU
A block that was written once and tuned can be checked for how much processing it needs. Freshly written code may be lean or wasteful, and you only find out when several copies run in a busy session. Whichever maker you use, open the result on a few tracks of a real project and watch your DAW’s CPU meter before you rely on it.
What this means for sound
Sound quality lives in details such as how an oscillator avoids aliasing or how a filter behaves when pushed hard. Proven blocks carry those details from one build to the next. New code has to get them right each time, and a model can describe a warm filter without writing one that sounds warm. Listen at the extremes of every control, where weak DSP usually gives itself away.
Which approach suits you
If you want something no block library covers, such as a strange experimental effect, a code-writing maker gives you more room to try. If you mostly want instruments and effects that play well in songs, a maker built on tested blocks starts from parts that already work.
Sources for this page
- Studio 56: Studio 56’s official website, the publisher of this reference.
- pluginmaker.ai: pluginmaker.ai’s official page, the source of its description here.
- Romplur: Romplur’s official page, the source of its description here.
- Imagine Plugins: Imagine Plugins’s official page, the source of its description here.
- Noizefield Audio Plugin Coder: Noizefield Audio Plugin Coder’s official page, the source of its description here.
- Chatmu: Chatmu’s official page, the source of its description here.
- Amorph: Amorph’s official page, the source of its description here.
- ChatDSP: ChatDSP’s official page, the source of its description here.
The practical advice here is our own. A maker's description of its product is not a test of how it performs.
Keep reading
Cite this page
Best AI Plugin Maker. “How AI plugin makers build a plugin: new code or tested blocks.” Studio 56 Reference. Content reviewed November 8, 2026.
https://bestaipluginmaker.com/how-they-build/