⚡ Microsound Sculptor

Some nodes are tools. MicrosoundSculptor is a weapon. It's granular processing aimed squarely at harsh noise and microsound art - the genre where audio is treated like material to be ground up, not music to be polished. If GranularProcessor is the friendly instrument, this is the one you feed a sound you don't like, and it comes back unrecognizable and angry.

What it is

Two independent damage systems stacked on each other. The destruction_mode picks how the signal gets wrecked at the sample level - bitcrush (crunchy digital aliasing), saturation (waveshaping grit), chaos (unpredictable sample mangling), ring_mod (metallic amplitude modulation), or frequency_shift (the detuned, sci-fi wobble). The amount of that is destruction_intensity (0–1). On top of that, sculpting_mode decides how the grains get treated: grain_filter, grain_morph, grain_feedback, or spectral_destroy - each a different way of mangling the granular layer, with sculpting_intensity (0–1) controlling how far it goes.

How it works

The engine is a microsound granular processor - grains here default to 25 ms, far tinier than GranularProcessor's typical settings, which is the micro in microsound. grain_size (1–200 ms) is the only real "musical" control; everything else is about destruction. chaos_rate (0.1–100) drives how fast the chaos mode churns, and feedback_amount (up to 0.9) feeds the output back into itself - with grains this small, feedback becomes a fast, snarling loop. microsound_density (up to 1000 grains/second) is the cloud thickness; near the top it stops being grains and becomes a wall of texture.

The optional controls: spectral_chaos adds FFT-domain randomization on top, grain_randomization spreads the grain timing/pitch unpredictably, and microsound_density you already met. Output is microsound_audio.

The inputs that matter

• destruction_mode - the flavor of the violence. bitcrush and ring_mod are the most controllable; chaos and frequency_shift are where it gets unpredictable.
• destruction_intensity and sculpting_intensity - your two main "how far" dials. Start low; this node goes from 0 to terrifying in a few ticks.
• grain_size - the one dial that feels like an instrument. 1–10 ms is pure destruction, 50+ ms starts to sound like chopped samples.

Install

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/eg0pr0xy/comfyui_noisegen.git
cd comfyui_noisegen
pip install -r requirements.txt

Windows embedded Python: ComfyUI\python_embeded\python.exe -m pip install -r requirements.txt. Or ComfyUI Manager, search "NoiseGen" / eg0pr0xy. No models, no keys.

Gotchas

Pack rule: don't skip the pip install - scipy is imported at module load and the whole pack fails to register without it. This is also the loudest node in the pack: high density + feedback + saturation is a recipe for brutal clipping, so check levels with the amplitude knob before you commit to a render. And honestly, this node is a one-trick pony in the best way - it does extreme, and it does extreme well. For anything subtler you want GranularProcessor or a SpectralProcessor instead. Bring a source you can afford to lose.