🔄 Feedback Processor

Feedback is how noise stops being static and starts being alive. Feed a signal back into itself, and it turns into a self-sustaining, evolving creature - the audio equivalent of a slow-motion avalanche. FeedbackProcessor is the pack's centerpiece for this, and it's the first processor in the README's own "harsh noise wall" recipe: NoiseGenerator → FeedbackProcessor → HarshFilter → MultiDistortion. If you build only one workflow in this pack, build that one.

What it does

It's a delay-based feedback loop with six personalities via feedback_mode: simple, filtered, saturated, modulated, complex, and runaway. The mode changes what sits inside the loop:

• simple - a plain delay feeding back on itself. Good clean echo.
• filtered - the default; a filter sits in the loop, so each pass gets shaped. This is where the interesting timbres come from.
• saturated - waveshaping in the loop. Each recycle gets dirtier until it's a wall of grit.
• modulated - the delay time wiggles via an LFO, giving you the warped, pitch-bending textures.
• complex - multiple paths, the "everything at once" mode.
• runaway - feedback so hot it self-oscillates. Expect screaming.

The core knobs are feedback_amount (0 to 0.95 - past about 0.8 is where it stops being echo and starts being a monster) and delay_time (0.1 to 100 seconds, so it can be anything from a slapback to a cavern). When filter_type is active (lowpass/highpass/bandpass/notch/allpass), filter_freq and filter_resonance tune it, and resonance near 1.0 will ring on its own. saturation adds drive to the loop. The modulation section - modulation_rate, modulation_depth, modulation_type (sine/triangle/sawtooth/square/noise) - drives the modulated mode but can add life to any of them. A wet_dry_mix (default 0.7) blends the processed signal with the original, and amplitude masters the output.

How it works

At its simplest: the output is delayed, scaled by the feedback amount, filtered, and mixed back into the input - repeatedly. Each pass around the loop either decays (low feedback) or accumulates and self-oscillates (high feedback). Because each recycle is filtered, the spectrum collapses toward whatever the filter emphasizes, which is why a lowpass filter with high feedback sounds like a roaring, evolving drone rather than a repeated echo.

The inputs that matter

• feedback_amount - the personality knob. Start at 0.3–0.4, explore past 0.8 carefully.
• feedback_mode - the character knob. filtered is the safest interesting place to start; runaway is where you go when you've stopped caring.
• delay_time and filter_freq - the two dials that decide what you hear: the rhythm of the repeats and the color of each pass.

Output is a single feedback_audio.

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. ComfyUI Manager has it too - search "NoiseGen" or eg0pr0xy. No models, no keys.

Gotchas

Feedback + high resonance + loud input is how you get instant digital clipping, so keep an eye on the output level and lean on amplitude and wet_dry_mix. Also remember the pack-wide rule: the pip install is mandatory - scipy and soundfile are imported at load, and without them the entire pack fails to register, not just this node. If the node exists in your menu, you already installed it right.