🎛️ Audio Mixer

Most of the fun in this pack is in the processors - the filters, the distortion, the chaos. AudioMixer is the unglamorous utility that lets you actually combine them into something. It's a four-input mixer with per-channel gain and pan, plus a master gain. Nothing clever, and that's the whole point: when you've got a noise bed, a tonal layer, and a feedback texture that each sound great alone, this is how they become one sound.

What it is

Four inputs - audio_a through audio_d - where only audio_a is required. Each channel gets its own gain (0 to 2) and pan (−1 hard left to +1 hard right). A master_gain rides on top of the whole sum. One output, mixed_audio, ready for the next processor or straight into AudioSave.

So the workflow shape you'll use it for is: split your sources, tune each one's level, then send the mix onward. It's the pack's version of a summing bus.

How it works

It's a weighted sum, exactly like a physical mixer channel: each input is scaled by its gain, panned between the left and right channels, and everything is added together before the master gain is applied. That's the whole mechanism. Don't overthink it.

The pan range is the thing to watch. Negative pans the left, positive pans the right, and 0 is center. If you're feeding mono material into a stereo bus, this is where you create the width - or kill it, if you forget and leave everything dead center.

The few controls that matter

• gain_a through gain_d - the levels. A noise wall from this pack can easily stack up loud; if your mixed_audio is clipping on the way into a filter, pull these down before you reach for the processor's own amplitude knob.
• pan_a through pan_d - where each source sits in the stereo field.
• master_gain - the safety valve. It also goes up to 2.0, which is more headroom than you usually want; treat anything above 1.0 as "I intend to clip."

Where it fits

The README keeps AudioMixer in the "Utility" bucket, and that's the right framing. It plays nice with the rest of the pack because everything speaks the standard ComfyUI audio format - a dict of waveform tensor plus sample_rate - so you can mix the output of a NoiseGenerator with, say, a VHS_LoadAudio sample and a FeedbackProcessor texture without any format conversion. That interoperability is the pack's quiet strength.

Install

Same as every node in this pack:

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, searching "NoiseGen" / eg0pr0xy. No models, no keys. Restart and it's in the 🎵 NoiseGen/Utility menu.

Gotchas

Two things worth knowing. First, don't skip the pip install - the pack imports scipy at load, so a missing requirements step takes down every node in it. Second, the gain and master_gain params max out at 2.0, and a handful of stacked 0.8-amplitude sources will sum past full scale fast. If things sound crunchy in a bad way, it's not the mixer's fault - check your gains first. That's genuinely the only way this node can bite you.