Random Noise

The name is exactly what it is: random noise. Every pixel gets three fresh random values, one per RGB channel. No fractal math, no smoothing, no hidden agenda - just static, the way nature (and a uniform random number generator) intended. If you've ever wanted a big sheet of TV static you can actually wire into a workflow, this is the node.

It's one of the five noise generators in comfy-plasma by Jordach, and it's the plainest of the bunch. The interesting ones (Plasma, Pink, Brown) shape the noise; this one just gives you the raw material plus very explicit controls over where each channel is allowed to fall.

How it works

For each pixel, the node picks random.randint for red, green, and blue - three independent values, each clamped to its own range. Then it writes the pixel and moves on. That's the whole loop, and it means the node is CPU-bound and a bit slow at big sizes, but perfectly fine at 512×512.

The controls are where it gets useful:

• value_min / value_max - the global brightness range. The default -1 means "use 0 and 255," so full range.
• red_min / red_max, green_min / green_max, blue_min / blue_max - per-channel overrides. Leave a channel at -1 and it inherits the global value range.

So you can generate, say, noise that only lives in the red channel between 100 and 200, or force a channel to a narrow band for a specific tint. The README's own example is "red set to work between 0 and 128," which gives you a reddish-noise wash. That's a genuinely handy way to make a subtle color-stained texture without doing a channel split.

width and height step in increments of 8, from 128 up to 8192. Default 512×512.

The seed thing you need to know

The seed input seeds Python's random module, so a fixed seed does reproduce the same noise. But read the author's own comment in the source: the seed "does nothing because ComfyUI doesn't understand static output nodes." What that means in practice: ComfyUI caches this node's output, so you can't just twiddle the seed and expect a fresh roll - you have to actually change a value (bump the seed, or any other input) to force it to re-run. There's no auto-randomize control here. It's a reproducible-seed node with extra steps, so treat the seed as "edit me when you want a different one."

Install

No requirements.txt, no model downloads. The pack needs only Pillow, numpy, torch, and requests - all standard ComfyUI.

ComfyUI Manager → Install Custom Nodes → search comfy-plasma → install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/Jordach/comfy-plasma

Restart ComfyUI. The node lives under image/noise.

Common issues

The main one is patience. At 512×512 this is instant-ish; crank it toward 8192 and you're doing 67 million Python putpixel calls, which is a real wait. If you need big noise fast, generate small and upscale. And remember the seed caching gotcha above - if you change the seed and the output doesn't change, it's ComfyUI holding the cache, not the node being broken. Nudge any other input and you'll get your new roll.