Omni Noise Generator

Noise is one of those things you only think about when you need it, and then you need five different kinds. OmniNoise is a single node that generates Random, Plasma, Grey, Pink, or Brown noise depending on one dropdown - it's the "unified generator" in a fork of Jordach's comfy-plasma, with the CPU loop rewritten in PyTorch so it runs on your GPU instead of crawling through pixels one at a time. The fork claims a 100–200x speedup over the original, and for a node that produces a 1024×1024 frame of noise, that's the difference between instant and "why is my workflow stuck on the noise node."

Pick noise_type, set width and height (128–8192, stepped in multiples of 8), drop in a seed, and out comes a single IMAGE. That's the whole thing. What you do with it is where the fun is.

The settings that matter

The per-channel min/max inputs look intimidating but have one trick that makes them easy: -1 means "use the global value." So value_min/value_max set the overall brightness range (0–255), and the six red/green/blue inputs are only there if you want, say, a red-only channel or a dim blue. Leave them all at -1 and you get full-range noise. In practice you'll touch value_min/value_max (to keep noise in a usable range) and mostly ignore the rest.

Two settings are noise-type-specific and hidden when they don't apply:

• random_distribution - only for Random noise. Uniform (TV Static) is the classic static look and is backwards-compatible with the original RandNoise node; Gaussian (Centered Gray) is centered gray noise, which diffuses more evenly and is what you want when you're using noise as a soft texture.
• turbulence - only for Plasma. Lower values make bigger, smoother fractal clouds; higher values get spikier and more static-like. Default 2.75 is a decent middle.

What you'd actually use it for

Honestly, this is a niche node with real users. Random noise is the obvious img2img starting texture - feed it as the image input and control exactly what initial chaos the sampler sees, reproducibly, since the seed is a widget. Plasma is the distinctive one: fractal cloud patterns that read as smoke, water, or texture when blended into an image. Grey/Pink/Brown are the film-grain family - Brown especially gives that classic warm analog grain, and a light grain pass over an otherwise too-clean render is a cheap way to kill the "generated" plasticky look. The same pack's Smart Resolution Calculator even uses these noise types as latent fill patterns, so if you've seen "DazNoise" as a fill option, this is the engine behind it.

Installing it

It ships in the DazzleNodes collection - search DazzleNodes in ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone --recursive https://github.com/DazzleNodes/DazzleNodes.git

Restart and find it under image/noise → Omni Noise Generator (the node class is JDC_OmniNoise, the display name is "Omni Noise Generator"). No models, no extra pip deps - it uses ComfyUI's own torch and falls back to CPU automatically if there's no GPU.

One honest caveat: the pack README says this fork "may be removed if the upstream project merges the optimization changes." That's the standard fate of a perf fork - if Jordach merges PR #7 upstream, this node's reason to exist shrinks to the one-node convenience. It works fine today, but don't build a critical production workflow around a node that might get retired in favor of its upstream equivalent.