AdvancedNoise

Every generation starts from noise, and by default that noise is plain Gaussian - the random static you never think about. AdvancedNoise lets you swap it. It produces a NOISE object you feed into a custom-sampler setup in place of the default noise source, and it can generate fractal noise, pyramid noise, and a range of other distributions instead of the standard bell curve.

Why bother? Because the noise you start from biases what the model finds in it. Fractal and pyramid noise carry structure across scales, which tends to encourage more detail and a different distribution of texture than flat Gaussian - a known trick for pushing extra crispness out of a model that otherwise renders a bit soft. It's a niche lever, but it's a real one, and it's the kind of low-level control RES4LYF exists to expose.

How it works

The node builds an initial noise tensor of the type you choose and hands it out as NOISE. That output goes into the noise input of a custom sampler (ComfyUI's SamplerCustomAdvanced, or RES4LYF's own samplers), which then uses it as the starting point instead of generating its own. The alpha and k parameters shape the character of the fractal/structured noise; noise_seed makes it reproducible.

The inputs and outputs that matter

• noise_type - the distribution: fractal, gaussian, uniform, and a family of pyramid-* and hires-pyramid-* variants, 17 in total. gaussian reproduces the standard behavior; fractal and the pyramid types are where the detail-shaping happens.
• alpha (FLOAT, default 1) - shapes the fractal noise's frequency falloff, i.e. how much fine vs coarse structure it carries.
• k (FLOAT, default 1) - a second shaping factor on the noise character.
• noise_seed (INT) - seeds the generation so a given setup is repeatable.

The single output is NOISE - wire it into a custom sampler's noise input.

Installing it

ComfyUI Manager: search RES4LYF, install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/ClownsharkBatwing/RES4LYF/
cd RES4LYF
pip install -r requirements.txt

pip install in the venv or portable Python, restart, hard-refresh (F5). No downloads.

Where people get tripped up

First, this only does anything if you're using a sampler that accepts a NOISE input. A plain KSampler generates its own noise and ignores you - you need SamplerCustomAdvanced or a RES4LYF sampler wired to take an external noise source. If swapping the noise seems to change nothing, that's usually the reason.

Second, the fancy types are subtle, not magic. Fractal and pyramid noise shift the texture balance; they don't rescue a bad prompt or fix composition, and pushed hard some types add a grain or pattern you didn't want. Change noise_type alone, keep the seed fixed, and compare against plain gaussian so you can actually see what it bought you. If you can't tell the difference, plain gaussian is fine and you've lost nothing.