LatentNoiseBatch_fractal

Most people never think about the noise a sampler starts from beyond "it's random" - but RES4LYF's own README makes a point of listing 20 different noise types available inside its main sampler, because the starting noise itself is a real lever on quality and character, not just an implementation detail. LatentNoiseBatch_fractal takes one of those noise types - fractal (colored) noise, as opposed to plain white gaussian noise - and exposes it as a standalone node, so you can generate it directly rather than only picking it from a dropdown buried inside the sampler.

Fractal noise, sometimes called colored or 1/f noise, has structure across frequencies instead of being uniformly random the way white noise is - think static with a texture to it rather than pure snow. Different samplers and schedules respond differently to what they start from, which is the whole reason RES4LYF bothers to expose so many noise options in the first place.

What you set

• latent (LATENT) - required, the latent shape to generate noise for.
• alpha (FLOAT, default 1) - controls how the noise's energy falls off across frequencies. Higher values push toward smoother, more low-frequency-dominated noise; lower values push toward something closer to flat white noise. Treat it as a texture dial rather than a precise, documented setting - the pack doesn't spell out exact behavior at each value, so the honest approach is to nudge it and look.
• k_flip (BOOLEAN, default false) - flips which end of the frequency spectrum gets emphasized. Exact effect isn't documented; try it both ways on a test run if alpha alone isn't giving you the character you want.
• steps (INT, default 0) - likely the number of layered passes (octaves) that go into building the fractal pattern, consistent with how fractal noise is typically generated elsewhere.
• seed (INT) - for reproducibility, same role as a sampler seed.
• Optional alphas, ks, steps_ (SIGMAS) - schedule any of the above across steps instead of holding them fixed, matching RES4LYF's broader "modulate parameters against time" design.
• Output: LATENT - the generated noise, ready to feed as a starting point into your sampler in place of the default.

Installing it

Part of the full RES4LYF pack:

• ComfyUI Manager - search "RES4LYF," install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/ClownsharkBatwing/RES4LYF, then cd RES4LYF && pip install -r requirements.txt, restart.

No extra downloads for this node.

Common issues

Not sure it's doing anything different from ordinary noise. Fractal noise's effect is subtle at moderate alpha values and much more visible at the extremes - if you're not seeing a difference, push alpha further from its default and compare against a plain gaussian start (the pack's LatentNoiseBatch_gaussian sibling) side by side.

Results look worse, not better. Colored noise isn't a universal improvement - it's a different starting condition that suits some samplers and subjects better than others. If it's hurting rather than helping, that's useful information, not a sign the node is broken; go back to standard noise for that particular combination.

Confusing this with the sampler's built-in noise type dropdown. If your ClownsharKSampler already has a noise type selected, this node's output is a separate, explicit noise latent you're substituting in - make sure you're actually wiring it to override the sampler's own noise generation rather than running both and getting an unexpected combination.