Sigmas ZetaEta

The Riemann zeta function, the Dirichlet eta function, and the Lerch transcendent are real functions from analytic number theory - the kind of thing that shows up in serious mathematics, not typically in a noise schedule. This node runs your sigma values through whichever one you pick, evaluated as a truncated series rather than a closed form, and hands back whatever comes out.

It's the number-theory sibling of Sigmas ReactionDiffusion: a genuinely obscure transform with no established use case, offered because the math is interesting to have available, not because a specific model or workflow calls for it. If you're not the type who wants to explore what an infinite-series function does to a decay curve out of curiosity, there's no downside to skipping this node entirely - nothing else in RES4LYF depends on it.

How it works

sigmas is the schedule to transform. function picks which of the three to apply: riemann_zeta, dirichlet_eta, or lerch_phi. Because these are infinite series in their exact mathematical form, the node approximates each one using a finite number of terms - approx_terms (default 100) controls how many; more terms means a closer approximation and more compute per value. offset and scale let you shift and stretch the input before the function is applied, which matters a lot here since these functions behave very differently across different input ranges. normalize_output (on by default) rescales the result back into a workable range afterward.

The inputs and outputs that matter

sigmas, function, offset, scale, normalize_output, and approx_terms are all required. One output: SIGMAS.

Installing it

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

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

Use the portable build's embedded pip.exe if that applies. Restart and hard-refresh the page afterward.

Where people get tripped up

offset and scale matter more here than they do on most of RES4LYF's other transform nodes, because functions like the Riemann zeta have specific input regions where they behave wildly or aren't well-approximated by a short series. If the output looks like noise rather than a coherent curve, try adjusting offset/scale to move your input values into a friendlier range, or raise approx_terms for a more faithful approximation, before assuming the node is broken.

There's no guarantee this produces a monotonically decreasing curve the way a standard schedule does - check the actual output (a plot or preview node) before wiring it straight into a sampler, the same caution that applies to the rest of RES4LYF's experimental sigma generators.