SamplingPercentToSigma (Ray)

Every custom-sampling workflow eventually needs the same bridge: "I think of guidance as 'first 15% of denoising,' but the sampler thinks in sigma values." RaySamplingPercentToSigma is that bridge for Raylight - it converts a 0–1 sampling percent into the model's sigma value, using the model's own sampling object.

The mechanism is a one-liner under the hood: model_sampling.percent_to_sigma(sampling_percent). The model already knows how its own noise schedule maps percent to sigma, so the node just asks it. Two inputs:

• sampling_percent (default 0.0) - 0.0 means "the very start of denoising" (maximum noise), 1.0 means "the end" (minimum noise). The output floats accordingly.
• return_actual_sigma (default false) - this is the subtle one, and the author's own tooltip spells out the gotcha: it returns the actual sigma value instead of the value used for interval checks. The difference only shows up at exactly 0.0 and 1.0 - at those endpoints, percent_to_sigma returns sentinel-ish values that work for interval comparisons but aren't the literal sigma_max/sigma_min. Flip this on and the node returns the true sigma_max at 0.0 and true sigma_min at 1.0. Leave it off if you're feeding the output into something doing range checks (like a sigma-threshold comparison).

Output is a single FLOAT (sigma_value). It's a utility, not a patch - it doesn't touch the model at all.

Where it fits

Mostly as plumbing for the pack's own guidance tooling. The natural use: compute a sigma threshold from a percent and feed it into a guider's sigma-based logic (e.g., "apply this only below 0.15"), or build a custom schedule by converting several percents and feeding the floats into list-driven nodes. It's in the custom_sampling/schedulers category, so think of it as a helper you chain before constructing your sampler/guider, not a node that runs during sampling.

Install

Part of raylight:

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

Restart, or install "raylight" via ComfyUI Manager.

Gotchas

• The return_actual_sigma toggle exists precisely because people hit the 0.0/1.0 discrepancy and thought the node was broken. If you're debugging "why is my sigma 0 at 0%", that's the default interval-check behavior - either flip the toggle or don't sample exactly at the endpoints.
• It needs a valid ray_actors input because the model lives on the workers - no Ray Init Actor upstream, no sigma. If you just want a standalone percent→sigma math helper on a single GPU, the value is model-dependent anyway, so you'd use ComfyUI's own model-sampling node instead.