Merge many sigmas by average

The kitchen-sink scheduler

Merge many sigmas by average from sigmas_tools_and_the_golden_scheduler is the same idea as its two-input sibling, scaled to the point of absurdity: it averages up to 25 SIGMAS tensors into one schedule, all at equal weight. Wire in every stock scheduler you can think of - karras, simple, exponential, sgm_uniform - and it returns their plain average, a sort of "the community's noise schedule" curve.

That's the whole mechanic, and it's worth being precise about it: equal weights, no knobs. The node's single real input is sigmas_1 (required); sigmas_2 through sigmas_25 are all optional, so you plug in whatever you have and the rest are ignored. The average is element-wise across all the provided tensors, so every schedule you connect must have the same number of steps - torch stacks them and takes the mean, and mismatched lengths will error.

One SIGMAS output, wired into a KSamplerAdvanced's sigmas input like any schedule.

When it's actually useful

Honestly? Less often than you'd think. Averaging a pile of schedulers tends to smooth out all their distinctive features into something bland - that's the nature of an average. But there are two legit uses. First, as an experiment: if you can't pick a scheduler, average them all and see what a "middle of the road" schedule does on your model. Second, as a baseline: render with the average, then compare against your favorite single scheduler, and the differences will tell you precisely what that favorite is contributing. If you find yourself wanting some of it but not all, the pack's plain Merge sigmas by average gives you a weight knob so you can lean a blend toward one side instead of taking a flat mean.

For the per-step version - start the image on one schedule, finish it on another - look at Merge sigmas gradually or Split and concatenate sigmas, both in the same pack.

Install & gotchas

ComfyUI Manager → search sigmas_tools_and_the_golden_scheduler, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Extraltodeus/sigmas_tools_and_the_golden_scheduler

Restart, no model downloads. Standard pack gotcha: requirements.txt lists only asteval while the code imports matplotlib and scipy, so pip install matplotlib scipy asteval if the pack won't load on a minimal install. And remember the step-count rule: 25 inputs is fun, but they all need to be the same length or you'll get a tensor-stack error before the node ever does anything useful.