GaussianScheduler

Most noise schedules are boring on purpose - they remove a steady or smoothly-varying amount of noise per step, because that's what DDPM-style models learned to expect. GaussianScheduler is not that. It shapes your sigma values as a Gaussian (normal) bell curve, which means the "middle" of your run carries the action and the two ends get comparatively little. It's the scheduling equivalent of "save your energy for the interesting part."

If that reminds you of why the Karras schedule became famous - it concentrates denoising effort in the middle steps on SD 1.5/SDXL - you've got the right mental model. This node just lets you dial that concentration precisely, with a mean and a standard deviation, instead of taking Karras as a fixed package.

How it works

It evaluates a normal distribution PDF across an x-window you define with a and b, then rescales the resulting curve so it fits between sigma_min and sigma_max. The mean positions the bump within that window, and standard_deviation controls how wide it is - small stdev, sharp narrow bump with most steps doing almost nothing; large stdev, a broad gentle dome. Since it's a PDF, the curve is tallest in the middle and trails off at both ends, and the rescale step means the endpoints always land exactly on your chosen sigma_min/sigma_max.

The inputs that matter

• steps (20) - number of sigma values.
• sigma_max (1.0) / sigma_min (0.0) - endpoints. The usual 0–1 defaults.
• mean (0.0) - where the bump sits in the x-window. Push it toward a to front-load the action, toward b to back-load it.
• standard_deviation (1.0) - bump width. This is the one you'll actually tweak.
• a (0.0) / b (1.0) - the x-window the Gaussian is sampled over. If a > b they're swapped for you; if they're equal the node raises an error, so don't.

Output is a SIGMAS array.

Install

Part of ComfyUI-ScheduledGuider-Ext. ComfyUI Manager - search ComfyUI-ScheduledGuider-Ext - or:

cd ComfyUI/custom_nodes
git clone https://github.com/mfg637/ComfyUI-ScheduledGuider-Ext

Restart ComfyUI. No models, no extra dependencies.

Common issues

Same trap as the pack's other schedulers: the default range is 0–1, which is perfect for a guider's CFG curve and useless as a raw sampler schedule. Rescale with ScaleToRange before it hits a KSampler.

One thing worth knowing before you chase a bell-curve schedule too hard: on flow-matching models, aggressive reshaping of the sigma curve is exactly what made Karras fail, and a sharp Gaussian bump is aggressive. This node earns its keep on SD 1.5 / SDXL-lineage models and as a CFG-curve source. On Z-Image or Flux-style flow-matching checkpoints, a balanced, conservative schedule is usually the safer bet.