ComfyUI Node

SigmasToPower

Reshape a schedule by raising every value to a power

By mfg637·Created 2 years ago·Updated about a year ago· 11
SigmasToPower
  • sigmas
  • SIGMAS
power2.0

SigmasToPower raises every sigma value in a schedule to a power you choose - σ becomes σ^power. It's a one-line math node, but it's the classic way to non-linearly reshape a curve: powers above 1 stretch the high end and compress the low end, powers below 1 do the opposite. A flat-ish schedule gets a meaningful low-end vs high-end split; a curve you like gets its distribution of effort bent in one direction or the other.

Where would that actually matter? Consider a CFG curve for a guider. Its shape is the entire point - where guidance ramps, holds, and fades. Squaring that curve pushes the values toward 0 (the low end) unless the original was already near 1, and a square root pulls them up. It's a coarse, predictable "bend this curve" knob you can apply to any schedule without rebuilding it from scratch.

How it works

For each sigma it computes math.pow(sigma, power) and rebuilds the schedule. That's it. There's no normalization afterward - if your input lives in 0–1, raising to a power keeps it in 0–1 but changes the distribution; if your input is a real-world schedule like 0.03–14.7, powers rapidly produce out-of-range values the node won't clamp. The power input has no stated min/max, so the guardrails are entirely up to you.

The inputs and output

  • sigmas - the schedule to transform.
  • power (2.0, step 0.1) - the exponent. Above 1 emphasizes the high end; below 1 emphasizes the low end.
  • Output: SIGMAS - the transformed 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

Watch your range, because this node won't. Raising real-world sigma values (say, up to 14.7) to power 2 gives you values in the hundreds - nonsense for a sampler, silent and undetectable unless you're looking. This transform behaves sanely on 0–1 curves (guider curves, mostly) and needs a ScaleToRange pass after it if you're feeding a real schedule.

Also keep the CFG-1 era in mind: reshaping curves is a fun toy, but on guidance-distilled models where CFG is fixed near 1, all this bending is aimed at a dial that's already been taken away. On SD 1.5/SDXL-lineage models, though, it's a legitimate tool for tuning how guidance effort distributes across the run.

Categorysampling/custom_sampling/sigmas

Inputs (2)

NameTypeDefaultDescription
sigmasSIGMAS
powerFLOAT2.0

Outputs (1)

NameTypeDescription
SIGMASSIGMAS