PredefinedExponent

PredefinedExponent applies base^sigma to every value in a sigma schedule, with the base chosen from a dropdown: e, 10, or 2. It's the inverse operation of a logarithm - apply PredefinedLogarithm to a schedule and then this node with the same base, and you get the original values back. It's a transform, not a generator: you feed it a schedule and it returns the exponential of each entry.

Why would you want that? Because "schedule" doesn't have to mean "the noise values themselves." People routinely work with schedules in a transformed space - the log of sigma, say - because it spreads out a compressed range so you can actually see and edit the shape. When you're done tweaking in log space, this node is how you translate back. The pack pairs the two nodes explicitly for that round-trip: the README's own framing is "ideal for reverting logarithmic transformations applied to sigma schedules."

How it works

For each sigma value it computes base^sigma. Base e uses the natural exponential; base 10 is a straight power of ten; base 2 uses exp2. The practical difference is just the magnitude of the swing - 2^x stays modest for small inputs, 10^x gets huge fast, and e^x sits in between. There's no rescaling and no clamping after the transform, which matters a lot when you see what happens to real schedules.

The inputs that matter

• sigmas - the schedule to transform.
• base - dropdown: e, 10, or 2. Default e.
• 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

This node is a scaler's nightmare if you feed it the wrong thing, because exponentials explode. A real sigma schedule in the 0.03–14.7 range, pushed through 10^x, produces numbers like 10^14 - garbage that no sampler can use, produced without a single error message. The sane use is on schedules that live in log space already, or small 0–1 curves where the transform is contained. If the output looks absurd, check what space your input was actually in. And as always in this pack: when a transformed schedule is headed for a real sampler, finish with ScaleToRange so the values land where the model expects them.