OffsetSigmas

OffsetSigmas adds a constant to every sigma value in a schedule. That's the entire job: one SIGMAS input, one offset number, output is your schedule shifted up or down by that amount. No scaling, no reshaping, no cleverness - a straight translation.

It's the kind of node that looks useless until you need it. The real-world use is adjusting where a schedule's values actually land without changing its shape. Say you've got a curve you like but its low end bottoms out at zero, and a sampler is treating that as "fully denoised" too early - a small positive offset lifts the whole floor. Or your CFG curve for a guider needs to sit in a slightly different part of the range while keeping its silhouette. That's an offset's job.

How it works

Loops over every sigma and adds offset to each. The offset input has a step of 0.1 and a generous range (you can go negative to push a schedule down). Crucially, it does not clamp anything: add a big enough positive offset and your sigmas can climb past what your model was trained to handle, and the node won't warn you. It trusts you.

The inputs and output

• sigmas - the schedule to shift.
• offset (1.0, step 0.1) - the amount added to every value. Negative values shift down.
• Output: SIGMAS - the shifted array, same length and shape.

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

The lack of clamping is the trap. Push a schedule up too far and you've invented sigma values your checkpoint never trained on, which tends to produce either washed-out results or an error when the sampler's internal checks trip. If an offset-fed run starts misbehaving, halve the offset and test again.

Also worth remembering: because it's a translation, not a scale, a big offset flattens the relative shape of a schedule. Adding 10 to a curve that ranges 0–2 makes it range 10–12, and the fine differences between steps become proportionally tiny. If you want to change the range while preserving relative variation, ScaleToRange is the right tool; OffsetSigmas is for nudges.