SplitSigmasByValue

SplitSigmasByValue is the pack's most workflow-shaped node: it cuts one sigma schedule into two at a threshold you choose. Everything above the cutoff goes to one output, everything at or below it goes to the other. It's the plumbing that lets you run the "high noise" part of a schedule through different machinery than the "low noise" part.

What would you do with that? The canonical idea is phase-based sampling: high-sigma steps build structure and composition, low-sigma steps refine detail. Run the high half through one sampler or CFG treatment and the low half through another, and you've got a two-stage pipeline without hand-counting which step is which. ComfyUI is uniquely suited to this kind of graph-building, and this node is the seam you split on.

How it works

It scans the schedule and finds how many values are greater than value, then slices the array at that point. high_sigmas is everything above the threshold, low_sigmas is the rest - and the output names are exactly what the brief says: high_sigmas and low_sigmas. Note there's a subtlety in the slice: the two outputs actually share one boundary value, so there's a one-element overlap rather than a clean divide. In practice that just means a single step appears in both halves, which is harmless for most uses but good to know if you're summing step counts.

The inputs and output

• sigmas - the schedule to split.
• value (1.0) - the cutoff. Default 1.0; the range extends to an absurd 2^60, so effectively you can place it anywhere your sigmas go.
• Outputs: high_sigmas and low_sigmas, two SIGMAS arrays.

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 overlap is the one thing that'll surprise you if you're counting steps - the two halves are not strictly disjoint. If you concatenate them back together you won't get the original schedule; you'll get a duplicate step at the seam.

The bigger planning note: this node splits by sigma value, not by position. A schedule that spends most of its steps at high sigma and drops fast will produce lopsided halves, which is usually what you want but worth checking if you expected an even 50/50 cut. And when you do run two different sampler treatments on the halves, remember each half still needs a full sampler pass with its own noise handling - the node only divides the schedule, it doesn't divide your workflow for you.