Concat Sigmas

ConcatSigmas does exactly one thing: takes two SIGMAS arrays and sticks them end to end. That's it. It's the node you grab when you want a sampling run that behaves differently in its first phase than its second - the denoising equivalent of "start aggressive, finish careful."

Why you'd actually want this

Most of the time you don't need a custom sigma schedule at all, and the stock KSampler scheduler is fine. But two-phase runs are a real thing. Maybe you want the first 40% of steps on a fast, coarse schedule and the last 60% on a slower, finer one - more steps per unit of progress where the detail lives. Or you want to prepend a few "warmup" steps to a main schedule. ConcatSigmas is the glue that lets you build those composite runs from two separately-generated schedules instead of hand-editing sigma lists.

It's also the natural bridge inside this pack: the pack's schedulers generate curves, and ConcatSigmas lets you splice a custom warmup curve onto a stock schedule before feeding the result somewhere that wants a full run.

How it works

Literally torch.cat([sigmas1, sigmas2]). Both inputs are SIGMAS tensors; the output is their concatenation in order - first all of sigmas1, then all of sigmas2. There's no deduplication, no interpolation, no magic. If your first schedule ends at a low sigma and your second starts high, you'll get a visible jump at the seam; if you're concatenating a coarse first phase with a fine second phase, that's usually fine because samplers read sigma values, not their index positions.

The inputs and output

• sigmas1 / sigmas2 - the two schedules to join, in that order.
• Output: SIGMAS - the combined array, ready to feed a KSampler's scheduler slot, a guider's curve input, or another sigma node.

Install

Part of ComfyUI-ScheduledGuider-Ext. Easiest via ComfyUI Manager - search ComfyUI-ScheduledGuider-Ext and install. Or:

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

Restart ComfyUI and you're done - no models, no extra dependencies.

Common issues

The only real footgun is ordering. sigmas1 comes first, and samplers expect their sigma schedule to run from high to low. If you concat a low-sigma second half onto a high-sigma first half and the jump looks wrong, check which end is which - swapping the inputs is a one-wire fix.

Also note this node works on schedule values, not step counts. Concatenating two 20-step schedules gives you a 40-step run, and your sampler needs to actually run that many steps or the tail gets ignored. If your sampler is fixed at 20 steps, you're throwing half the concatenation away.