Split and concatenate sigmas

The franken-schedule node

Split and concatenate sigmas from sigmas_tools_and_the_golden_scheduler is the blunt instrument of schedule mixing: it takes the first N steps from one schedule and stitches on the remaining steps from a second schedule. No blending, no crossfade - a hard cut at a step you choose. If you like how exponential opens an image but prefer simple's landing, this is the node that gives you exactly that.

Under the hood it's a straight tensor cut and join:

torch.cat((sigmas_1[:sigmas_1_until], sigmas_2[sigmas_1_until:]))

Everything from sigmas_1 up to and including step sigmas_1_until is kept; the rest of the schedule comes from sigmas_2.

What you set

Four inputs:

• sigmas_1 / sigmas_2 - the two schedules
• sigmas_1_until - an integer (default 10) marking the cut point: how many leading steps come from schedule 1
• rescale_sum - a boolean, default off. Flip it on and the result is scaled so its total matches sigmas_1's total, which keeps the overall noise budget consistent when you're borrowing a tail from a very different schedule

One SIGMAS output, wired into a KSamplerAdvanced's sigmas input like any external schedule.

The honest take

This is the most surgical of the pack's merge nodes, and the cut point is doing all the work. Early steps shape composition, so keeping those from your favorite "structure" scheduler and letting the tail from a gentler one handle the finish is a clean way to get two schedulers' best halves in a single pass. It's also a good debugging tool: cut a problem schedule at various points and swap in a known-good tail to isolate which phase of the schedule is producing a bad render.

Two practical things to watch. First, sigmas_2 needs to be long enough to cover the tail - if it's shorter than the remainder, the joined result will be shorter than you expect and the sampler's step count won't line up. Second, this is a hard cut, not a smooth transition; if you want a handoff that ramps gradually instead of snapping, the pack's Merge sigmas gradually does that.

Install & gotchas

ComfyUI Manager → search sigmas_tools_and_the_golden_scheduler, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Extraltodeus/sigmas_tools_and_the_golden_scheduler

Restart, no model downloads. Pack-level gotcha applies: requirements.txt lists only asteval while the code imports matplotlib and scipy, so a minimal install that errors on import needs pip install matplotlib scipy asteval.