Merge sigmas gradually

Two schedulers, one handoff

Merge sigmas gradually from sigmas_tools_and_the_golden_scheduler is the more interesting cousin of the plain average merge. Where the average applies one constant blend weight to every step, this one ramps the blend as the schedule progresses: it starts dominated by sigmas_1, smoothly crosses over, and ends dominated by sigmas_2. The README's framing is the clearest: "like if you want to start with karras and end with simple."

Under the hood each step gets a blend factor that's linearly remapped from proportion_1 (at the first step) down to 1 - proportion_1 (at the last step), with

result[i] = sigmas_1[i] * factor + sigmas_2[i] * (1 - factor)

Same three inputs as the average merge: sigmas_1, sigmas_2, and proportion_1 (float, default 0.5, range 0–1). One SIGMAS output, wired into a KSamplerAdvanced's sigmas input like any external schedule.

Where it shines - and the trap

This is a genuinely nice way to get "good structure up front, gentle finish" out of a single KSampler pass. Early steps set the composition, so starting with your sharper scheduler there keeps that phase decisive; late steps refine detail, so finishing on a gentler one avoids the harsh end that aggressive curves can produce. It's a scheduler version of doing a two-phase render without actually running two renders.

The trap is proportion_1 at its default of 0.5. Because the ramp endpoints are proportion_1 and 1 - proportion_1, setting it to 0.5 makes both endpoints 0.5 - the "ramp" is flat, and you've just built a plain average, defeating the whole point. Crank it toward 0.8 or 0.9 and the handoff becomes real. Start debugging there if the blend feels like it's doing nothing.

Both schedules still need the same step count, since the merge is element-wise - mismatch the lengths and you'll get an error rather than a resized schedule.

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. Standard pack gotcha: requirements.txt only lists asteval while the code imports matplotlib and scipy - pip install matplotlib scipy asteval fixes a minimal install that errors on import.