VisualizeSigmasKJ

Sigmas are the per-step noise levels a sampler walks down from full noise to a clean image. Every scheduler - karras, exponential, beta, whatever custom thing you've bolted together - produces a different curve, and until you see that curve it's all abstract numbers. Visualize Sigmas plots the schedule as an image so you can actually look at the shape. If you've ever wondered why a custom sigma schedule gave you weird results, this is the node that shows you what you built.

It's part of KJNodes, kijai's utility pack, filed under misc because it's a debugging and teaching tool rather than something in the generation path.

How it works

You feed it a SIGMAS input - the schedule coming out of a scheduler or a custom-sigmas node - and it draws that sequence as a graph in an image output. The clever bit for real use is that it also passes the sigmas straight through unchanged, so you can splice it inline between your scheduler and your sampler without altering anything: the sampler gets the same schedule, and you get a picture of it on the side. That makes it a zero-cost inspector you can leave in a graph.

The inputs and outputs that matter

• sigmas - the schedule to plot.
• start_step / end_step - crop the range you're plotting. end_step defaults to -1, meaning "to the end"; set these when you want to zoom in on part of the curve rather than the whole thing.

Outputs are sigmas_out (the untouched schedule, to continue into your sampler) and image (the plotted curve, to send to a preview). Wire sigmas_out onward and drop the image onto a preview node to watch it.

Installing it

ComfyUI Manager → search KJNodes for ComfyUI → install → restart. Or cd ComfyUI/custom_nodes && git clone https://github.com/kijai/ComfyUI-KJNodes, then pip install -r ComfyUI-KJNodes/requirements.txt, and restart.

Where it earns its place

• Custom schedules. If you're hand-building sigmas, splitting a run into high-noise and low-noise passes, or playing with the detail-schedule tricks people layer onto Wan and Flux, seeing the curve is the difference between tuning and guessing. A schedule that dumps most of its steps at high noise looks very different from one that lingers near the end, and now you can tell at a glance.
• It doesn't change anything. Because it's a passthrough, there's no risk to inserting it - the sampler runs the exact schedule it would have. Leave it in while you tune, pull it out (or don't) when you're done.
• It's a readout, not a fix. It tells you the schedule is weird; it won't fix it. Pair it with whatever node produced the sigmas to actually change the shape.