Visualize Sigma Schedule

Every scheduler you've ever picked - Karras, exponential, the fancy custom one you found in a workflow - is really just a shape drawn across the sampling steps. That shape is the sigma schedule: the amount of noise removed at each step, from "a lot" at the start to "a little" at the end. And most of the time you're choosing it blind. Visualize Sigma Schedule is the node that finally draws the picture.

Wire a SIGMAS tensor into it and it renders a chart of the schedule in the node's preview panel. That's the whole job, and it's genuinely useful once you've seen it, because the difference between schedulers is completely obvious in plot form - and often surprising. Your "smooth" custom schedule might have a brutal cliff in the first two steps; a Karras curve shows exactly why it spends so many steps at low noise.

How it works

The node takes the sigmas output from any scheduler (the same SIGMAS type you already pipe into a sampler), plots noise level against step index with matplotlib, and pushes the chart to the UI as a temp preview. No file is saved to your output folder - it goes to ComfyUI's temp directory and lives in the preview, same RAM-preview philosophy as the rest of the pack.

Where it shines

• Comparing schedulers. Build a couple of scheduler loaders, visualize them side by side, and the "why does exponential behave differently" question answers itself.
• Debugging custom schedulers. WtlNodes ships a Dual Ease Cosine Scheduler and there are endless community sigma generators; the fastest way to catch a bad curve (sigmoid shape, non-monotonic, huge first jump) is to plot it. A sigma schedule that isn't monotonically decreasing is almost always a bug.
• Understanding what "start_percent/end_percent" actually mean. If you use the pack's Cache Accelerator or any percent-based tooling, this is how you see which part of the schedule that window covers.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Scorpiosis0/ComfyUI-WtlNodes.git

Restart ComfyUI, or search "WtlNodes" in ComfyUI Manager. The pack's deps are just numpy, scipy, pillow - plus matplotlib for this node, which is a standard ComfyUI dependency anyway.

Gotchas

• It's an output node with no outputs. It terminates a wire, so put it after a scheduler and don't expect it to pass sigmas onward - branch off the scheduler for the sampler path.
• The preview is temp-only. If you want the chart as a file, screenshot it; the node isn't built to save.
• Big schedules make busy charts. A 40-step schedule renders fine, but it's a line, not a histogram - don't expect per-step detail at high step counts, just the overall shape. Which is, frankly, what you need.