RerangeSigmas

RerangeSigmas is a tiny sigma-utility node from the SpliceTools pack (AustinMroz), and it solves one genuinely annoying problem: you've built a custom-sampling graph with SplitSigmas handing different ranges to different samplers, and now you want one of those samplers to run at a different step count. Normally that means editing the scheduler that feeds the whole graph - which reflows every branch and marks half your workflow stale. RerangeSigmas lets you take the sigmas you already have and re-grid them onto a new number of steps, keeping the same sigma_max and sigma_min. In other words: resize one stage's schedule in place.

How it works

You feed it two things: a sigmas (SIGMAS) wire from wherever your schedule currently comes from, and a steps INT (default 10, minimum 1). It reads the endpoints of the incoming schedule - the first sigma and the last one - then rebuilds an exponential schedule between them at your new step count. If your schedule ends in a real 0 (a full-denoise pass), it handles that correctly instead of producing a degenerate last step.

That's the whole mechanism. It doesn't invent a new range or a new scheduler - it just resamples the existing one onto a new grid, which is exactly why it slots into a split-sigmas graph without disturbing the rest. The author's own note calls the scheduling a TODO, so for now it's exponential and it matches what most stock schedulers do anyway. If you need a different sigma distribution you're better off swapping the scheduler; this node is about step counts, not curves.

What you actually do with it

The README's framing is right: combined with SplitSigmas, it's the clean way to change the number of steps a single sampling node gets. In a two-pass graph (say a low-noise refinement pass and a main pass), drop RerangeSigmas on the branch you want to retune, set steps to the new count, and leave the scheduler alone. No stale-flagging, no rewriting the whole graph.

One honest caveat: this is a niche tool. If you're not doing custom sampling with split schedules yet, you don't need it - a plain KSampler's step count is one field. It earns its place the moment you've got a graph where the schedule and the samplers stopped being one-to-one.

Installing it

It ships in the SpliceTools pack, which is about as light as custom nodes get - no extra Python dependencies and no model downloads, just pure torch and ComfyUI's own helpers. Install via ComfyUI Manager by searching SpliceTools, or:

cd ComfyUI/custom_nodes
git clone https://github.com/AustinMroz/ComfyUI-SpliceTools

Restart ComfyUI, and you'll find it under sampling → custom_sampling → sigmas in the node menu.

Gotchas

• The steps you ask for is what you get in the output, but remember the source schedule's endpoints are preserved - if the incoming sigmas had been through a split, you're re-ranging a piece of the curve, not the whole thing. Feed it the sigmas you actually want rescaled.
• Since it rebuilds an exponential grid, don't use it to add fine detail steps at the end of a Karras-style curve and expect the Karras distribution back. For that you'd regenerate the schedule properly.

Experimental, unmaintained-feeling, and genuinely handy for one specific pattern. If that pattern is yours, it's the easiest tool in the pack.