🐧 Sigma Remove

If Sigma Insert is the "add a step" tool in the 🐧 SuperNodes scheduling family, this is its mirror: remove one sigma value from the schedule. One step less means the sampler denoises one fewer time - a cheap, surgical speedup, or a way to skip a specific noise level that's misbehaving.

How it works

Two inputs: sigmas (the schedule) and index (which value to remove). Indexing is the friendly part - 0 is the first step, and negatives count from the end: -1 is the last, -2 second-to-last, and so on. That's the convention people actually use, because "the step just before zero" is a far more natural thing to ask for than counting forward from the top.

The node removes exactly that one element and returns the shortened SIGMAS. If the index is out of range it raises a clear IndexError rather than silently doing nothing.

When you'd reach for it

• Shaving a step from a long schedule to speed up renders. Each removed sigma is one fewer denoise pass, and removing one from a 30-step run is usually invisible to quality.
• Skipping a specific noise level that produces a banding artifact or a weird transition in your custom schedule. A single offending sigma is a real phenomenon with hand-built schedulers.
• Cleaning up a schedule that another node (or a previous edit) left with a redundant step - e.g. a near-duplicate pair of small sigmas where one clearly isn't doing work.

Install

ComfyUI Manager, search "ComfyUI-SuperNodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/sonnybox/ComfyUI-SuperNodes

Restart ComfyUI. No model downloads; the pack's only extra dependency is matplotlib (used by the Sigmas Graph node - handy for checking your schedule before and after removal). This pack is written against ComfyUI's newer extension API, so update ComfyUI if the nodes won't load.

Gotchas

• Removing a sigma genuinely removes a denoising step - it doesn't "merge" neighboring steps. If you remove a mid-schedule value, the sampler now jumps from the previous noise level to the next, which is a slightly different trajectory than running one fewer step with a rescaled schedule.
• Negative indexing is resolved against the current length, so if you edit a schedule, remove a step, and then remove by -1 again, you're removing the new last step - not the original last step. Edits compound; re-check after chaining.
• The schedule must still end in 0.0 for the sampler. If you remove the final zero you're handing the sampler a schedule without a clean end state, which is on you, not something the node guards against.

It's the simplest node in the family and the least likely to surprise you. One input, one index, one fewer step.