🐧 Sigma Insert
Add a denoising step exactly where the schedule needs one
- sigmas
- SIGMAS
A sigma schedule is the list of noise levels the sampler walks down - the scheduler's output, one value per denoising step, descending to zero. When you use a CustomScheduler node you're effectively hand-authoring that list, and the standard complaint is that the stock schedulers never put a step where you want it. This node is surgical editing for that list: insert one extra noise value at a chosen index, and you've added a denoising step at exactly that point in the curve.
It's part of the 🐧 SuperNodes pack by SuperCC, one of six sigma-editing nodes in the pack's Scheduling category.
How it works
- sigmas - the schedule tensor you're editing.
- index - where to insert, meaning "preceding this position."
0is the front of the schedule.-1is special: it inserts just before the first0.0- i.e. between the last real step and the final clean step - or appends at the end if there's no0.0. That-1behavior is the one you'll actually use. - value - the sigma to insert.
The one rule that matters: the inserted value must be smaller than the sigma before it. Sigmas describe a strictly descending noise curve, so a value that's out of order isn't just ugly, it breaks the sampler's assumption. If you insert a value ≥ the previous sigma, the node raises a clear error rather than quietly corrupting your schedule. So for the -1 case (inserting before the final zero), pick something small - a gentle extra step, not a second full-denoise.
Output is the modified SIGMAS tensor, which feeds straight back into the sampler's sigmas input.
When you'd reach for it
- Add a refinement step near the end of the curve, where schedules tend to jump from a small noise level straight to clean.
- Add a "warm-up" step at the very front (index 0) - though note the value still has to be below the next sigma, so this nudges rather than extends the top of the curve.
- Experimenting with step placement without rebuilding a whole custom schedule.
Install
ComfyUI Manager, search "ComfyUI-SuperNodes", or:
cd ComfyUI/custom_nodes
git clone https://github.com/sonnybox/ComfyUI-SuperNodes
Restart ComfyUI. No models to download. The pack's only extra dependency is matplotlib (used by the Sigmas Graph node, which is the natural debugging partner here - insert, then graph, then run). This pack targets ComfyUI's newer extension API, so keep ComfyUI updated.
Gotchas
- Every sigma you insert is one more step the sampler runs, which is linear cost in time. A couple of extra steps is cheap; don't reach for this to add ten.
- The monotonicity error is strict:
value >= prevfails. The tooltip's phrasing - "must be smaller than the sigma value before it" - is not advisory. - Editing schedules is most predictable on models that respect schedule shape. On flow-matching models the community has found aggressive schedule reshaping can hurt rather than help, so add steps conservatively and graph the result before trusting it.
It's a small, sharp tool. The -1 insert-before-zero behavior alone makes it worth remembering for anyone who's ever wanted "just one more gentle step at the end."
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| sigmas | SIGMAS | Input sigma schedule. | |
| index | INT | 0-1–10000 | Index of the sigma to insert preceding. 0 = first, -1 = before 0.0 (or at the end if 0.0 not present). |
| value | FLOAT | 1.000–10000 | Sigma value to insert. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| SIGMAS | SIGMAS | Sigma schedule with the value inserted. |