Choose SplitSigmas when the boundary must stay at a specified sequence index. Choose SplitSigmasDenoise when the boundary should be determined from a denoise proportion and the sequence’s number of steps. Before choosing, write down what the boundary means in your workflow and where each output should go. These nodes split sigma sequences; this is not a choice between two image-strength sliders.
Decide what the boundary means, then connect it
Use a fixed step when a downstream stage must begin at a particular index, even if you later change the denoise setting elsewhere. Use denoise when you want the split position to be derived from a proportion of the available steps. The official documentation describes that calculation as the total number of steps, one less than the number of sigma values, multiplied by the denoise factor. It does not specify here how any fractional result is rounded or handled at the endpoints.
The nodes also describe their outputs differently: SplitSigmas returns two sigma sequences, while SplitSigmasDenoise labels them high_sigmas and low_sigmas. Do not infer output identity from node color alone. Confirm the actual socket type labels and connect each sequence to the intended downstream stage.
- Identify the input
SIGMASsequence and record its length as shown by a reliable workflow or inspection method. Do not assume the step count equals the number of sigma values. - State the intended boundary in plain language, such as “the second stage starts at the fixed index I” or “the split follows proportion D of the steps.”
- Choose the matching node and enter the
stepordenoisevalue. The documented denoise range is 0.0–1.0, in increments of 0.01. - Connect both output sockets to their named stages, checking the socket labels and receiving input types.
- Record the observed split and output lengths for this workflow. If the observed boundary differs from your intended one, inspect the sequence and connections before changing other sampling settings.
Boundary planning and acceptance worksheet
This worksheet is an editorial comparison method, not a promise about exact output sizes or rounding. Suppose, hypothetically, an input has 13 sigma values, so the documented “one less than the number of sigma values” wording gives 12 steps. A denoise value of 0.50 corresponds to a calculated position of 6 before any unspecified handling of that result. A fixed step of 4 instead expresses a boundary at the chosen index. Record what the installed workflow actually outputs; the documentation does not establish whether every sequence treats the boundary index as an included element in the same way.
| Input length | Requested boundary | Does it follow a changed denoise value? | Observed outputs and destinations | Decision |
|---|---|---|---|---|
| 13 values, 12 steps (hypothetical) | Fixed step 4 | No | Record both output sequences; route each to its planned stage | Keep this node if index 4 is the required boundary |
| 13 values, 12 steps (hypothetical) | denoise 0.50; calculated position 6 | Yes, by the documented proportional rule | Record both output sequences and the actual split shown by the workflow | Keep this node if the boundary should track the proportion |
| Any observed length | Desired index or proportion not confirmed | Unknown | Mark the unexpected output or connection; verify socket types and sequence input | Resolve the mismatch before interpreting downstream results |
Limits and what to do when it fails
The documentation defines the input and boundary controls, but does not fully describe index inclusivity, rounding of a calculated position, or behavior at every extreme value. Do not turn the hypothetical calculation into a guaranteed count for either output. Inspect the resulting sequences in your actual workflow, particularly when exact downstream compatibility matters.
If the output appears empty, reversed, or routed to the wrong stage, first verify that the node received the intended SIGMAS sequence and that both receiving sockets accept the displayed type. Then compare the observed outputs with your worksheet and confirm whether you chose a fixed index or a proportion intentionally. If the boundary convention remains unclear, consult the current node documentation or implementation for your installed ComfyUI version; do not infer image quality from the split alone. This guide describes ComfyUI node behavior, not an Imagild feature.
Common questions
Does `denoise` mean the image will look 50% denoised? No. Here it is an input used to determine a sigma-sequence split position. The documentation does not define it as a guarantee about visual appearance.
Should I switch to denoise when the sequence length changes? Only if the boundary should continue to represent a proportion of the available steps. If the required boundary is a particular index, keep using `step` and review whether that index is still appropriate.
Can the two nodes be compared by output size alone? Output sizes can help you record what happened, but they do not establish that one split is better. Use the intended boundary and correct downstream connections as your acceptance criteria.
Sources
Sources and next step
- SplitSigmas - ComfyUI Built-in Node Documentation
- SplitSigmasDenoise - ComfyUI Built-in Node Documentation
This guide interprets public documentation and uses explicitly hypothetical acceptance examples, not application tests. Controls and availability can change; verify the cited manual against your installed version.
For work created in Imagild, carry the input requirements and acceptance checklist into the photo editor. The external software controls discussed above belong to their respective products.