Open-H3 IC · Sigma Grid:
OpenH3ICSigmas creates a noise-level schedule for the Open-H3 Regenerate 2K IC workflow. A noise-level schedule, also called a sigma grid, defines the sequence of noise levels that a compatible sampling process follows as it refines a latent representation toward a finished result. This node lets you use a named preset or define your own sequence, while checking that the values stay within the supported range and do not increase as the sequence progresses. It helps you reproduce the schedule used in the supplied regeneration workflow or experiment with a controlled alternative without manually constructing a tensor. The node prepares the schedule only; it does not generate images, upscale latents, or run sampling itself. In ComfyUI, its display name is “Open-H3 IC · Sigma Grid,” under the “Open-H3/Regenerate 2K IC” category.
Open-H3 IC · Sigma Grid Input Parameters:
sampler
A required selection that chooses the preset noise-level schedule when you leave custom_sigmas empty or enter only whitespace. The available options are up3, up4, and hf8, and the default is up3; numerical minimum and maximum values do not apply. The supplied workflow uses up3, whose schedule is approximately 0.834942, 0.696852, 0.468753, 0.0, giving three transitions between four noise levels. These transitions are counted as three steps in the report. The provided context does not include the schedule definitions for up4 or hf8, so their exact values cannot be documented here. Despite its name, this control selects a sigma preset, not the sampling algorithm used elsewhere in your workflow. If you provide a nonblank custom schedule, that schedule takes priority and this selection does not determine the resulting values.
custom_sigmas
A required, single-line text field for an optional custom schedule, with an empty string as its default. Leave it blank to use the selected preset, or enter comma-separated numbers such as 0.8, 0.5, 0.2, 0.0 to replace the preset entirely. Each value must be between the minimum of 0.0 and the maximum of 1.0, inclusive, and the sequence must be non-increasing, meaning each value must be equal to or lower than the preceding value. The ordering check allows a numerical tolerance of 1e-12. The final value must be zero within an absolute tolerance of 1e-12; using an explicit 0.0 is the clearest choice. Spaces around numbers and empty comma-separated entries are ignored, but a nonblank entry must still produce at least one number. No maximum number of entries is specified, and there is no requirement to start at 1.0. The reported step count is the number of values minus one, so the example above produces three steps; a schedule containing only 0.0 is accepted but produces zero steps.
Open-H3 IC · Sigma Grid Output Parameters:
sigmas
A SIGMAS output containing the selected or custom schedule as a one-dimensional PyTorch tensor with float32 precision. Each entry represents a noise level, and adjacent entries define the transitions that a compatible downstream sampling node can follow. For the default up3 preset, the tested tensor contains approximately 0.834942, 0.696852, 0.468753, 0.0. Connect this output to a compatible SIGMAS input in your sampling workflow. This output carries only the schedule: it does not contain a model, latent data, or generated media. Small differences between the displayed decimal values and the original schedule values can occur because the tensor stores numbers at 32-bit floating-point precision.
report
A STRING output containing a JSON-formatted summary of the schedule. It includes three fields: sampler, which identifies the selected preset or reads "custom" when you supplied a custom schedule; sigmas, which lists the schedule values before their conversion into a float32 tensor; and steps, which is calculated as the number of schedule entries minus one, with a minimum of zero. For example, the custom sequence 0.8, 0.5, 0.2, 0.0 produces a report identifying a custom schedule with four values and three steps. This summary helps you verify which schedule was actually used, record settings for reproducibility, and detect cases where a custom entry has overridden the preset selection. Because the report uses the values before tensor conversion, its decimal representation can differ slightly from the tensor output.
Open-H3 IC · Sigma Grid Usage Tips:
- To match the supplied Open-H3 IC regeneration workflow, select
up3and leavecustom_sigmasempty. This uses the documented three-step schedule of approximately0.834942, 0.696852, 0.468753, 0.0. - For a custom experiment, start with an easy-to-check sequence such as
0.8, 0.5, 0.2, 0.0. Keep every value within[0, 1], arrange the values from higher to lower, and finish with an explicit0.0. - Clear
custom_sigmaswhenever you want to return to a preset. Changingsampleralone has no effect while a nonblank custom schedule is present. - Inspect the
reportwhen checking your workflow. Itssamplerfield confirms whether a preset or custom schedule was used, and itsstepsfield confirms the number of transitions. - Remember that four sigma values define three steps. Adding entries increases the number of scheduled transitions, but the provided code does not establish that a longer custom schedule improves quality or is appropriate for every model.
- Use the preset as your baseline before changing the schedule. Validation checks numerical constraints, not whether a custom sequence is suitable for your model or regeneration task.
- Write custom values as plain comma-separated numbers without brackets or labels. Use
0.8, 0.4, 0.0rather than[0.8, 0.4, 0.0]orsigmas: 0.8, 0.4, 0.0.
Open-H3 IC · Sigma Grid Common Errors and Solutions:
custom_sigmas must end in 0.0
- Explanation: Your nonblank custom schedule either contains no usable numbers or ends with a value whose absolute distance from zero exceeds
1e-12. For example,0.8, 0.4fails because its final value is not zero, and an entry containing only commas also fails. - Solution: Enter a complete sequence ending in an explicit
0.0, such as0.8, 0.4, 0.0. If you intended to use a preset, clear the field completely instead.
custom_sigmas must be non-increasing
- Explanation: A later value is larger than the value immediately before it by more than the allowed
1e-12tolerance. For example,0.8, 0.4, 0.6, 0.0fails because the schedule rises from0.4to0.6. Equal adjacent values are permitted. - Solution: Arrange the sequence so it never rises, such as
0.8, 0.6, 0.4, 0.0. Check the entire list rather than only its first and last values.
custom_sigmas values must be in [0, 1]
- Explanation: At least one schedule value is below
0.0or above1.0. These range limits are inclusive and are checked without the tolerance used for the ordering and final-zero checks. - Solution: Replace out-of-range entries with suitable numbers between
0.0and1.0. For example, change1.2, 0.5, 0.0to a valid sequence such as1.0, 0.5, 0.0, or clear the custom field to use a preset.
could not convert string to float: '...'
- Explanation: A comma-separated entry cannot be read as a number. This can happen if you include square brackets, explanatory text, or unsupported separators; for example,
[0.8, 0.4, 0.0]includes brackets that the parser does not remove. The quoted text in the actual error identifies the entry that failed. - Solution: Use only numbers separated by commas, such as
0.8, 0.4, 0.0. Remove brackets, labels, and comments, and use a decimal point rather than a decimal comma.
Unknown IC sampler: ...
- Explanation: When no custom schedule is supplied, the requested preset name is not found in the preset collection. The normal interface restricts selection to
up3,up4, andhf8, so this error is most relevant to programmatic calls or altered workflow data. - Solution: Select one of the available preset names using the node’s dropdown. If you are calling the node programmatically, use the exact supported spelling, such as
up3, or supply a valid custom schedule.
