Krea 2 Sigma Scheduler:
Sigmax.Krea2SigmaScheduler builds validated Krea 2 noise schedules for ComfyUI custom sampling. A noise schedule determines how sampling progresses from higher to lower noise; this node provides that progression explicitly, without patching the model’s sampling configuration. It supports Krea 2 Turbo and RAW schedules, plus experimental RAW-to-Turbo LoRA schedules. You can generate a complete schedule or select a portion of one, making the node useful when you need a schedule matched to a particular Krea 2 workflow or sampling stage.
Krea 2 Sigma Scheduler Input Parameters:
variant
Selects the schedule family. The available options are Turbo, RAW, LoRA Experimental (RAW mu), and LoRA Experimental (Turbo mu); Turbo is the first option in the node’s selection list. Turbo uses a fixed shift value, while RAW derives its shift from the image dimensions. The two experimental LoRA options use RAW-derived or Turbo-derived shift behavior, respectively. Choose the option that matches the workflow you are running; the LoRA options are explicitly experimental rather than official RAW or Turbo recipes.
steps
Sets the number of sampling steps in the complete schedule before any step-range selection. The default is 8; the minimum is 1, and the maximum is the node’s configured _MAX_STEPS value, whose numeric value is not included in the provided source. With strict_official enabled, Turbo requires exactly 8 steps and RAW requires exactly 52. With strict mode disabled, Turbo can use other permitted step counts, but RAW still accepts only its named 28-step and 52-step recipes. The experimental LoRA variants use the requested step count within the node’s allowed range.
width
Sets the image width used to construct the schedule. The default is 1024, the minimum is 16, and the interface changes it in increments of 16; its maximum is the configured _MAX_DIMENSION value, which is not numerically specified in the provided source. Enter the width of the image you intend to sample. Width contributes to the resolution-based shift used by RAW and the experimental RAW-mu option, so changing it can change their schedules. The node also records requested and effective dimensions in its schedule information.
height
Sets the image height used to construct the schedule. The default is 1024, the minimum is 16, and the interface changes it in increments of 16; its maximum is the configured _MAX_DIMENSION value, which is not numerically specified in the provided source. Enter the height of the image you intend to sample. Like width, it contributes to the resolution-based shift for RAW and the experimental RAW-mu option. The schedule information records both requested and effective dimensions so you can inspect what was used.
strict_official
Controls whether the supported official step-count restriction is enforced. The default is True; the options are True and False. When enabled, Turbo accepts only 8 steps and RAW accepts only 52 steps. Turning it off permits other valid Turbo step counts and allows RAW’s named 28-step recipe as well as its 52-step recipe; it does not allow arbitrary RAW step counts. For either experimental LoRA variant, the builder sets effective strict mode to False regardless of this setting.
start_step
Sets where the selected portion of the complete schedule begins. The default and minimum are 0; the interface maximum is `_MAX_STEPS
- 1
, whose numeric value is not supplied. Leave it at0` to begin at the first step, or increase it when your workflow needs a later portion. The selected range must fit within the steps available in the complete schedule.
end_step
Sets the end of the selected portion of the complete schedule. The default and minimum are -1; the interface maximum is _MAX_STEPS, whose numeric value is not supplied. The builder treats -1 as the unspecified-end setting and reports the complete schedule’s final step as the effective end. Set a specific end step when you need a shorter portion, ensuring it forms a valid range with start_step and does not exceed the available steps.
Krea 2 Sigma Scheduler Output Parameters:
sigmas
A SIGMAS tensor containing the selected noise schedule for a compatible ComfyUI custom-sampling workflow. Each value represents a noise level in the progression used by sampling. If you select only part of the complete schedule, this output contains that portion rather than the entire schedule. The node constructs and validates the schedule before converting it to a Torch float tensor for the host.
schedule_info
A STRING containing structured JSON information about the schedule. It identifies the profile, recipe, and variant; records requested and effective dimensions; describes the shift used; reports the selected step range and output step count; and includes warnings supplied by the schedule builder. Its fingerprints identify both the complete schedule and the values actually returned through the host tensor. This is useful when you need to inspect a configuration or compare the schedules produced by two runs.
Krea 2 Sigma Scheduler Usage Tips:
- For the strict Turbo configuration, select
Turbo, keepstrict_officialenabled, and setstepsto8. - For strict RAW, select
RAWand setstepsto52. To use the named28-step RAW recipe, turnstrict_officialoff; turning it off does not make other RAW step counts valid. - Set
widthandheightto the image dimensions you intend to use, particularly for RAW orLoRA Experimental (RAW mu), where the shift is derived from resolution. - Keep
start_stepat0andend_stepat-1when you need the complete schedule. Change them only when the rest of your sampling workflow is intended to consume a selected portion. - Inspect
schedule_infowhen comparing runs. Its recipe, effective dimensions, shift, slicing details, warnings, and fingerprints provide a more precise record than the visible settings alone.
Krea 2 Sigma Scheduler Common Errors and Solutions:
strict official Turbo requires exactly 8 steps
- Explanation:
Turbowas selected withstrict_officialenabled, butstepsis not8. - Solution: Set
stepsto8, or disablestrict_officialif you deliberately need a modified Turbo step count.
strict official RAW requires exactly 52 steps
- Explanation:
RAWwas selected withstrict_officialenabled, butstepsis not52. - Solution: Set
stepsto52. For the named28-step RAW recipe, disablestrict_officialand setstepsto28.
RAW steps must select the named 28-step or 52-step recipe
- Explanation: RAW supports only its named
28-step and52-step recipes, even when strict mode is off. - Solution: Choose
28or52steps. If you need a different step count, select a variant that supports it rather than changing RAW’s recipe count.
strict_official must be boolean
- Explanation: The schedule builder received a value other than
TrueorFalseforstrict_official. - Solution: Use the node’s Boolean control. If another node supplies this setting, make sure it passes a Boolean value.
ComfyUI host execution requires Torch FloatTensor support
- Explanation: The node built the schedule but could not access Torch’s
FloatTensorsupport when converting it into the host output. - Solution: Check that the ComfyUI environment running the node has a working Torch installation, then restart ComfyUI after correcting the environment.
ComfyUI host tensor must expose numeric tolist output
- Explanation: The tensor created by the host did not provide numeric values through
tolist(), which the node needs to bind the schedule information to its actual output. - Solution: Check the ComfyUI and Torch environment for an incompatible or modified tensor implementation, then retry with a standard working host setup.
Invalid step range or input value
- Explanation: The builder validates step counts, dimensions, and the selected step range. An out-of-range value or a range that does not fit the complete schedule can prevent it from producing an output. The precise message depends on the validation path.
- Solution: Keep
steps,width, andheightwithin their interface limits. Begin withstart_stepset to0andend_stepset to-1, then adjust the range within the available steps shown inschedule_info.
