MiniMax H3 LoRA Schedule:
H3LoraSchedule creates a time-based strength schedule for selected rows in a Power LoRA Stack. Instead of applying the same LoRA strength throughout the entire sampling process, you can make the strength increase, decrease, or change according to a chosen curve over the model’s sampling timeline. The schedule can be based on sampler step positions or on the scheduler’s normalized sigma values, allowing you to control when a LoRA has the most influence. You can target every row or only specific rows, combine several schedule nodes in a chain, and use different schedules for different groups of LoRAs. When schedules overlap, the later schedule in the chain takes priority for the affected rows. The node reads the sampler timeline automatically, so it does not require a separate SIGMAS connection.
MiniMax H3 LoRA Schedule Input Parameters:
rows
This text value selects which Power LoRA Stack rows receive the schedule. Use all to affect every row, a single row number such as 2, a comma-separated selection such as 1,3, or a range such as 2-4. The selected rows use the strength pattern created by this node, while unselected rows remain controlled by other schedules or their normal stack settings. The default is all. The source does not define a minimum or maximum numeric value because this is a row-selection string; row numbers should correspond to rows that exist in the connected Power LoRA Stack.
start_strength
This floating-point value sets the LoRA strength at the beginning of the selected scheduling interval. A positive value applies the LoRA in its normal direction, zero disables its influence at that point, and a negative value reverses the strength effect when supported by the connected LoRA stack. The value is interpolated toward end_strength according to the selected curve, unless curve is step or explicit. The minimum is -10.0, the maximum is 10.0, and the default is 1.0.
end_strength
This floating-point value sets the LoRA strength at the end of the selected scheduling interval. It determines the target strength reached at end_percent, or the final value of the generated pattern when the selected curve uses a continuous transition. For example, a start strength of 1.0 and an end strength of 0.0 gradually fades the selected LoRA out over the chosen interval when using a suitable curve. The minimum is -10.0, the maximum is 10.0, and the default is 0.1.
curve
This option determines how the strength changes between the start and end of the scheduling interval. linear changes strength at a constant rate; cosine creates a smoother eased transition; smoothstep eases the beginning and ending of the transition; power applies a power-shaped transition controlled by curve_power; step holds one value and changes to the other rather than gradually interpolating; and explicit uses the values entered in explicit_strengths instead of generating values from the start and end strengths. The available options are linear, cosine, smoothstep, power, step, and explicit. There is no separate numeric minimum, maximum, or default shown for this option; the source lists the available choices in the order shown, with linear appearing first.
curve_power
This floating-point value controls the shape of a power curve. Values near 1.0 produce a shape closer to a direct progression, while values further from 1.0 make the transition spend more of its interval near one end before changing more rapidly toward the other. This setting affects the distribution of intermediate LoRA strengths, not the start and end strength values themselves. It is also accepted as part of the schedule configuration for other curve choices, but it has its primary visible effect with power. The minimum is 0.01, the maximum is 20.0, the default is 2.0, and the control uses increments of 0.05.
explicit_strengths
This text field contains manually specified strength values for an explicit curve. Enter strengths separated by commas or spaces, with one value for each intended sampling step or schedule position, for example 1.0, 0.8, 0.5, 0.2, 0.0. The node records these values instead of calculating a transition from start_strength and end_strength; the number of values is reported in the info output. The default is an empty string. The source does not define a minimum or maximum text length, but this field should contain valid numeric strength values, and the number of values should match the intended schedule positions.
start_percent
This floating-point value specifies where the schedule begins within the sampler timeline, expressed as a percentage. A value of 0.0 starts the schedule at the beginning of the selected domain, while a later value delays the schedule until that percentage of the timeline has been reached. The minimum is 0.0, the maximum is 100.0, the default is 0.0, and the control uses increments of 1.0.
end_percent
This floating-point value specifies where the schedule ends within the sampler timeline, expressed as a percentage. The selected curve is applied between start_percent and end_percent; outside that interval, the schedule follows the behavior associated with the generated schedule and selected curve. Setting this to a value below 100.0 makes the scheduled interval finish before the end of sampling. The minimum is 0.0, the maximum is 100.0, the default is 100.0, and the control uses increments of 1.0.
domain
This option chooses how the percentage positions are mapped onto the sampling process. steps follows model-call indices, meaning the schedule is positioned according to the sequence of sampling steps. sigma follows the scheduler’s actual normalized noise values, which can produce timing that reflects the noise schedule rather than treating every step interval as equally spaced. The available options are steps and sigma. The source does not show a separate default declaration; steps is the first listed option and is also the method used by the node’s build method when no domain value is supplied.
schedule
This optional H3_SCHEDULE input accepts an earlier schedule so you can chain multiple H3LoraSchedule nodes. The current node appends its schedule to the incoming chain, and when row selectors overlap, the later node wins for those rows. Leave this input unconnected when creating the first schedule. The input accepts an existing schedule chain or a compatible empty value; no explicit default, range, or option list is defined in the node interface.
MiniMax H3 LoRA Schedule Output Parameters:
schedule
This H3_SCHEDULE output contains the newly created schedule, including the selected rows, strength settings, curve, percentage interval, and domain. If an earlier schedule input is connected, the output contains the earlier schedule followed by this node’s schedule, allowing several nodes to be combined before connecting the result to the Power LoRA Stack’s schedule input. The output is the main functional result of the node and is intended to be wired into that schedule input rather than directly into a sampler.
info
This string output provides a short human-readable summary of the schedule. It reports the selected rows, starting and ending strengths, curve type, domain, and percentage interval, such as rows all: 1 → 0.1 linear (steps 0–100%). When curve is explicit, the summary also reports the number of explicit values, such as 10 values. This output is useful for checking the node’s configuration while building or debugging a workflow, but it does not replace the actual schedule output.
MiniMax H3 LoRA Schedule Usage Tips:
- Start with
rowsset toallwhen you want one schedule to control the complete Power LoRA Stack. Use specific row numbers, comma-separated selections, or ranges when different LoRAs need different timing. - Use a positive
start_strengthand a lowerend_strengthto fade a LoRA out during sampling. For example,1.0to0.0with alinearorsmoothstepcurve gradually reduces the LoRA’s influence. - Use a lower starting strength and a higher ending strength when you want the selected LoRA to become more influential later in the sampling process.
- Choose
smoothsteporcosinewhen you want a softer transition with less abrupt change at the beginning and end. Chooselinearwhen predictable, even change is more important. - Use
powerwithcurve_powerto bias when the transition happens. Adjust the power gradually because values far from1.0can concentrate most of the change near one end of the interval. - Use
stepwhen you need a clear switch between two strength states rather than a gradual transition, such as keeping a LoRA active and then turning it down after a specific point. - Use
explicitwhen you need direct control over individual schedule positions. Enter comma- or space-separated numeric strengths inexplicit_strengths, and check theinfooutput to confirm how many values were detected. - Set
start_percentandend_percentto limit the period in which the transition occurs. For example,25to75confines the scheduled change to the middle portion of the selected timeline. - Use
stepswhen you want the schedule tied to the model-call sequence. Usesigmawhen the scheduler’s actual normalized noise progression should determine the schedule timing. - Chain nodes when different rows need different schedules. Connect the previous node’s
scheduleoutput to the next node’sscheduleinput, and place the schedule intended to take priority later in the chain. - Connect the final
scheduleoutput to the Power LoRA Stack’s schedule input. The node does not require a separate SIGMAS connection because the stack reads the sampler timeline automatically.
MiniMax H3 LoRA Schedule Common Errors and Solutions:
ValueError while parsing explicit strengths
- Explanation: The
explicitcurve requiresexplicit_strengthsto contain numeric values separated by commas or spaces. Text, unsupported separators, or empty entries can prevent the schedule from being created. - Solution: Enter values such as
1.0, 0.8, 0.5, 0.2or1.0 0.8 0.5 0.2. Remove labels, brackets, and other non-numeric text. Uselinear,cosine,smoothstep,power, orstepinstead if you do not need manually entered values.
explicit schedule has an unexpected number of values
- Explanation: An explicit schedule depends on the number of strengths provided. Too few or too many values can make the timing different from what you intended for the available sampling positions.
- Solution: Count the values in
explicit_strengthsand provide one value for each intended schedule position. Confirm the detected count in theinfooutput and adjust the list or use a generated curve when exact per-position control is unnecessary.
schedule input is not a valid H3_SCHEDULE
- Explanation: The optional
scheduleinput can only accept a schedule produced by a compatible H3 schedule node. Connecting another data type, such as a model, conditioning, string, or image, prevents the chain from being interpreted correctly. - Solution: Connect an H3LoraSchedule
scheduleoutput, or leave the input disconnected when this is the first schedule node in the chain.
rows does not affect the intended LoRA rows
- Explanation: The node applies its settings according to row selectors, so a row number that does not exist or a selector with an invalid format may not target the intended stack entries.
- Solution: Confirm the row numbering in the Power LoRA Stack and use supported forms such as
all,2,1,3, or2-4. Check the connected stack and theinfooutput to verify the selection you intended.
start_percent or end_percent is outside the allowed range
- Explanation: Timeline percentages must be between
0.0and100.0. Values outside that range cannot represent a valid position in the sampling timeline. - Solution: Set both values between
0and100. Use0for the beginning and100for the end. Make sureend_percentis not earlier thanstart_percentunless the connected schedule implementation explicitly supports a reversed interval.
curve_power is outside the allowed range
- Explanation: The node accepts curve power values from
0.01through20.0. Values outside this range are rejected by the input control or can produce an invalid configuration. - Solution: Choose a value within the allowed range, such as the default
2.0. Use smaller adjustments around1.0when you want a mild change in curve shape.
curve is not a supported option
- Explanation: The curve setting must match one of the choices defined by the node. An unsupported value cannot determine how strengths should be generated.
- Solution: Select one of
linear,cosine,smoothstep,power,step, orexplicit. If entering values manually, selectexplicitand provide numeric values inexplicit_strengths.
schedule output is not connected to the LoRA stack
- Explanation: The node can create a valid schedule without changing the image if its
scheduleoutput is not passed to the Power LoRA Stack’s schedule input. - Solution: Connect the final H3LoraSchedule
scheduleoutput to the Power LoRA Stack schedule input. When using several schedule nodes, connect the final node in the chain.
MiniMax H3 LoRA Schedule cannot be found or loaded
- Explanation: This usually indicates that the custom node package containing H3LoraSchedule is not installed, is disabled, or failed during startup. The provided node source does not define a more specific loading message.
- Solution: Confirm that the package is installed in the correct ComfyUI custom-nodes directory, restart ComfyUI, and review the startup console for the first import or dependency error. Ensure the package’s required dependencies are available before rebuilding the workflow.
