HyperFlow Sigmas:
HyperFlowSigmas provides the sampling schedule designed for HyperFlow checkpoints, helping you use the noise-level progression intended for their short sampling process rather than relying on a general-purpose schedule. A sampling schedule acts like a roadmap for generation: it tells the sampler which noise levels to move through as it develops the final result. The node supports three predefined schedules, with 4 steps favoring speed, 6 balancing speed and quality, and 8 favoring quality. When constructing a schedule, it applies HyperFlow’s fixed video-flow shift of 12 to the predefined noise levels, preserving the checkpoint-specific progression instead of spacing the levels evenly. It can also reuse a schedule already attached to the loaded model for the default configuration. Its main purpose is to make HyperFlow’s intended low-step sampling setup convenient and consistent; it does not perform image or video generation by itself.
HyperFlow Sigmas Input Parameters:
model
A required input of type MODEL that provides access to the loaded diffusion model and any HyperFlow schedule attached to it. For the default 6-step configuration, the node checks whether the diffusion model has a stored schedule named hyperflow_sigmas and uses it when available. If that schedule is absent, or you choose 4 or 8 steps, the node constructs the corresponding predefined HyperFlow schedule instead. You should connect the HyperFlow model you intend to sample with, typically from the HyperFlowLoader or HyperFlowGGUFLoader node, and use that same model in your sampler. Although a schedule can be constructed when the stored attribute is absent, this does not establish that an unrelated checkpoint is compatible with HyperFlow sampling. This input has no numeric range, selectable options, or default value.
steps
An optional INT input that selects the predefined sampling schedule. The default is 6, the minimum is 4, the maximum is 8, and the interface increment is 2; the only supported values are 4, 6, and 8. Choose 4 when you prioritize faster generation, 6 for the documented balance between speed and quality, or 8 when you prioritize quality and can allow more sampling work. These choices select different predefined progressions, not simply a longer or shorter version of an evenly spaced schedule. At 6 steps, an existing model-attached schedule takes priority when available; at 4 or 8 steps, the node always constructs the selected schedule from its predefined values. If you omit this input, the node uses 6. Unsupported values are rejected rather than rounded to a supported step count.
HyperFlow Sigmas Output Parameters:
sigmas
An output of type SIGMAS containing the sequence of noise levels that guides the sampler through generation. When the node constructs a predefined schedule, it returns 5 values for 4 steps, 7 values for 6 steps, or 9 values for 8 steps: each sampling step moves between two neighboring values, so the sequence needs one more value than the step count. These constructed schedules descend from 1.0 to 0.0, with their intermediate values transformed using 12 × raw / (1 + 11 × raw). The final zero marks the end of the noise-level progression. Constructed schedules use a float32 tensor; when an existing model-attached schedule is selected, its stored values and data type are preserved. In either case, the node returns an independent copy on the CPU, avoiding direct changes to the original schedule. Connect this output to the sigmas input of SamplerCustom to apply the selected HyperFlow progression.
HyperFlow Sigmas Usage Tips:
-
Start with 6 steps, the documented default, to establish a baseline before comparing faster or higher-step configurations.
-
Use 4 steps for quick previews or rapid prompt comparisons, and try 8 steps when quality is your priority. The node describes these as speed-versus-quality preferences, not guarantees that every result will improve with more steps.
-
Follow the recommended sampling setup: connect the schedule to SamplerCustom, select
eulerthrough your sampler-selection node, and set SamplerCustom’scfgto 1. - Connect the same HyperFlow model to this node and to SamplerCustom. This keeps schedule selection aligned with the model you are actually sampling. -
Keep the supplied schedule unchanged while establishing a working workflow. Nodes that flip the schedule, change its first value, or insert intermediate values alter the predefined progression and no longer reproduce the original HyperFlow schedule.
-
If you control
stepsthrough another node or an automated workflow, restrict the supplied integer to 4, 6, or 8. Values such as 5 and 7 are not supported even though they fall between the interface’s minimum and maximum. -
When comparing step counts, keep your prompt, seed, model, and other sampling settings unchanged so you can better judge the effect of the schedule choice.
HyperFlow Sigmas Common Errors and Solutions:
HyperFlow samples [4, 6, 8] steps, got 5
- Explanation: The requested step count is not one of the three supported schedules. The number after
gotreflects the unsupported value supplied to the node; 5 is an example. - Solution: Set
stepsto 4, 6, or 8. If another node supplies this value, correct that source rather than only changing the visible widget.
Required input is missing: model
- Explanation: ComfyUI cannot execute the node without its required MODEL connection. The exact validation message may vary by ComfyUI version.
- Solution: Connect the MODEL output from your HyperFlow loader to
modeland confirm that the loader completes successfully.
ValueError: invalid literal for int() with base 10: 'six'
- Explanation: The node converts
stepsto an integer before selecting a schedule. A text value such assixcannot be converted. This is primarily relevant to scripted calls or malformed workflows, because the normal interface uses an integer control. - Solution: Supply an integer value of 4, 6, or 8. Check any script or upstream control that sets the step count.
AttributeError: 'NoneType' object has no attribute 'model'
- Explanation: The node received no usable model object when it tried to access the diffusion model. Other malformed objects can produce a similar AttributeError mentioning
modelordiffusion_model. A missinghyperflow_sigmasattribute alone is not an error: the node handles that case by constructing a schedule. - Solution: Check the loader for earlier errors, reconnect its MODEL output, and confirm that you are supplying a properly loaded ComfyUI model rather than a different data type.
AttributeError: 'list' object has no attribute 'clone'
- Explanation: In the default 6-step configuration, a model-attached
hyperflow_sigmasvalue was found, but it was not a tensor supporting the copying operation the node expects. A Python list is one possible example; the exact type named in the error can differ. - Solution: Reload the model through a compatible HyperFlow loader and check whether another custom node or script replaced its stored schedule. If needed, update the HyperFlow node pack and reload the model. Selecting 4 or 8 steps bypasses the stored default schedule, but does not repair the underlying model configuration.
