Skip to main content
This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub
The SkipLayerGuidanceSD3 node enhances guidance towards detailed structure by applying an additional set of classifier-free guidance with skipped layers. This experimental implementation is inspired by Perturbed Attention Guidance and works by selectively bypassing certain layers during the negative conditioning process to improve structural details in the generated output.

Inputs

ParameterData TypeRequiredRangeDescription
modelMODELYes-The model to apply skip layer guidance to
layersSTRINGYes-Comma-separated list of layer indices to skip (default: “7, 8, 9”)
scaleFLOATYes0.0 - 10.0The strength of the skip layer guidance effect (default: 3.0)
start_percentFLOATYes0.0 - 1.0The starting point of guidance application as a percentage of total steps (default: 0.01)
end_percentFLOATYes0.0 - 1.0The ending point of guidance application as a percentage of total steps (default: 0.15)

Outputs

Output NameData TypeDescription
modelMODELThe modified model with skip layer guidance applied