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 FreSca node applies frequency-dependent scaling to guidance during the sampling process. It separates the guidance signal into low-frequency and high-frequency components using Fourier filtering, then applies different scaling factors to each frequency range before recombining them. This allows for more nuanced control over how guidance affects different aspects of the generated output.

Inputs

ParameterData TypeRequiredRangeDescription
modelMODELYes-The model to apply frequency scaling to
scale_lowFLOATNo0-10Scaling factor for low-frequency components (default: 1.0)
scale_highFLOATNo0-10Scaling factor for high-frequency components (default: 1.25)
freq_cutoffINTNo1-10000Number of frequency indices around center to consider as low-frequency (default: 20)

Outputs

Output NameData TypeDescription
modelMODELThe modified model with frequency-dependent scaling applied to its guidance function