Skip to main content
The SUPIRApply node applies a SUPIR model patch to a diffusion model. It uses the patch to modify the model’s behavior, allowing it to incorporate guidance from an input image during the sampling process. The node also provides controls for adjusting the strength of this guidance over time and includes an optional feature to help maintain fidelity to the original input.

Inputs

ParameterDescriptionData TypeRequiredRange
modelThe base diffusion model to which the SUPIR patch will be applied.MODELYes-
model_patchThe SUPIR model patch containing the weights and configuration for modifying the model.MODELPATCHYes-
vaeThe VAE (Variational Autoencoder) used for encoding the input image into a latent representation.VAEYes-
imageThe input image used to guide the generation process. Only the first three color channels (RGB) are used.IMAGEYes-
strength_startControl strength at the start of sampling (high sigma). The influence of the image guidance begins at this value. (default: 1.0)FLOATNo0.0 - 10.0
strength_endControl strength at the end of sampling (low sigma). Linearly interpolated from start. The influence of the image guidance ends at this value. (default: 1.0)FLOATNo0.0 - 10.0
restore_cfgPulls denoised output toward the input latent. Higher = stronger fidelity to input. 0 to disable. (default: 4.0)FLOATNo0.0 - 20.0
restore_cfg_s_tminSigma threshold below which restore_cfg is disabled. (default: 0.05)FLOATNo0.0 - 1.0
Note: The image input is processed to extract only the RGB channels. If an image with an alpha channel is provided, the alpha channel is ignored.

Outputs

Output NameDescriptionData Type
modelThe diffusion model with the SUPIR patch applied and any additional post-CFG functions configured.MODEL
This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub

Source fingerprint (SHA-256): 41900f166aa7c0e325da874a3e250c56b94569822292785b1b3165705eb22e0c