Automatic CFG - Post rescale only

Documentation

• Class name: Automatic CFG - Post rescale only
• Category: model_patches/automatic_cfg/presets
• Output node: False

This node is designed to apply a post-processing rescaling operation to the output of a generative model, specifically targeting the adjustment of the model's output based on a rescaling factor. It aims to refine the model's predictions by adjusting the scale of the output, enhancing the balance between conditioned and unconditioned components of the generation.

Input types

Required

• model
  • The generative model to which the post-rescaling operation will be applied. This parameter is crucial as it determines the base model whose output will be adjusted.
  • Comfy dtype: MODEL
  • Python dtype: torch.nn.Module
• subtract_latent_mean
  • A boolean flag indicating whether the latent mean should be subtracted from the model's output, affecting the final generation's characteristics.
  • Comfy dtype: BOOLEAN
  • Python dtype: bool
• subtract_latent_mean_sigma_start
  • Defines the starting sigma value for subtracting the latent mean, influencing when this operation is applied during the generation process.
  • Comfy dtype: FLOAT
  • Python dtype: float
• subtract_latent_mean_sigma_end
  • Defines the ending sigma value for subtracting the latent mean, marking the end of the range within which this operation is applied.
  • Comfy dtype: FLOAT
  • Python dtype: float
• latent_intensity_rescale
  • A boolean flag indicating whether the intensity of the latent space should be rescaled, impacting the visual quality and characteristics of the generated output.
  • Comfy dtype: BOOLEAN
  • Python dtype: bool
• latent_intensity_rescale_method
  • Specifies the method used for rescaling the intensity of the latent space, affecting how the rescaling operation is performed.
  • Comfy dtype: COMBO[STRING]
  • Python dtype: str
• latent_intensity_rescale_cfg
  • The configuration value for the latent intensity rescale operation, determining the intensity of rescaling applied.
  • Comfy dtype: FLOAT
  • Python dtype: float
• latent_intensity_rescale_sigma_start
  • Defines the starting sigma value for the latent intensity rescale operation, influencing when this adjustment is applied in the generation process.
  • Comfy dtype: FLOAT
  • Python dtype: float
• latent_intensity_rescale_sigma_end
  • Defines the ending sigma value for the latent intensity rescale operation, marking the end of the range within which this adjustment is applied.
  • Comfy dtype: FLOAT
  • Python dtype: float

Output types

• model
  • Comfy dtype: MODEL
  • The modified generative model with the post-rescaling operation applied. This output reflects the adjustments made to the model's output scaling, aimed at enhancing generation quality.
  • Python dtype: torch.nn.Module

Usage tips

• Infra type: GPU
• Common nodes: unknown

Source code

class postCFGrescaleOnly:
    @classmethod
    def INPUT_TYPES(s):
        return {"required": {
                                "model": ("MODEL",),
                                "subtract_latent_mean" : ("BOOLEAN", {"default": True}),
                                "subtract_latent_mean_sigma_start": ("FLOAT", {"default": 15,  "min": 0.0, "max": 10000.0, "step": 0.1, "round": 0.1}),
                                "subtract_latent_mean_sigma_end":   ("FLOAT", {"default": 7.5, "min": 0.0, "max": 10000.0, "step": 0.1, "round": 0.1}),
                                "latent_intensity_rescale"     : ("BOOLEAN", {"default": True}),
                                "latent_intensity_rescale_method" : (["soft","hard","range"], {"default": "hard"},),
                                "latent_intensity_rescale_cfg" : ("FLOAT", {"default": 7.6,  "min": 0.0, "max": 100.0, "step": 0.1, "round": 0.1}),
                                "latent_intensity_rescale_sigma_start": ("FLOAT", {"default": 15,  "min": 0.0, "max": 10000.0, "step": 0.1, "round": 0.1}),
                                "latent_intensity_rescale_sigma_end":   ("FLOAT", {"default": 7.5, "min": 0.0, "max": 10000.0, "step": 0.1, "round": 0.1}),
                              }}
    RETURN_TYPES = ("MODEL",)
    FUNCTION = "patch"

    CATEGORY = "model_patches/automatic_cfg/presets"

    def patch(self, model,
              subtract_latent_mean, subtract_latent_mean_sigma_start, subtract_latent_mean_sigma_end,
              latent_intensity_rescale, latent_intensity_rescale_method, latent_intensity_rescale_cfg, latent_intensity_rescale_sigma_start, latent_intensity_rescale_sigma_end
              ):
        advcfg = advancedDynamicCFG()
        m = advcfg.patch(model=model, 
                         subtract_latent_mean = subtract_latent_mean,
                         subtract_latent_mean_sigma_start = subtract_latent_mean_sigma_start, subtract_latent_mean_sigma_end = subtract_latent_mean_sigma_end,
                         latent_intensity_rescale = latent_intensity_rescale, latent_intensity_rescale_cfg = latent_intensity_rescale_cfg, latent_intensity_rescale_method = latent_intensity_rescale_method,
                         latent_intensity_rescale_sigma_start = latent_intensity_rescale_sigma_start, latent_intensity_rescale_sigma_end = latent_intensity_rescale_sigma_end,
                         ignore_pre_cfg_func = True
                         )[0]
        return (m, )