CFGNorm - ComfyUI Built-in Node Documentation

CFGNorm applies a normalization technique to the classifier-free guidance (CFG) process in diffusion models. It adjusts the scale of the denoised prediction by comparing the norms of the conditional and unconditional outputs, then applies a strength multiplier to control the effect. By default the normalization only attenuates the guidance output, but enabling pre_cfg rescales the combined noise before the sampler’s CFG combine without clamping, which can amplify.

Inputs

Parameter	Description	Data Type	Required	Range
`model`	The diffusion model to apply CFG normalization to	MODEL	Yes	-
`strength`	Controls the intensity of the normalization effect applied to the CFG scaling (default: 1.0)	FLOAT	Yes	0.0 to 100.0
`pre_cfg`	If true, rescale the combined noise BEFORE the sampler’s CFG combine, without clamping (can amplify). Matches the norm-scaled CFG used by models like Lens. Default false keeps the original post-CFG x0-space attenuate-only behavior. (default: False)	BOOLEAN	No	true / false

Note: This node is marked as experimental.

Outputs

Output Name	Description	Data Type
`patched_model`	Returns the modified model with CFG normalization applied to its sampling process	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): 427dda498ebda2642489aa2591f063355a77b1e064e0dc7fa8570ed8200c2734

​Inputs

​Outputs

Inputs

Outputs