RenormCFG - ComfyUI Built-in Node Documentation

The RenormCFG node modifies the classifier-free guidance (CFG) process in diffusion models by applying conditional scaling and normalization.

The RenormCFG node modifies the classifier-free guidance (CFG) process in diffusion models by applying conditional scaling and normalization. It adjusts the denoising process based on a specified timestep threshold and a renormalization factor, controlling the influence of conditional versus unconditional predictions during image generation. The resulting model is returned with this patched CFG behavior.

Inputs

Parameter	Description	Data Type	Required	Range
model	The diffusion model to apply renormalized CFG to	MODEL	Yes	-
cfg_trunc	Timestep threshold for applying CFG scaling. When the current timestep is below this value, CFG scaling and renormalization are applied; otherwise, only the conditional prediction is used (default: 100.0)	FLOAT	No	0.0 - 100.0 (step 0.01)
renorm_cfg	Renormalization factor that limits the maximum norm of the CFG-scaled prediction relative to the original conditional prediction. A value of 0.0 disables renormalization (default: 1.0)	FLOAT	No	0.0 - 100.0 (step 0.01)

Note: cfg_trunc and renorm_cfg are advanced parameters. Renormalization only takes effect when renorm_cfg is greater than 0.0 and the current timestep is below cfg_trunc; if the new prediction norm is already below the computed maximum, no rescaling is performed.

Outputs

Output Name	Description	Data Type
model	The modified model with the renormalized CFG function applied	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): 5925bdfe2d62ef7261d73cda661834102ae6600b1afe53f4093568a6e83ec2ab