FreeU - ComfyUI Built-in Node Documentation - ComfyUI

The FreeU node applies frequency-domain modifications to a model’s output blocks to enhance image generation quality. It works by scaling different channel groups and applying Fourier filtering to specific feature maps, allowing for fine-tuned control over the model’s behavior during the generation process.

Inputs

Parameter	Description	Data Type	Required	Range
`model`	The model to apply FreeU modifications to	MODEL	Yes	-
`b1`	Backbone scaling factor for model_channels × 4 features (default: 1.1)	FLOAT	Yes	0.0 - 10.0
`b2`	Backbone scaling factor for model_channels × 2 features (default: 1.2)	FLOAT	Yes	0.0 - 10.0
`s1`	Skip connection scaling factor for model_channels × 4 features (default: 0.9)	FLOAT	Yes	0.0 - 10.0
`s2`	Skip connection scaling factor for model_channels × 2 features (default: 0.2)	FLOAT	Yes	0.0 - 10.0

Note: FreeU adjustments are applied only to feature maps whose channel count equals model_channels × 4 (using b1 and s1) or model_channels × 2 (using b2 and s2). The Fourier filter scales only the central low-frequency region of the skip connection feature maps; all other frequency components remain unchanged.

Outputs

Output Name	Description	Data Type
`model`	The modified model with FreeU patches 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): 7f7bd34964218ed16c9e58caa446d0c1e69f116607334df4a114cdc4adaf047f