LatentApplyOperationCFG - ComfyUI Built-in Node Documentation

The LatentApplyOperationCFG node applies a latent operation inside the classifier-free guidance (CFG) step of a model’s sampling process. It intercepts the conditioning outputs produced before CFG, applies the connected operation to the latent values, and returns the model with this modified sampling behavior.

Inputs

Parameter	Description	Data Type	Required	Range
`model`	The model to which the CFG operation will be applied	MODEL	Yes	-
`operation`	The latent operation to apply during the CFG sampling process	LATENT_OPERATION	Yes	-
`start_percent`	Fraction of the denoising schedule at which the operation starts being applied; 0 is the beginning of the schedule (default: 0.0)	FLOAT	No	0.0 to 1.0 (step 0.001)
`end_percent`	Fraction of the denoising schedule at which the operation stops being applied; 1 is the end of the schedule (default: 1.0)	FLOAT	No	0.0 to 1.0 (step 0.001)

Note: This node is marked as experimental. The operation is applied to the model’s conditioning outputs during the CFG sampling process. When two conditioning outputs are present, the operation is applied to the difference between the first and second output, and the second output is added back to the result. When only one conditioning output is present, the operation is applied directly to it. The operation only runs between the start_percent and end_percent points of the denoising schedule, so it can be limited to part of the schedule; outside that window the conditioning outputs are returned unchanged.

Outputs

Output Name	Description	Data Type
`model`	The modified model with the CFG operation 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): 6a5f59f02eaec38334c63d871e48e89aa983a5ac2ca10801161cdc9e13cacdf2

​Inputs

​Outputs

Inputs

Outputs