APG - ComfyUI Built-in Node Documentation

The APG (Adaptive Projected Guidance) node modifies the sampling process by adjusting how guidance is applied during diffusion. It separates the guidance vector into parallel and orthogonal components relative to the conditional output, allowing for more controlled image generation. The node provides parameters to scale the guidance, normalize its magnitude, and apply momentum for smoother transitions between diffusion steps.

Inputs

Parameter	Description	Data Type	Required	Range
`model`	The diffusion model to apply adaptive projected guidance to	MODEL	Yes	-
`eta`	Controls the scale of the parallel guidance vector. Default CFG behavior at a setting of 1 (default: 1.0).	FLOAT	Yes	-10.0 to 10.0
`norm_threshold`	Normalize guidance vector to this value, normalization disabled at a setting of 0 (default: 5.0).	FLOAT	Yes	0.0 to 50.0
`momentum`	Controls a running average of guidance during diffusion, disabled at a setting of 0 (default: 0.0).	FLOAT	Yes	-5.0 to 1.0

Note: When the noise level (sigma) increases during sampling, the momentum running average resets to zero. If the model provides only a single conditioning output (no separate unconditional conditioning), the guidance adjustment is skipped and the conditioning is left unchanged.

Outputs

Output Name	Description	Data Type
`model`	Returns the modified model with adaptive projected guidance 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): df0c76aee28479d49c4e471e54d1d32082adc6921a6a50b506675144a79e018a

​Inputs

​Outputs

Inputs

Outputs