SamplerCustomAdvanced - ComfyUI Built-in Node Documentation - ComfyUI

The SamplerCustomAdvanced node performs advanced latent space sampling using custom noise, guidance, and sampling configurations. It processes a latent image through a guided sampling process with customizable noise generation and sigma schedules, producing both the final sampled output and a denoised version when available.

Inputs

Parameter	Description	Data Type	Required	Range
`noise`	The noise generator that provides the initial noise pattern and seed for the sampling process	NOISE	Yes	-
`guider`	The guidance model that directs the sampling process toward desired outputs	GUIDER	Yes	-
`sampler`	The sampling algorithm that defines how the latent space is traversed during generation	SAMPLER	Yes	-
`sigmas`	The sigma schedule that controls the noise levels throughout the sampling steps	SIGMAS	Yes	-
`latent_image`	The initial latent representation that serves as the starting point for sampling. Supports optional `noise_mask` for selective denoising, and optional `downscale_ratio_spacial` and `downscale_ratio_temporal` keys for advanced latent handling	LATENT	Yes	-

Outputs

Output Name	Description	Data Type
`output`	The final sampled latent representation after completing the sampling process. Any `downscale_ratio_spacial` or `downscale_ratio_temporal` keys from the input latent are removed from this output	LATENT
`denoised_output`	A denoised version of the output when the sampling process produces an intermediate clean prediction (x0), otherwise returns the same as the output. When available, this represents the model’s best estimate of the clean latent at each step	LATENT

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): 23cffad0f7cf74dcd494c2828b2116bb4d00a1e55e42ded074b587ac20183290