KSampler (Advanced)

Documentation

• Class name: KSamplerAdvanced
• Category: sampling
• Output node: False

The KSamplerAdvanced node is designed to enhance the sampling process by providing advanced configurations and techniques. It aims to offer more sophisticated options for generating samples from a model, improving upon the basic KSampler functionalities.

Input types

Parameter	Comfy dtype	Description
model	MODEL	Specifies the model from which samples are to be generated, playing a crucial role in the sampling process.
add_noise	COMBO[STRING]	Determines whether noise should be added to the sampling process, affecting the diversity and quality of the generated samples.
noise_seed	INT	Sets the seed for noise generation, ensuring reproducibility in the sampling process.
steps	INT	Defines the number of steps to be taken in the sampling process, impacting the detail and quality of the output.
cfg	FLOAT	Controls the conditioning factor, influencing the direction and space of the sampling process.
sampler_name	COMBO[STRING]	Selects the specific sampler to be used, allowing for customization of the sampling technique.
scheduler	COMBO[STRING]	Chooses the scheduler for controlling the sampling process, affecting the progression and quality of samples.
positive	CONDITIONING	Specifies the positive conditioning to guide the sampling towards desired attributes.
negative	CONDITIONING	Specifies the negative conditioning to steer the sampling away from certain attributes.
latent_image	LATENT	Provides the initial latent image to be used in the sampling process, serving as a starting point.
start_at_step	INT	Determines the starting step of the sampling process, allowing for control over the sampling progression.
end_at_step	INT	Sets the ending step of the sampling process, defining the scope of the sampling.
return_with_leftover_noise	COMBO[STRING]	Indicates whether to return the sample with leftover noise, affecting the final output's appearance.

Output types

Parameter	Comfy dtype	Description
latent	LATENT	The output represents the latent image generated from the model, reflecting the applied configurations and techniques.