EasyUse/Samplergenerated

EasyKsampler (Downscale Unet)(samplerSimpleDownscaleUnet)

EasyKsampler (Downscale Unet) node from ComfyUI-Easy-Use.

sampler Simple Downscale Unet

pipe
model
pipe
image
downscale_mode
Auto
block_number
3
downscale_factor
2.000
start_percent
0.000
end_percent
0.350
downscale_after_skip
downscale_method
COMBO
upscale_method
COMBO
image_output
Preview
link_id
INT
save_prefix
ComfyUI
Easy Use

Description

The EasyKsampler (Downscale Unet) node is a pipeline-based sampler that applies U-Net downscaling during part of the denoising process. It is designed to reduce memory usage and accelerate sampling by processing the U-Net at a reduced resolution for the selected step range, then restoring the normal U-Net resolution.

It accepts an EasyUse PIPE_LINE, optionally takes a separate model, and returns the updated pipeline and a decoded image.

Inputs

NameTypeDefaultDescription
pipePIPE_LINE—The EasyUse pipeline containing the model, positive/negative conditioning, latent, and sampler settings.
downscale_modeCOMBOAutoNone disables U-Net downscaling. Auto lets the node decide suitable downscale settings. Custom uses the manually entered block_number and downscale_factor.
block_numberINT3The U-Net block number targeted by the downscale operation. Range: 1–32.
downscale_factorFLOAT2How strongly the U-Net's working resolution is reduced. A value of 2 typically halves the spatial dimensions; larger values apply a stronger reduction. Range: 0.1–9.
start_percentFLOAT0The point in the sampling process where downscaling begins. 0 is the first step, 1 is the last step.
end_percentFLOAT0.35The point in the sampling process where downscaling stops. Defaults to 0.35, so downscaling is active for the first 35% of steps.
downscale_after_skipBOOLEANtrueWhen enabled, the downscale operation is applied after the U-Net skip connection step.
downscale_methodCOMBO—The interpolation/resize method used to downscale U-Net features. Available options are populated by ComfyUI.
upscale_methodCOMBO—The interpolation/resize method used to restore U-Net features to full resolution. Available options are populated by ComfyUI.
image_outputCOMBOPreviewControls how the generated image is handled: Hide, Preview, Save, Hide&Save, Sender, or Sender&Save.
link_idINT—Used by Sender and Sender&Save modes to identify the destination link for sending the generated image.
save_prefixSTRINGComfyUIThe filename prefix used when saving generated images.
modelMODEL—Optional model override. When omitted, the model from the pipe is used.

Outputs

TypeDescription
PIPE_LINEThe updated EasyUse pipeline containing the sampled result. Can be passed to downstream EasyUse nodes.
IMAGEThe generated image tensor. Previewing, saving, and sending behavior is controlled by image_output.

Usage Notes

  • Set downscale_mode to None to disable U-Net downscaling and use the node as a normal KSampler-style pipeline sampler.
  • In Custom mode, block_number and downscale_factor are exposed for manual tuning.
  • The default start_percent/end_percent window of 0.0–0.35 applies downscaling only during the early part of denoising, leaving later steps at full resolution for finer detail.
  • Choose downscale_method and upscale_method that are compatible with your ComfyUI installation and model type.
  • image_output modes that save images use save_prefix as the base filename.
  • The optional model input is useful for sampling with a model that differs from the one stored in the pipeline, without rebuilding the pipe.

Comments

Sign in with GitHub to join the discussion.

Loading comments…