EasyCascadeKsampler (Full)(samplerCascadeFull)

Description

The EasyCascadeKsampler (Full) node is the full form of the EasyUse Cascade sampler. It takes an EasyUse PIPE_LINE, runs it through the selected sampler settings, and returns the updated pipeline along with the model and latent that were used in sampling. It is designed for Stable Cascade workflows, where the _c-suffixed inputs refer to the Stage C latent/model side of the cascade pipeline.

In addition to the standard sampler controls, this node exposes VAE selection for encoding images to latents and decoding latents back to images, image preview/save/send options, and optional direct inputs for overriding the model, starting latent, or input image.

Inputs

Required

• pipe (PIPE_LINE): The EasyUse pipeline containing the model, positive/negative conditioning, latent, and related workflow settings. The node uses this as the base state and returns an updated pipe after sampling.

• encode_vae_name (COMBO): The VAE used to encode an input image into a latent when image_to_latent_c is provided.

• decode_vae_name (COMBO): The VAE used to decode the sampled latent into an image for previewing, saving, or sending.

• steps (INT, default 20, range 1–10000): The number of sampling steps to run.

• cfg (FLOAT, default 4.0, range 0.0–100.0): The classifier-free guidance strength. Higher values generally make the result follow the conditioning more strongly.

• sampler_name (COMBO, default "euler_ancestral"): The sampler to use during sampling.

• scheduler (COMBO, default "simple"): The scheduler used to control the sampling schedule.

• denoise (FLOAT, default 1.0, range 0.0–1.0, step 0.01): The amount of the latent to denoise. Use 1.0 for full sampling from scratch. Lower values preserve more of the input latent structure, which is useful for img2img-style refinements.

• image_output (COMBO): Controls what is done with the decoded image:

  • Hide: Do not preview, save, or send the image.
  • Preview: Show the image in the UI.
  • Save: Save the image to the output directory.
  • Hide&Save: Save the image without displaying it.
  • Sender: Send the image to a matching EasyUse image receiver.
  • Sender&Save: Send and save the image.

• link_id (INT): The link/receiver ID used when image_output is Sender or Sender&Save.

• save_prefix (STRING, default "ComfyUI"): The filename prefix used when saving generated images.

• seed (INT): The seed for random noise generation. Using the same seed with the same settings produces reproducible results.

Optional

• image_to_latent_c (IMAGE): An optional input image to encode into the Stage C latent using encode_vae_name. This is useful when you want to start sampling from an image rather than an existing latent.

• latent_c (LATENT): An optional Stage C latent to use as the starting point for sampling. If this is provided, it takes precedence over encoding image_to_latent_c.

• model_c (MODEL): An optional Stage C model to use for sampling. If omitted, the model in pipe is used.

Outputs

• pipe (PIPE_LINE): The updated EasyUse pipeline, carrying the sampled latent and the settings used during sampling.

• model (MODEL): The model that was used for sampling.

• latent (LATENT): The latent produced by the final sampling step.

Usage Notes

• The node expects the pipe to contain the conditioning and model state needed by the sampler. Use an EasyUse pipeline source node to build it before connecting this node.
• If both image_to_latent_c and latent_c are supplied, the provided latent is used as the starting point and image_to_latent_c is ignored.
• The two VAE dropdowns allow the encode and decode stages to use different VAEs. Make sure the selected VAEs are compatible with the checkpoint and latent format used by the pipeline.
• denoise is especially useful when starting from an encoded image: setting it below 1.0 will keep part of the original image structure while allowing the sampler to refine it.
• link_id only matters when image_output is set to Sender or Sender&Save; it routes the output to the matching receiver in the workflow.