PreSampling (LayerDiffuse)(layerDiffusionSettings)

This node is a pre-sampling control for Layer Diffusion workflows. It attaches Layer Diffusion settings and sampler parameters to a PIPE_LINE, so the downstream sampler nodes run with transparency-aware attention or convolution behavior instead of a plain latent denoise.

Description

PreSampling (LayerDiffuse) is used before the KSampler in an EasyUse pipeline. It takes an existing PIPE_LINE, applies the selected LayerDiffusion method and weight, and passes the modified PIPE_LINE forward. It also exposes the typical sampling controls — steps, CFG, sampler, scheduler, denoise, and seed — so you can configure the whole generation in one place. When optional image, blended_image, and mask inputs are provided, the node can participate in image-to-image or masked LayerDiffuse generation.

Inputs

• pipe (PIPE_LINE, required):
  The upstream pipeline containing model, positive/negative conditioning, latents, and other sampler settings. This node reads from and writes back to this pipeline.

• method (COMBO, required):
  Selects the LayerDiffusion injection method to apply. The available choices depend on your LayerDiffuse installation. Pick the method that matches the LayerDiffuse model and workflow you are using.

• weight (FLOAT, required):
  Default: 1.0 — Minimum: -1.0 — Maximum: 3.0 — Step: 0.05
  Controls the strength of the Layer Diffusion effect. Higher values make the transparency-aware layers more influential; lower or negative values weaken or invert the effect.

• steps (INT, required):
  Default: 20 — Minimum: 1 — Maximum: 10000
  Number of sampling steps to run.

• cfg (FLOAT, required):
  Default: 8.0 — Minimum: 0 — Maximum: 100
  Classifier-free guidance scale.

• sampler_name (COMBO, required):
  Default: euler
  The sampler to use for denoising. Options follow ComfyUI's standard sampler list.

• scheduler (COMBO, required):
  Default: normal
  The scheduler used with the sampler.

• denoise (FLOAT, required):
  Default: 1.0 — Minimum: 0 — Maximum: 1 — Step: 0.01
  The fraction of the latent to denoise. 1.0 runs a full denoise; lower values preserve more of the original structure when an input image is used.

• seed (INT, required):
  Seed for the sampling process. Set a fixed value to make the generation reproducible. If the widget is left randomizing, the value may change on each run.

• image (IMAGE, optional):
  Optional input image for image-to-image or masked Layer Diffusion workflows.

• blended_image (IMAGE, optional):
  Optional pre-blended image used by Layer Diffusion as the compositing target. Supplying this can help the model understand how the foreground layer should fit into the scene.

• mask (MASK, optional):
  Optional mask that defines the region to generate or edit. The masked area receives the Layer Diffusion treatment while unmasked areas are preserved.

Outputs

• pipe (PIPE_LINE):
  The updated pipeline, ready to be passed to a downstream sampler node. The emitted pipeline includes the configured sampling parameters and LayerDiffuse settings.

Usage Notes

• Use this node as the last pre-sampling step before a sampler that supports PIPE_LINE, such as the KSampler variants in ComfyUI-Easy-Use.
• The optional image, blended_image, and mask inputs are not all required for every method. For a plain LayerDiffuse generation, you can leave them disconnected, and the node will work in a text-to-image style with Layer Diffusion applied.
• When using denoise < 1.0, connect image (and usually mask) so the sampler has something to align the partial denoise against.
• weight can go negative. In some LayerDiffuse methods this can produce an inverted or corrective effect; experiment with small steps around typical values such as 0.8, 1.0, or 1.2.
• If the output looks unblended or the foreground/background separation is too weak, raise weight or check that method matches the LayerDiffuse model you loaded.
• steps, cfg, sampler_name, and scheduler behave exactly like the same settings on a regular sampler node, so existing sampler presets can be reused here.