PreSampling (Advanced)(samplerSettingsAdvanced)

Description

The PreSampling (Advanced) node (samplerSettingsAdvanced) is a pipeline configuration node in the EasyUse/PreSampling category. It does not perform sampling itself; instead, it prepares a PIPE_LINE with advanced sampler settings for a downstream sampling node. It exposes the same style of controls as a KSampler Advanced node — steps, CFG, sampler, scheduler, step range, noise injection, seed, and leftover-noise behavior — and returns the modified pipeline.

The node also accepts optional image_to_latent and latent inputs so you can establish the latent that will be sampled from, either from an input image or from an existing latent.

Inputs

Required

• pipe (PIPE_LINE): The incoming pipeline carrying model, conditioning, and other state. The node reads from this pipe and returns it with updated sampling parameters.
• steps (INT, default 20, range 1–10000): Number of sampling steps. Higher values generally improve quality at greater computation cost.
• cfg (FLOAT, default 8, range 0–100): Classifier-free guidance scale. Higher values strengthen conditioning; lower values allow more variation.
• sampler_name (COMBO): The sampler to use. Available options depend on the installed samplers.
• scheduler (COMBO): The scheduler to use. Available options depend on the installed schedulers.
• start_at_step (INT, default 0, range 0–10000): First step of the sampling range. Keep 0 to sample from the very beginning.
• end_at_step (INT, default 10000, range 0–10000): Last step of the sampling range. The default 10000 effectively disables an early cutoff; lower values stop sampling before the full schedule.
• add_noise (COMBO, default "enable (CPU)"): How initial noise is added. "enable (CPU)" adds noise using CPU generation, "enable (GPU=A1111)" uses GPU/A1111-compatible noise generation, and "disable" starts without adding initial noise.
• seed (INT): Random seed for noise generation. Use the same seed to reproduce the same noise.
• return_with_leftover_noise (COMBO, options "enable" / "disable"): Controls whether the latent returned for the next stage keeps residual noise. "enable" is often used when another sampler or refinement pass will follow; "disable" returns a cleaner latent.

Optional

• image_to_latent (IMAGE): An input image to use as the latent source. It will be converted into latent space so sampling can operate on it.
• latent (LATENT): A latent to use directly. Provide this when you already have a latent and do not need to encode an image.

Outputs

• PIPE_LINE: The updated pipeline with the chosen pre-sampling settings and the resolved latent. Connect it to a sampler node to execute the sampling.

Usage Notes

• This node is intended to be connected to a downstream sampling node that consumes a PIPE_LINE; it prepares the settings, not the final image.
• The step-range controls (start_at_step and end_at_step) let you run only part of the denoising schedule. For a full normal generation, leave start_at_step at 0 and end_at_step at 10000.
• Use add_noise set to "disable" when starting from a latent that should not receive a fresh burst of noise.
• return_with_leftover_noise matters most for multi-pass workflows: enabling it can help a subsequent refinement pass, while disabling it gives a more finalized latent.
• If the pipeline already contains a latent, you can leave both optional latent inputs disconnected. Otherwise, connect either image_to_latent or latent to establish the latent to be sampled.