KSampler (Advanced) node in ComfyUI

KSampler (Advanced) is KSampler with step-range control: add_noise, start_at_step, end_at_step and return_with_leftover_noise, used for base + refiner chains and two-model sampling.

Node pack: ComfyUI core

Category: Sampling

What KSampler (Advanced) does

The advanced sampler runs the same denoising loop but lets you sample only a slice of the schedule. With return_with_leftover_noise enabled and end_at_step set to, say, 20 of 25, the first sampler stops early and hands a still-noisy latent to a second KSampler (Advanced) that starts at step 20 with add_noise disabled. This is how the SDXL base/refiner workflow and many two-model tricks are built.

There is no denoise slider; denoise strength is expressed through the step range instead. start_at_step 10 of 20 steps is the same as denoise 0.5 on a plain KSampler.

Inputs

Name	Type	What it is
model / positive / negative / latent_image	various	Same as KSampler.
add_noise	COMBO	enable to add fresh noise to the latent before sampling (first stage), disable to continue a partly sampled latent.
noise_seed	INT	The seed for the noise.
steps / cfg / sampler_name / scheduler	various	The full schedule; both stages should use the same total steps.
start_at_step	INT	First step to run. 0 for the first stage.
end_at_step	INT	Step at which to stop. 10000 means run to the end.
return_with_leftover_noise	COMBO	enable when another sampler will continue from this latent.

Outputs

Name	Type	What it is
LATENT	LATENT	The (possibly partially sampled) latent.

Skip the setup: ComfyUI workflows preinstalled on the cloud

BitVector runs ready-made ComfyUI workflows on its own GPUs. No install, no missing nodes, no red boxes. Open it on your phone or laptop and generate in a minute.

Try BitVector

How to use KSampler (Advanced)

Stage one: add_noise enable, start_at_step 0, end_at_step 20, return_with_leftover_noise enable, steps 25.
Stage two with the second model: add_noise disable, start_at_step 20, end_at_step 10000, return_with_leftover_noise disable, same steps, scheduler and seed.
Decode the second latent.

Settings and tips

Keep steps, scheduler and seed identical in both stages; only the model, prompts and step range change.
For img2img without a denoise slider: steps 20, start_at_step 10 equals denoise 0.5.
SDXL refiner was tuned for the last 20 percent of steps; with modern SDXL checkpoints it is usually unnecessary.

Troubleshooting KSampler (Advanced)

Second stage output is pure noise

Why it happens

add_noise is enabled in the second stage, or return_with_leftover_noise was disabled in the first, so noise levels do not line up.

How to fix it

First stage: leftover noise enable. Second stage: add_noise disable, start_at_step equal to the first end_at_step.

Second stage looks over-smoothed or ignores the prompt

Why it happens

The refiner stage runs too many steps or uses a different scheduler than the base stage.

How to fix it

Give the second stage the last 15-25 percent of steps and the same scheduler.

Nothing happens in the second sampler (output equals input)

Why it happens

start_at_step is at or beyond steps, so there is nothing left to sample.

How to fix it

Make start_at_step smaller than steps and smaller than end_at_step.

Run this workflow from your phone

Every workflow on this page is preinstalled on BitVector with the models and custom nodes already in place. Pick one, type a prompt, done. Zero setup, nothing to download.

Try BitVector

Questions about KSampler (Advanced)

Can I use it as a normal sampler?

Yes: add_noise enable, start 0, end 10000, leftover disable is identical to KSampler at denoise 1.0.

How do I get denoise 0.35?

With 20 steps, start_at_step 13.

Related nodes

KSampler

ComfyUI core

KSampler runs the denoising loop: it takes the model, prompts and a latent and produces the finished latent image, controlled by seed, steps, cfg, sampler, scheduler and denoise.

Upscale Latent and Upscale Latent By

ComfyUI core

Upscale Latent resizes a latent to a new size (or by a factor) so a second KSampler can add detail at higher resolution: the classic hires fix.

VAE Decode

ComfyUI core

VAE Decode converts the sampled LATENT into a pixel IMAGE with the VAE; VAE Decode (Tiled) does the same in tiles for very large images.

Tired of fixing nodes? Let the cloud do it

BitVector keeps hundreds of ComfyUI workflows installed, updated and tested on fast cloud GPUs. No Python, no CUDA errors, no VRAM limits. Works from any browser.

Try BitVector