PreSampling (Advanced)

This is easy preSampling with the advanced sampling controls bolted on. Same job - it configures your sampler settings into the pipe rather than running them - but it exposes the extra knobs you need for multi-pass and partial-denoise setups: where in the step schedule to start, where to stop, and how noise is handled. If you're building a two-stage sampler (base pass then a refiner pass) or wiring up an SDXL base+refiner handoff, this is the pre-sampler you reach for instead of the plain one.

How it works

Like every Easy-Use pre-sampler, it takes a PIPE_LINE in, writes your parameters onto it, and hands the pipe back out without sampling anything. The difference is the four extra fields that mirror ComfyUI's own KSamplerAdvanced: you're no longer telling it "denoise this fully from noise," you're telling it which slice of the denoising schedule this pass owns. That's what makes chained passes possible - pass one runs steps 0-15 and leaves noise on the latent, pass two picks up at step 15 and finishes it.

The inputs that matter

• start_at_step and end_at_step - the slice of the schedule this pass handles. For a base pass you might run 0 to 15; the refiner then starts at 15. Leaving them at 0 and 10000 (the max) means "do the whole thing," same as the basic node.
• return_with_leftover_noise - set this to enable on any pass that hands off to another sampler. It leaves the latent noisy so the next stage has something to finish. disable it on the final pass, or your image stays grainy.
• add_noise - enable (CPU) is the default and the reproducible choice. enable (GPU=A1111) matches how Automatic1111 seeds noise on the GPU, which is handy if you're trying to reproduce a webui image. disable skips adding noise entirely (for a pass that continues an already-noised latent).
• steps, cfg, sampler_name, scheduler, seed - the usual crew, same as the basic pre-sampler.

Optional image_to_latent and latent inputs set what you're sampling from. Output is a single pipe for your sampler.

How to install it

Grab the whole pack once and this comes with it. Via ComfyUI Manager, search "ComfyUI Easy Use" and install. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/yolain/ComfyUI-Easy-Use

Run install.bat or pip install -r requirements.txt, restart ComfyUI. It's preinstalled on comfy.icu.

Common issues

The classic mistake is leaving return_with_leftover_noise disabled on your first pass - the base pass fully denoises, the refiner then has clean pixels to work on and either does nothing or oversharpens. If your two-pass setup looks worse than a single pass, that flag is the first thing to check, followed by making sure the refiner's start_at_step matches the base's end_at_step.

The enable (GPU=A1111) noise mode is genuinely useful but it's a reproducibility choice, not a quality one - it doesn't make images better, it makes them match a different tool. Don't flip it expecting sharper output.

And the standing Easy-Use rule applies: this node needs a pipe from an Easy-Use loader upstream, and it needs an Easy-Use sampler downstream to actually render. It's the middle of a sandwich, useless on its own.