PreSampling

The whole idea behind ComfyUI-Easy-Use is that it splits a normal KSampler in two. Instead of one fat node that holds your steps, cfg, sampler, scheduler, seed and runs the diffusion, Easy-Use gives you a pre-sampling node that just configures those settings, and a separate sampler node that does the actual work. easy preSampling is the config half.

Why bother? Because it lets you preview and tweak your sampling parameters without committing to a full render, and because those settings ride inside the pipe - so once you set them here, every downstream Easy-Use node knows about them. If you've used the pack's loaders, this is the node that catches the pipe they hand you and stamps your sampler config onto it. It's TinyTerra's "pipe" pattern (the pack is built on top of TinyTerra) taken to its logical end: one wire carrying the whole state of your generation.

How it works

Easy-Use nodes pass around a bundle called PIPE_LINE - model, CLIP, VAE, positive, negative, latent, all in one connection. easy preSampling takes that pipe in, writes your sampling parameters into it, and passes it back out. Nothing is sampled yet. The image gets made later, when a node like easy kSampler receives this pipe and runs it. Splitting it this way is what lets you rewire samplers, fork the pipe into two different sampler branches, or A/B two parameter sets without duplicating your loader.

The inputs that matter

• steps (default 20) and cfg (default 8) - the two you'll touch most. Twenty steps and cfg 7-8 is the classic SD 1.5 / SDXL starting point. If you're on a distilled or flow-matching model, both of those defaults are wrong for you - distilled checkpoints want cfg 1 and a handful of steps, so don't just accept 8.
• sampler_name and scheduler - 44 samplers and 11 schedulers, including the good stuff like align_your_steps and gits that Easy-Use bundles in. euler / normal is a safe default; dpmpp_2m + karras is the reliable SDXL pick.
• denoise (default 1.0) - leave it at 1.0 for txt2img. Drop it to 0.4-0.6 when you feed an image in for img2img.
• seed - your reproducibility knob.

Two optional inputs decide what you're sampling from: image_to_latent encodes an image straight into a latent (img2img without a separate VAE Encode node), and latent takes a latent you already have. Wire in neither and it uses the empty latent from your loader.

The single output is pipe - send it to easy kSampler, easy kSamplerTiled, or any Easy-Use sampler.

How to install it

Easiest path is ComfyUI Manager: open it, search "ComfyUI Easy Use", install, restart. Manual install is a git clone into your custom_nodes folder:

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

Then run install.bat (Windows) or pip install -r requirements.txt from the pack folder to pull dependencies, and restart ComfyUI. On comfy.icu the pack is already installed, so you can drop the node straight in.

Common issues

The number one gotcha is nothing to do with the node: this thing only speaks pipe. If you're trying to feed it a raw MODEL or CONDITIONING, it won't fit - you need one of the Easy-Use loaders (easy fullLoader, easy a1111Loader, etc.) upstream to build the pipe first. That trips up everyone who drops a single Easy-Use node into an otherwise vanilla workflow.

Second: if your image comes out looking half-cooked or noisy, check that a sampler is actually consuming this pipe. easy preSampling configures but never renders - forget the sampler and you get nothing. And remember the defaults are tuned for SD 1.5/SDXL; on a Turbo or Lightning model, cfg 8 will burn the image every time.