EasyKSampler (Custom)

This is Easy-Use's pipe-native version of ComfyUI core's SamplerCustom pattern - the sampler that hands you back two latents instead of one. Where easy kSampler just gives you a final image, easy kSamplerCustom also gives you the raw sampled latent and a separately-computed fully-denoised latent, which matters the moment your workflow does something clever with noise: custom guiders, advanced schedulers, or anything that wants to inspect or reuse the "what did the model think the clean image looks like" estimate rather than just the end result.

Why you'd reach for it

Most people never need this - easy kSampler covers the normal case fine. Reach for the Custom version when your workflow is doing sampler-level work rather than just "run the standard pipeline": chaining custom noise sources, feeding the denoised latent into a second pass that needs the clean estimate rather than the noisy working latent, or building something that mirrors what core ComfyUI's SamplerCustom / SamplerCustomAdvanced nodes are for, but wired through Easy-Use's pipe so it slots into the rest of an Easy-Use graph instead of needing a separate model/conditioning/latent hookup.

How it works

Like every sampler in this pack, it reads its configuration off the pipe - model, conditioning, latent, sampler, scheduler, steps, CFG, all of it set upstream by a loader and a presampling node. What's different from the plain easy kSampler is the output side: instead of collapsing everything down to just an image, it exposes the sampling process's two distinct latent states separately, so downstream nodes can pick whichever one they actually need.

The inputs and outputs that matter

• pipe - the one required wire, carrying model, prompts, latent, and sampling settings from your loader/presampling chain.
• image_output - what to do with the rendered image: Hide, Preview, Save, Hide&Save, Sender, Sender&Save, or None. Note the extra None option here that plain easy kSampler doesn't have - pick it if you only care about the latent outputs and don't want an image rendered at all.
• link_id - pairs with a receiver node when using a Sender output mode.
• save_prefix - filename prefix if saving.
• Optional model - override the pipe's model at the last moment, same pattern as the plain sampler (handy for a model patch applied right before sampling).

Outputs: pipe (pass-through for chaining), output (the raw sampled LATENT), denoised_output (the fully-denoised LATENT estimate), and image.

Installing it

Ships with the base pack. ComfyUI Manager: search ComfyUI Easy Use, install, restart. Manually:

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

then install.bat on Windows or pip install -r requirements.txt, restart. No extra models for the sampler itself - whatever checkpoint your pipe already carries.

Common issues & troubleshooting

Not sure which latent output to use. If you're not deliberately working with a custom guider or a two-stage denoising setup, you almost certainly want denoised_output, not output - it's the cleaner, fully-resolved latent. output carries whatever the raw sampling step produced, which can include leftover noise depending on your sampling configuration. When in doubt, VAE-decode both and look at them; the difference is usually obvious.

This feels like overkill for a normal generation. It probably is - if you don't have a specific reason to need two separate latent outputs, easy kSampler is the simpler, more common node and does everything a standard generate step needs. Reach for Custom only when something downstream specifically wants the raw/denoised split.

Set image_output to Sender and nothing shows up elsewhere. Sender mode only works if a matching receiver node elsewhere in the graph shares the same link_id. If nothing's receiving it, either add a receiver node or just wire the image output directly to a Preview/Save node instead.