KSampler

This is a drop-in replacement for ComfyUI's core KSampler - same inputs, same sampling behavior, nothing lost - with one addition: an INFO output that hands back a text string of the settings you sampled with, plus how long it took. If you've ever burned a generation onto disk and then, later, had no idea what steps/cfg/sampler combination made it, this is the node that fixes that at the source rather than after the fact.

How it works

It's the exact same denoising loop as the standard KSampler - model in, conditioning in, latent in, latent out - so there's no new sampling behavior to learn. What's different is that the node also assembles a readable summary of what it just did and exposes it as a second output. Wire that INFO string into this pack's Add Text Overlay and you get images that are self-labeled with the exact settings that produced them - handy for A/B comparisons, or for building a contact sheet with Grid Filler where every tile needs to say what it was.

The inputs and outputs that matter

All standard KSampler fields, with the author's own tooltips (these are the pack's descriptions, not guesses):

• model - the model doing the denoising.
• seed - the noise seed.
• steps - default 20, "the number of steps used in the denoising process."
• cfg - default 8, "balances creativity and adherence to the prompt. Higher values result in images more closely matching the prompt however too high values will negatively impact quality." Worth a flag: 8 is ComfyUI's long-standing stock default, but a lot of current checkpoints are guidance-distilled and run best at CFG 1 - Z-Image Turbo, Flux 2 Klein distilled, and most Lightning/Turbo/Hyper variants included. If you're on one of those, check what the checkpoint actually wants rather than trusting this default; at CFG 1 the negative-conditioning pass isn't even computed, so your negative prompt is doing nothing regardless of what you type into it.
• sampler_name - 44 built-in choices (euler, dpm_2, heun, and so on). This is ComfyUI's stock sampler list, not the larger set from a dedicated pack like RES4LYF - if you need something like res_2m or an implicit solver, that's a separate install, not something hiding in this dropdown.
• scheduler - 9 choices (simple, karras, exponential, and others), "controls how noise is gradually removed to form the image."
• positive / negative - your conditioning.
• latent_image - the latent to denoise.
• denoise - default 1, "lower values will maintain the structure of the initial image allowing for image to image sampling."

Outputs: LATENT (the denoised result, same as stock) and INFO - the settings-and-timing string, unique to this node.

Installing it

Via ComfyUI Manager: search "iTools" or "ComfyUI-iTools," install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/MohammadAboulEla/ComfyUI-iTools

then restart. No special dependencies beyond what standard KSampler already needs.

Troubleshooting

Output looks identical to what stock KSampler would give you. That's by design - this node's sampling math is unchanged. If your results seem off, that's a settings problem (checkpoint mismatch, CFG too high for a distilled model, wrong scheduler for a flow-matching model) rather than anything this node changed.

INFO string isn't showing what you expect, or isn't connected to anything. It's a plain STRING output like any other - nothing renders it automatically. Route it to a Text Preview node to actually see it, or into Add Text Overlay to burn it onto the image itself.

Negative prompt seems to do nothing at all. If cfg is set to 1, that's expected and not specific to this node - classifier-free guidance's unconditioned pass isn't run at CFG 1, so the negative conditioning has nothing to act through. Raise CFG if you need the negative prompt to matter, or accept that a CFG-1 checkpoint wants constraints written positively instead.