Sigmas Rescale

Sigmas Rescale is a better way to set denoise strength - and the node's own description says so plainly. The problem it fixes: ComfyUI's KSampler (and most nodes with a denoise value) implement partial denoise by slicing the sigma schedule according to step count. That's a crude cut. Sigmas Rescale instead rescales the schedule by noise level, which the pack's author reckons generally gives better results for img2img. Same goal - start from a noisier or cleaner point - different, more principled mechanism.

It's the kind of small, sharp utility that RES4LYF is full of, and its solid click-through suggests people doing img2img go hunting for exactly this.

How it works

You feed it a full SIGMAS schedule plus a start and end noise level, and it rescales the curve to span that range rather than chopping steps off the front. Because you're setting the actual noise endpoints, the denoise behavior tracks more consistently across different schedulers than the slice-based approach does. There's a bonus baked in: if you reverse start and end, it flips the schedule, which is what unsampling wants.

The inputs and output

• sigmas (SIGMAS) - the schedule to rescale. Required; it modifies an incoming curve, so wire a scheduler into it.
• start (default 1) - the noise level to begin at. For img2img, lowering this is roughly "denoise less / keep more of the input."
• end (default 0) - the noise level to finish at. Normally 0 (fully denoised).

Set start lower than 1 to preserve more of your source image; set start below end and the node flips the schedule for unsampling workflows. The single output is sigmas_rescaled (SIGMAS) → your sampler's sigmas input.

Installing it

ComfyUI Manager: search RES4LYF, install, restart. Manually:

cd custom_nodes
git clone https://github.com/ClownsharkBatwing/RES4LYF/
cd RES4LYF
pip install -r requirements.txt

pip install inside the venv (or portable Python), restart, hard-refresh F5. No downloads.

Common issues

The main confusion is expecting this to behave like the denoise slider on a KSampler - it's deliberately different, and that's the point, so a start of 0.7 here won't map one-to-one to denoise 0.7 elsewhere. Treat it as its own thing and dial it in by eye. Because feeding a sampler sigmas overrides that sampler's own scheduler and its denoise, set your denoise here, not in both places, or they'll fight. The flip-on-reverse behavior is easy to trigger by accident: if your output comes out looking inverted or wrong for a normal generation, check that you didn't set start below end. And preview the rescaled curve with SigmasPreview before trusting it. If the node's missing, RES4LYF isn't installed.