ClownOptions Sigma Scaling

This is one of the deeper cuts in the RES4LYF options family, and it does two loosely related jobs: scaling the noise the sampler injects, and running the "lying sigma" trick where the model is told a slightly different noise level than the one actually present. Both are trajectory-shaping tools - the kind of "sigma, latent, and noise manipulation" the pack was built to expose.

Fair warning: this is advanced, taste-driven territory. You do not need it for a good first result, and if you're new to the pack, skip it. It earns its place when you already have a solid base and want to squeeze out grain, sharpness, or a particular texture.

What the two halves do

Noise scaling. The s_noise family multiplies how much SDE noise gets added. Turn it up for more texture and grain, down for a cleaner, smoother finish. Because RES4LYF samples with substeps, there's a matching control for those, and an anchor value that keeps SDE noise consistent across steps.

Lying sigmas. The lying controls feed the model a fudged sigma - telling it the image is noisier or cleaner than it really is at that point. It's a deliberate mismatch that can sharpen detail or shift how aggressively the model denoises. There's a forward version and an inverse version, each with a step at which it kicks in.

The inputs that matter

• s_noise (default 1) - the main noise-scaling multiplier. 1.0 is neutral. This is the one most people touch: nudge up for grain, down for smooth.
• s_noise_substep (default 1) - the same, for substeps on _2s/_3s samplers.
• noise_anchor_sde (default 1) - anchors SDE noise level for consistency; leave at 1 unless you know why you're changing it.
• lying / lying_inv (default 1 each) and lying_start_step / lying_inv_start_step - the lying-sigma scale factors and the steps they start on. Experimental; move in small amounts.

Optional s_noises and s_noises_substep accept SIGMAS curves if you'd rather schedule the scaling per-step instead of using a flat value, plus the usual options chain input. Output is a single options (OPTIONS) for the sampler.

Installing RES4LYF

ComfyUI Manager: search RES4LYF, install, restart. Or clone:

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

Portable users: embedded pip under python_embedded/Scripts/. Restart, hard-refresh with F5. No model files.

Common issues

Everything's grainy now. You pushed s_noise too high. It's a multiplier on noise - back toward 1.0. Remember it interacts with your sampler's eta: high eta plus high s_noise compounds.

The lying controls did something weird. They're supposed to - deliberately misreporting sigma is an inherently unstable trick. Move them in tiny increments from 1.0, and if in doubt leave them at the defaults; the noise scaling is the part most people actually want.

No change at all. Defaults are neutral (everything at 1). And, as with all ClownOptions nodes, the options output has to reach a RES4LYF sampler to take effect - a dangling chain does nothing.