Sigmas Mult

Sigmas Mult is the multiply-by-a-number cousin of Sigmas Add. Where Add shifts every noise level by a constant, Mult scales the whole schedule by a factor. Same family of tiny sigma-surgery nodes from RES4LYF, same job: reshape the curve your sampler walks down without swapping to a different scheduler.

If sigmas are new to you: your sampler denoises in steps, and the sigma schedule is the list of noise levels for those steps - high at the start, near zero at the end. The scheduler node builds it. RES4LYF exists because on flow-matching models (Flux, Chroma, HiDream, Wan, SD3.5) the exact shape of that curve is where a lot of the quality lives, and the pack gives you nodes to bend it. Multiplying is one of the cleanest ways to bend it, because it stretches the whole thing proportionally instead of lopping off the top.

How it works

Feed it a SIGMAS schedule and a multiplier. Every sigma gets multiplied by that number, and you get a scaled schedule back. A multiplier above 1 pushes the noise levels up (more aggressive denoising per step); below 1 pulls them down (gentler, which can read as cleaner but also flatter). There's also an optional second schedule input, sigmas2 - when you provide it, you're multiplying one schedule against another rather than against a flat constant, which lets you modulate one curve by the shape of a second.

The inputs and outputs that matter

• sigmas (SIGMAS) - the schedule you're scaling.
• multiplier (FLOAT, default 1) - the scaling factor and the main knob. 1 is a no-op; values a little under 1 (say 0.9–0.98) are a common, safe first experiment.
• sigmas2 (SIGMAS, optional) - a second schedule to multiply against instead of using the flat multiplier. Leave it unconnected for normal use.

Output is a single SIGMAS, which goes into the sigmas input of your sampler. As always with an external schedule, connecting it overrides the scheduler and step count you set inside the sampler node.

How to install it

ComfyUI Manager: search RES4LYF, install, restart. Manual: cd ComfyUI/custom_nodes && git clone https://github.com/ClownsharkBatwing/RES4LYF/, activate your venv, pip install -r requirements.txt (portable build: use the embedded python's pip), then restart Comfy and hard-refresh with F5.

For the nested sampler dropdown, install rgthree-comfy too and enable "Auto Nest Subdirectories" in its settings.

Common issues

The classic gotcha is confusing this with denoise strength. It isn't that. Scaling sigmas changes the levels the sampler targets, not how many steps run or how much of the image is repainted - if you want partial denoise for img2img, use the sampler's own denoise setting. Reach for Mult when you specifically want the whole noise curve stretched.

Two failure signs to read correctly, both of which are the node doing its job: crank the multiplier well above 1 and you can hand the model noise levels it wasn't trained to see, which shows up as artifacts or blowout. Pull it far below 1 and the image goes soft and under-cooked because the sampler never climbs high enough to make real changes. Small moves, fixed seed, one variable at a time - this is experimental territory and the point of a fixed seed is to actually see what the multiplier did.