☁️BizyAir Multiply Sigmas (stateless)

Everything in and out of this node is SIGMAS, a list of noise levels, no BIZYAIR_MODEL, no cloud model call, no account-specific state. That's almost certainly what the "(stateless)" in the display name is pointing at: unlike most of this pack, which routes through your SiliconFlow account and a hosted model, this is a pure math operation on a plain list of numbers. It's the cloud pack's version of ComfyUI's own custom-sampling "Multiply Sigmas" node, same idea, same job.

What it actually does

A sigma schedule describes how much noise gets removed at each sampling step, front-loaded (a lot of noise removed early) versus back-loaded (noise removal concentrated later) shapes the whole character of a generation. This node takes an existing sigmas list (typically produced upstream by a scheduler node) and multiplies a portion of it by factor, where start and end (both 0-1, defaulting to 0 and 1, the full range) define which fraction of the schedule gets scaled. Set factor below 1 to reduce noise over that window, sampling effectively skips ahead relative to what the base schedule intended, or above 1 to add more perceived noise there, telling the sampler it has more work left to do than the schedule originally said.

If that sounds close to what BizyAir_LyingSigmaSampler does, that's because it's a related idea from a different angle: Lying Sigma Sampler wraps a sampler and applies its lie continuously across a percentage window as sampling runs, this node instead directly edits the sigma list itself before sampling even starts, and slots into a custom-sampling graph as a discrete step rather than a sampler wrapper. Which one to use depends on how you've built the rest of the chain, both are ways of telling the model something other than the literal truth about how much noise remains.

Where it fits

This is strictly a custom-sampling node, it has no place in a plain KSampler workflow. Typical chain: a scheduler node produces a base SIGMAS list, this node scales part of it, and the result feeds into whatever node in your graph actually runs the denoise loop against a BIZYAIR_MODEL. start/end let you target the effect narrowly, scaling only the early steps to change how aggressively composition locks in, or only the late steps to change how much fine detail gets added, rather than applying factor uniformly across the whole schedule.

Installing it

Through ComfyUI Manager: search "BizyAir", install, restart. Or clone directly:

cd ComfyUI/custom_nodes && git clone https://github.com/siliconflow/BizyAir.git

This node itself does pure local math and likely needs no API call on its own, but it's only useful wired into a chain that eventually does need a BizyAir/SiliconFlow API key for the actual model call, set via the "click to login" prompt the README points to on first use. Signup involves phone-number verification, and some users outside China have reported friction there.

Common issues

An oversharpened or artifact-heavy result from a custom-sampling chain using this node is usually factor pushed too far from 1.0, or a start/end window that overlaps awkwardly with another sigma-manipulation node elsewhere in the graph, back it off before assuming the base sampler or model is at fault. If the effect seems to do nothing, confirm sigmas is actually connected to a real scheduler output and not left on a default or disconnected socket.

As with the rest of this pack, if the whole chain stalls or errors and nothing here looks obviously wrong, check the BizyAir API key on the downstream model-calling nodes first. Separately: BizyAir occasionally ends up installed without you asking for it, a known ComfyUI Manager quirk (misresolving an unrelated missing-node error on an older shared workflow), not malware, and removable via the Manager.