☁️BizyAir SplitSigmasDenoise

This is the sibling to BizyAir SplitSigmas, and it solves the same problem - cutting a sigma schedule in two so you can hand each half to a different sampler - but it lets you pick the cut point by how much denoising has happened instead of by counting raw steps. That matters more than it sounds, because "step 8" means something different on a 12-step schedule than a 40-step one, while "40% denoised" means the same thing regardless of how many steps you're using.

Two-stage sampling is the pattern this exists for: Wan 2.2's high-noise/low-noise split is the clearest real-world example in current workflows, where an early pass handles composition and motion on a noisier schedule and a later pass refines detail once most of the denoising work is done. Anywhere you're chaining two samplers or two models across one generation, this is the node that decides where the handoff happens.

Same as its sibling, this one is a plain local utility rather than a cloud call - its sigmas input and both outputs are the standard SIGMAS type, not one of BizyAir's cloud-routed BIZYAIR_* types, so it runs without needing your API key.

How it works

You give it a schedule and a denoise fraction between 0 and 1. It finds the point in the schedule corresponding to that fraction of total denoising progress and splits there, returning everything before as one schedule and everything after as another.

Inputs and outputs

• sigmas - the schedule to split, type SIGMAS.
• denoise - a float from 0 to 1, defaulting to 1, in steps of 0.01. This is the fraction of denoising to complete before the split. A value of 0.5 splits roughly at the halfway point of the denoising process, regardless of how many actual steps that spans.

It outputs high_sigmas (the earlier, noisier segment) and low_sigmas (the later segment closer to a clean result) - wire high_sigmas into your first-pass sampler and low_sigmas into whatever resumes the job.

Installing BizyAir

Standard pack install applies. Through ComfyUI Manager: search "BizyAir," install. Or clone it yourself:

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

Restart ComfyUI afterward, or use comfy node install bizyair via Comfy-CLI. You don't strictly need an API key for this specific node to function, but you'll want one set up if your workflow uses any of BizyAir's other, cloud-backed nodes alongside it - the README treats setting your key as a first-use step for the pack as a whole.

Common issues

The main thing to get right is understanding that denoise here describes a position in the schedule, not the strength of an img2img-style denoise pass - don't confuse it with the denoise slider on a KSampler. Setting it to 1 (the default) means "split at the very end," which will leave low_sigmas essentially empty; if you want a real two-stage split, pick something in the middle, like 0.3–0.6, depending on how much work you want each stage to do.

Beyond that, the same caveat as SplitSigmas applies: this node only earns its place in a workflow that's genuinely built around two sequential sampling passes. If you're running a single model end to end, there's nothing here for you - feed your full schedule to one KSampler and skip this entirely.