☁️BizyAir RandomNoise

A standard KSampler node bundles seed handling and noise generation invisibly inside itself - you type a number, it works, you never think about the step in between. This node exists because ComfyUI also supports building a sampler out of separate pieces (a noise source, a scheduler, a sampler algorithm, wired together manually instead of packed into one big node), and something has to be the piece that actually turns a seed into a literal starting-noise tensor. That's this node's entire job.

Why the modular approach exists

Splitting sampling into separate nodes is what lets advanced sampler packs plug in - swapping in a different solver or a different sigma schedule without ComfyUI needing a bespoke "KSampler but with this one custom scheduler" node for every combination. If you've ever seen a workflow that chains something like a noise node into a scheduler into a SamplerCustom, this is exactly that pattern, just running BizyAir's cloud model instead of a locally loaded one. It's more wiring for something a stock KSampler does in one box, but it's the entry point for anyone who wants more control over exactly how the sampling trajectory is built.

The input and output

• noise_seed - the only field. Default 0, and the range goes essentially to the top of a 64-bit unsigned integer (roughly 1.8 × 10¹⁹), which is a much wider ceiling than the 32-bit seed ranges you'll see on other nodes in this same pack. In practice this doesn't change how you use it - pick a number, or leave it to whatever your workflow randomizes it to, the same as any other seed field. It just means you're not going to run out of unique seeds.

Output: NOISE, which feeds into the scheduler/sampler chain rather than directly into a KSampler - it's built for the modular custom-sampling path, not the all-in-one node.

Installing it

Through ComfyUI Manager: search BizyAir, click Install, restart. Or manually:

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

Set your BizyAir API key on first use.

Where people get burned

The most common confusion is expecting this node to behave like a KSampler's seed field and being surprised when nothing happens on its own - a NOISE output by itself doesn't generate an image; it has to flow through a scheduler and a sampler node before it does anything visible. If your workflow "runs" with no errors but nothing comes out the other end, check that the noise output actually made it all the way through the chain to a sampling node, rather than assuming the seed itself is the problem.

The second thing worth knowing: if you're chasing the same kind of sampler control that advanced local node packs offer - dozens of solver choices, unconventional sigma schedules built specifically for flow-matching models - that ecosystem exists and is worth knowing about, but it's a separate local install, not something this cloud noise node unlocks by itself. This node gets you the entry point into modular sampling on BizyAir's hosted models; the rest of that chain (scheduler, sampler selection) still has to be built out of whatever BizyAir offers alongside it or handed off to plain NOISE/LATENT-typed nodes once you're past this step.