☁️BizyAir Conditioning (Average)

This is BizyAir's cloud version of one of ComfyUI's oldest conditioning-math nodes, and it does exactly what the name says: it takes two already-encoded conditionings and blends them by weight, at the embedding level, rather than making you write one prompt that tries to describe both things at once. It's the node you reach for when "write a prompt that's 70% cyberpunk city and 30% medieval castle" is a worse plan than just encoding each idea separately and mixing the vectors.

How it's different from just combining prompts as text. If you concatenated "cyberpunk city" and "medieval castle" into one string and encoded that, the text encoder would have to reconcile two competing ideas inside a single pass - you get whatever compromise the model lands on, and it's mostly out of your hands. Averaging two separately-encoded conditionings instead blends the results of two clean, independent encodes, which gives you an actual dial: push the weight toward one side or the other and watch the balance shift, predictably, without re-encoding anything.

Inputs that matter. conditioning_to and conditioning_from are the two conditionings being blended - both required, both BIZYAIR_CONDITIONING, meaning they need to come from a BizyAir CLIP/text-encode node upstream, not a stock one. conditioning_to_strength is the actual dial: a float from 0 to 1, default 1, step 0.01. At 1.0 you get pure conditioning_to and conditioning_from contributes nothing; at 0.0 it's the reverse; 0.5 is an even split. The naming is easy to misread at a glance - it's specifically the weight on conditioning_to, not a generic "blend amount," so 0.7 means 70% conditioning_to and 30% conditioning_from, not the other way around.

Output: a single BIZYAIR_CONDITIONING, the blended result - wire it into whatever expects a positive (or negative) conditioning further down your BizyAir chain, same as any other conditioning node in the pack.

A word on when this actually helps. Conditioning-level blending is an SD 1.5/SDXL-era technique through and through, built around CLIP's fairly literal, additive relationship between prompt and output. If you're running this against a newer LLM-encoded base through BizyAir's catalog, the text encoder is reading your prompt more like an instruction than a bag of tokens, and a mathematical blend of two encoded instructions doesn't necessarily land where you'd expect. It's still worth trying, but don't be surprised if it behaves less predictably on a modern base than the classic "blend two SDXL prompts" use case it was designed around.

Install: the pack comes as one unit - search "BizyAir" in ComfyUI Manager and install, git clone https://github.com/siliconflow/BizyAir.git into custom_nodes and restart, or comfy node install bizyair. There's no separate download for this node specifically; what you do need separately is a BizyAir account with your API key set on first use, or this node (and everything feeding it) just fails to run.

Troubleshooting: if the blend doesn't seem to be doing anything, double-check you're actually moving conditioning_to_strength and not leaving it at the 1.0 default, which silently ignores conditioning_from entirely - an easy thing to forget when you've just wired the node up and haven't touched its widget yet. If the node errors or hangs outright, that's almost always the API key, not the math - confirm you're authenticated before assuming the blend logic is broken.