☁️BizyAir ConditioningZeroOut

The simplest node in this whole pack, and also one of the easier ones to reach for without fully understanding why. It takes a conditioning and zeroes out its embedding - same shape and type, but stripped of whatever it used to encode, functionally "nothing." The classic use case: a lot of sampler and guider setups require something wired into a negative-conditioning slot, even on models where a negative prompt doesn't actually mean anything anymore. Rather than encoding an empty string through a text encoder (extra round trip, extra cost, and on a cloud node like this one, an extra API call), you take an existing conditioning and zero it out directly.

Where this genuinely matters: guidance-distilled models - most of what's shipping in 2026 - bake their guidance into the weights and run at CFG 1 by default, which means there's no unconditional pass for a negative prompt to steer at all. Under most samplers, the negative slot's forward pass isn't even computed at CFG 1. On those setups, this node is a cheap, correct way to satisfy a graph that structurally needs a negative conditioning input without pretending a real negative prompt is doing anything.

Where it can bite you, and this is a real, reported trap rather than a hypothetical one: some of the newer low-CFG sampler variants (the ones purpose-built for the sub-1 guidance regime some cloud catalog models are moving toward) don't treat a zeroed-out conditioning identically to a genuinely-encoded empty string. People have reported extremely grainy output specifically when a ConditioningZeroOut-style empty negative is fed into that sampler family, with the grain clearing up once they swap in a real (even if literally empty-string) encoded negative instead - an unresolved quirk, not user error, and one worth knowing about before you assume this node is a universal drop-in replacement for "encode an empty negative prompt." If your negative slot is feeding one of those samplers and you're seeing unexpected grain, that's the first thing to test - swap this node out for an actual empty-string encode and see if the grain goes away.

Inputs that matter: just conditioning, required, BIZYAIR_CONDITIONING. Nothing else to configure - the node has exactly one job.

Output: one BIZYAIR_CONDITIONING, zeroed out, ready to drop into a negative slot (most commonly) or anywhere else a conditioning input needs to exist but shouldn't meaningfully influence generation.

Install: search "BizyAir" in ComfyUI Manager and click install, git clone https://github.com/siliconflow/BizyAir.git into custom_nodes and restart, or comfy node install bizyair - one install for the whole pack. It's a cloud-executed node like everything else in BizyAir, so it needs your BizyAir account and API key set on first use, even though the operation itself is trivial.

Troubleshooting: if you're getting grainy or noisy output and you've got this node feeding a negative slot on a low-CFG or CFG++-style sampler, try the real-empty-negative swap described above before chasing anything else. Otherwise, a stall or generic failure is, as with every node in this pack, worth checking your API key over first.