☁️BizyAir Conditioning (Concat)

Three conditioning nodes in this pack look similar on the surface - Average, Combine, Concat - and each one does something genuinely different, so it's worth being clear about which lever you're actually pulling. Average blends two conditionings by weight into one. Combine keeps two conditionings fully separate and hands both to the sampler as parallel entries. Concat does neither - it stitches the underlying embedding tensors of two conditionings end-to-end, extending the effective token sequence, the way you'd tape two sentences together rather than average them or run them as two independent thoughts.

What that's actually useful for: appending. If you've got a conditioning_to that already represents your main prompt, and you want to tack on extra descriptive text - think "and also, painted in the style of Moebius" - without re-encoding the whole combined string from scratch, Concat lets you encode the addition separately and splice it onto the end. It's also a workaround for text encoders with a tight practical prompt length: rather than hoping one giant string encodes cleanly, you build it in pieces and concatenate the pieces' embeddings after the fact.

Inputs that matter: conditioning_to and conditioning_from, both required, both BIZYAIR_CONDITIONING - the naming mirrors ConditioningAverage's inputs, but don't let that fool you into expecting a strength dial here. There isn't one. conditioning_from gets appended onto conditioning_to at full weight, every time; this node has exactly one job and no parameters to tune.

Output: one BIZYAIR_CONDITIONING, the concatenated result, ready to wire into a sampler slot or another conditioning node downstream - same as everything else in this family, as long as the chain stays BizyAir end to end.

A caveat worth having going in: concatenation is a token-sequence-level operation, and how well it behaves depends on what the underlying text encoder actually does with a longer sequence. On classic CLIP-based models this is fairly well-trodden ground. On the newer LLM-encoded architectures that read your prompt more like an instruction than a token bag, splicing two independently-encoded chunks together doesn't necessarily read the same way a single coherent instruction would - the model wasn't trained on "here's a sentence, then here's an unrelated second sentence bolted onto the same conditioning." Worth testing on whatever base model you're actually using through BizyAir's catalog rather than assuming it behaves identically everywhere.

Install: search "BizyAir" in ComfyUI Manager and hit install, or git clone https://github.com/siliconflow/BizyAir.git into custom_nodes and restart, or comfy node install bizyair - one pack install brings in this node with the rest. It's a cloud operation like everything else here, so register a BizyAir account and set your API key before expecting anything to run; the README calls this out explicitly as a first-use requirement.

Troubleshooting: if the concatenated conditioning doesn't seem to be picking up the second half of your intended prompt, double-check that both inputs are actually wired - it's easy to leave conditioning_from connected to the wrong upstream node when you're juggling several conditioning branches in a busy graph. Beyond that, treat any stall or unclear failure the same as anywhere else in this pack: confirm your API key is set before assuming the node logic itself is at fault.