Concat Conditionings

Concat Conditionings takes several encoded prompts and joins them into one, end to end. Not blended, not averaged - concatenated, the way you'd stick two strings together. If you've got conditioning coming from three different CLIP encodes and you want the model to see all of it as a single prompt, this is the node that stitches them.

The subtlety worth getting right up front: this is concat, not combine, and Impact Pack ships both. Concat glues the conditioning tensors into one longer sequence, so the model reads them as a continuous prompt. Combine (the sibling node) keeps them as separate conditionings that get sampled together and averaged in effect. They sound similar and behave differently, and picking the wrong one is the usual source of "why does my prompt feel mushy." If you want the model to attend to a long, extended description as one thing, concat. If you want to mix two independent prompt intents, combine.

Why you'd want this

The everyday reason is prompt length and structure. Some setups want you to build a prompt in pieces - a base description here, a style block there, a quality tail somewhere else - each encoded separately, then assembled. Concat Conditionings lets you keep those pieces as distinct nodes on the canvas (easier to toggle, reorder, or reuse) while still feeding the sampler one unified conditioning. It's plumbing for people who like their prompts modular.

It also shows up when a workflow merges conditioning from different sources that need to read as one continuous instruction rather than as competing signals. Concatenation preserves all of it in sequence instead of diluting any one part.

The inputs

Here's where the info tells a small story. The node's schema shows a single required input, conditioning1 (type CONDITIONING) - but this is one of ComfyUI's growing-slot nodes. Connect something to the first conditioning input and another slot appears; connect to that and you get another, and so on. So "one input" really means "one or more, added as you wire them." You feed in as many conditionings as you have, and they're concatenated in order.

The output is a single CONDITIONING, the joined result, which plugs straight into a KSampler's positive or negative input, or into any node downstream that eats conditioning.

Order matters, because concatenation is sequential - the pieces are joined in the order of the slots. If a result feels like it's weighting the wrong part, try reordering which conditioning feeds which slot.

Common issues

The big one, again, is choosing concat when you meant combine or vice versa. If your assembled prompt is behaving like an over-mixed average instead of a clear extended description, you probably wanted concat and used combine, or the reverse. Swap the node and re-judge.

Beyond that, this is a lightweight utility - no models, no GPU cost beyond the trivial tensor op, no real failure modes. One thing to keep in mind on modern architectures: some newer models (FLUX.1 and friends) don't use negative conditioning the normal way, so how much prompt-assembly gymnastics helps depends on the model. On classic SD/SDXL-style stacks where positive and negative conditioning both do work, concat is a clean way to build up a long positive prompt from parts.

Installing it

Concat Conditionings comes with ComfyUI Impact Pack; you don't install it on its own. Easiest: open ComfyUI-Manager, search ComfyUI Impact Pack, click Install, restart. By hand: cd ComfyUI/custom_nodes && git clone https://github.com/ltdrdata/ComfyUI-Impact-Pack, then install the pack's requirements.txt in ComfyUI's Python environment (portable: ..\..\..\python_embeded\python.exe -m pip install -r requirements.txt), and restart. Auto-install went away in v7.6, so a manual clone without the requirements step just won't load. The pack is ltdrdata's - same author as ComfyUI-Manager and the Inspire Pack - which is why it's one of the most trusted things in the ecosystem. This node in particular is pure utility: no downloads, nothing heavy, works the moment the pack loads.