Concat (cond)

Concat (cond) is the node that answers "my prompt is too long for one CLIP encode." It takes two CONDITIONING objects - to and from - and splices the token embeddings of from onto the end of to, giving you one CONDITIONING that's effectively the sum of both prompts. Same trick as core ComfyUI's ConditioningConcat, reimplemented in this pack.

The mechanism, from the source: for each conditioning entry in to, it takes the embedding tensor and runs torch.cat along the token dimension, appending from's tokens. On a CLIP model that means from gets its own 77-token chunk instead of being truncated at the boundary - it's the "BREAK but automatic" move, where the KB's prompt-engineering essay describes BREAK as forcing a split at the chunk boundary so one concept doesn't bleed into another. Encode both halves separately, then splice: that's the entire point.

There's one guardrail baked in: if from contains more than one conditioning entry, the node raises a warning that only the first will actually be applied. That mirrors core ComfyUI's own caution, and it's the failure mode you'll hit if you feed it the output of a node that emits multiple conditionings.

The inputs that matter

• to - the base conditioning; the splice target
• from - the conditioning whose tokens get appended

One output: CONDITIONING, wire it into your sampler's positive (or negative) input.

Where beginners get burned

The two inputs have to be compatible, and "compatible" here means encoded by the same CLIP. Both halves should come from the same checkpoint's text encoder (or at least the same CLIP family) - splicing a T5-encoded Flux chunk onto an SDXL CLIP chunk is asking for nonsense. And like everything in this family, it's a CLIP-era tool: on Flux and other LLM-encoded models, long prompts are handled natively and chunk-splicing is the wrong tool. If you're on SDXL and your prompt keeps getting its tail chopped, this is your fix.

Installing it

No dependencies beyond ComfyUI's own, no models to fetch. ComfyUI Manager → search ComfyUI-Prompt-Helper, or:

cd ComfyUI/custom_nodes
git clone https://github.com/elypha/ComfyUI-Prompt-Helper

Restart, and it's under prompt_helper.

Fair warning: it's a niche tool - you'll reach for it specifically when a long CLIP prompt is truncating. When that happens, it's precisely what you needed.