Combine Texts

Combine Texts is the pack's version of the classic "merge two prompts" utility, except the merging is done by Claude instead of a string-concat node. You give it two texts, it hands both to the model with an instruction, and out comes one combined_texts string - de-duplicated, coherent, and with whatever redundancy you had in the inputs quietly removed. If that last part is what you want, this is the node you'd reach for over a plain text-concatenation node.

The realistic use case is joining a character description with a style or scene block, or merging a LoRA trigger phrase into a base prompt without ending up with a run-on paragraph that repeats itself. For LLM-encoded checkpoints that read prompts as instructions, a clean merged sentence genuinely beats two pasted fragments.

How it works

The node assembles {prompt}\n{text_1_prefix} {text_1}\n{text_2_prefix} {text_2} and sends it to Anthropic's messages API with a fixed 1024-token budget. That's the whole mechanism. Those two prefix fields are the only thing separating this from Transform Text: they're just labels so Claude can tell your two inputs apart. They default to 1 and 2, and nobody's stopping you from renaming them character and style to steer the model's reading of what each block is.

Inputs and outputs that matter

• text_1 and text_2 - the two strings to combine. Both multiline.
• text_1_prefix / text_2_prefix - the labels described above. Defaults 1 and 2.
• model - nine Claude aliases, default claude-opus-4-6. This is simple text work; claude-haiku-4-5 handles it fine and won't cost you opus money.
• api_key - your Anthropic key, typed in, baked into the workflow JSON. Scrub before sharing.
• prompt - the instruction to Claude. Optional, and here's the catch: it defaults to "Describe this image in detail.", copied from the pack's image node. The author even wrote a proper default ("Combine the following two texts into one coherent prompt without redundancies") and never wired it in. So unless you type your own prompt, you'll get Claude describing two texts like they're photographs. Set it, every time.
• system_prompt - optional, your "act as" box.

The combined_texts output is a plain STRING that wires into any string input - a CLIP Text Encode, a text display node, or back through Transform Text for another pass.

How to install

Same pack as the rest, one install covers all three nodes. ComfyUI Manager: search "ComfyUI Claude". Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/tkreuziger/comfyui-claude.git
pip install -r comfyui-claude/requirements.txt

Restart, done. Dependencies are light (anthropic, pillow, comfyui-types) and there's nothing to download - the heavy lifting is an API call, not a model file.

Common issues

• Empty output. Errors are swallowed and logged to the ComfyUI console while the node returns ''. Empty string back = read the console for the auth or rate-limit error.
• The default prompt. It's the image-description text. If your "combined" result reads like alt-text, you forgot to set it.
• It's an API call, not a utility. It costs money per run, and your text goes to Anthropic's servers. For zero-cost joining, a plain text-concatenation node is still the right tool - this one earns its keep on the de-duplication and rewriting, not on the joining itself.

Honestly, this is the thinnest node in the pack: a labeled wrapper around one prompt. If you've used Transform Text, you've basically used this. Its one trick is the prefix labels letting Claude treat your two inputs as distinct things, and that trick is only as good as the prompt you remember to type.