String Function

The name says "String Function," but what you actually get from this node is a prompt-tag combiner and a find-and-replace tool bolted into one box. It ships inside the ComfyUI Deploy pack (the 🔗ComfyDeploy category), and its real job is letting a deployed workflow stitch together a prompt from a few pieces - or scrub one clean - before it hits the CLIP encoder.

Why it exists makes sense once you've deployed a workflow. When you hand a txt2img API to other people, they want to send you a prompt string, not rebuild your graph. String Function is how you let a caller's tags merge with the baked-in style tags your workflow needs, with tidy_tags cleaning up the mess afterwards. You'll usually find it sitting right before a CLIPTextEncode in deployed SDXL or Flux pipelines.

How it works

The mechanism is simpler than the two-hat framing suggests. Set action to append and it joins text_a, text_b, and text_c with ", " (or nothing, if tidy_tags is off). Set it to replace and it does literal find-and-replace on text_a - unless text_b starts and ends with /, in which case it treats it as a regex and substitutes matches with text_c.

The tidy_tags toggle is the sleeper feature. It collapses double spaces, strips stray commas, kills duplicated commas, and trims whitespace. For tag-heavy anime workflows this is genuinely the difference between cat, , hat, blue,, sky and cat, hat, blue, sky - the kind of cleanup that quietly improves your sampling, because a junk-padded prompt is a drift-padded prompt.

The inputs that matter

• action - append or replace. Required, and it decides which half of this node you get.
• tidy_tags - yes or no. Leave it on.
• text_a, text_b, text_c - all optional multiline strings. In append mode they're the pieces; in replace mode text_a is the haystack, text_b is the needle (or /regex/), text_c is the replacement.

Output is a single STRING. Despite being marked as an output node in the schema, it still returns a result value, so you can wire it straight into a CLIPTextEncode or a text-prompt node.

Installing it

It's part of ComfyUI Deploy, so the install is the pack install. Either search "ComfyUI Deploy" in ComfyUI Manager, or do it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/BennyKok/comfyui-deploy.git

Restart ComfyUI after cloning. The pack pulls in a few pip deps (aiofiles, pydantic, opencv-python, imageio-ffmpeg, brotli, tabulate) - opencv is the chunky one, so first load can take a minute. No model files, no API key, no extra downloads. It works fine as a plain string utility even if you never touch the ComfyDeploy dashboard.

Where people get burned

Nothing dramatic here, just one trap: if you leave text_a, text_b, or text_c empty in append mode, the empty strings are filtered out, which is usually what you want - but in replace mode an empty text_b turns into .replace("", ...) behavior that inserts text_c everywhere. Weird, but predictable. And remember this node is the string prep stage, not the prompt node itself - people reach for it expecting a prompt-weight control and find only text out. It outputs plain text; the weighting happens downstream in your text encoder.