AGSoft Text Replace

"Replace all" is something you expect from a text editor, not a ComfyUI node - until you're cleaning up danbooru tags or normalizing a batch of prompts and realize there's no built-in for it. AGSoft Text Replace from comfyui-AGSoft fixes that with one text input, up to three find/replace pairs, and per-pair case-sensitivity and whole-word toggles. One node replaces the pile of chained text nodes you were about to build.

How it works

Mechanically it's Python's re.sub under a friendly wrapper, and the implementation choices are smart ones:

• Your find strings are re.escaped, so special characters like ., *, or ( are treated literally - no regex surprises unless you intentionally use the features below.
• case_sensitive toggles re.IGNORECASE (default on, so it behaves like a plain search).
• whole_word wraps the pattern in \b word boundaries, so find="cat" with whole-word on matches "the cat" but not "catalog".

The three pairs - find1/replace1, find2/replace2, find3/replace3 - each with their own case_sensitive and whole_word booleans, run in order 1 → 3. Empty find fields are skipped entirely, so you can leave pair 2 and 3 blank and use this as a plain single-replace node.

What you'll actually set

Honestly, just three fields most of the time: text, find1, and replace1. The whole-word and case-sensitivity toggles matter when you're editing natural-language prompts - replacing "cat" without whole-word will happily mangle "catwoman" into "dogwoman". The second and third pairs earn their keep when you're normalizing a long tag string, like swapping three synonyms in one pass instead of chaining nodes.

The single output is a STRING, which feeds straight into whatever downstream text consumer you have - a CLIP Text Encode, a Show Text node for checking, or another replacement pass.

Where you'd reach for it

• Scrubbing a prompt: strip negative phrases, replace best quality with masterpiece, drop artist names you don't want.
• Translating a style vocabulary - map English tags to whatever the model was trained with.
• Tag cleanup on captioned datasets before a training run.

It's a small node with a boring job, which is exactly why it's worth having.

Installing it

It's part of the big comfyui-AGSoft utility pack, so install once and everything comes along:

cd ComfyUI/custom_nodes
git clone https://github.com/Art-xmaster/comfyui-AGSoft.git
# restart ComfyUI

Or search comfyui-AGSoft in ComfyUI Manager. No models, no heavy deps - this one is pure Python re, so nothing extra to install.

Gotchas

The big one is order. Replacements run sequentially, and pair 2 operates on the output of pair 1. If your patterns overlap, results depend on the order you list them - which is a feature once you know it, and a bug until you do.

Two smaller ones: whole-word matching uses \b boundaries, so it plays well with spaces and punctuation but won't help you with underscore-heavy tag strings. And because it's built on re.sub, a backslash in a replace string is treated as an escape - fine for real use, but don't go hunting for a bug in your own backslashes. That's really the whole failure surface; there isn't much else to trip on.