Text Replace

Find-and-replace is the most taken-for-granted text operation in existence, and in ComfyUI it's usually done with regex nodes that silently do nothing when your pattern doesn't match. VLMTextReplace does the operation properly: literal or regex substitution, case sensitivity control, a hard cap on replacements, and - the part that saves real debugging time - explicit behavior when the search string isn't found.

You reach for it whenever a model's output needs a surgical fix: strip a prefix the model keeps prepending, rename a label across a report, normalize a value, scrub a specific token before it reaches an API or a downstream parse. The pack's text toolkit is dependency-free, so this is a plain-Python find-and-replace that works anywhere.

Inputs that matter

• text - what you're editing.
• search - the literal string or regex to find.
• replacement - what replaces it (regex capture groups work in regex mode, \1 and friends).
• mode - Literal or Regular expression. Literal is the safe default; switch to regex when you need patterns.
• case_sensitive - on by default. Turn off for "replace regardless of case."
• max_replacements - caps how many matches get replaced; 0 replaces every match.
• if_not_found - Keep text (default) or Error. This is the underrated one: Error turns "my pattern silently matched nothing" into an immediate, loud failure instead of a wrong-looking downstream result.

Outputs

• text - the result.
• replacement_count - exactly how many replacements happened. Wire this into a conditional or use it to confirm the operation did what you expected.
• diagnostics_json - the operation's details for inspection.

The replacement_count output is the quiet superpower. "Did my substitution actually fire?" is a question every automation has, and most find-and-replace tools make you eyeball the output. Here it's a number you can check programmatically.

Install

Part of ComfyUI VLM Nodes (gokayfem/ComfyUI_VLM_nodes). ComfyUI Manager → search "VLM Nodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/gokayfem/ComfyUI_VLM_nodes
python -m pip install -r ComfyUI/custom_nodes/ComfyUI_VLM_nodes/requirements.txt

Run with ComfyUI's Python; the repo won't touch torch. No models, no downloads.

Gotchas

Regex mode is Python regex - if your workflow expects JavaScript or shell semantics, escape patterns will differ. And a subtle one: in Literal mode, the search is a literal string, so characters like * or ( are treated as plain text. That trips people coming from regex-first tools more than anything else here. When in doubt, check mode before wondering why nothing matched.