Regex Match To Float

A switch, not a parser

Regex Match To Float doesn't pull a number out of your text. It asks one yes/no question - does this regex pattern appear in this string? - and then hands back one of two floats you typed in. Match means output_on_match, no match means output_on_non_match. That's the whole node, and honestly that's fine.

It ships in db0's personal utility pack, so it's small, does one thing, and makes a couple of opinionated calls you should know about. (Same db0 who runs AI Horde and wrote the Civitai Helper extension - a real name in the scene, but this is his "custom nodes for my purposes" drawer, not a polished library.)

Why you'd bother

ComfyUI workflows are programs, and programs need branches. Most switch nodes route on numbers or booleans, but the decision in your head usually lives in a string: the prompt, a tag list, the output of an LLM node, a filename. Regex Match To Float is how you turn that text into a knob.

Reachable uses:

• Prompt mentions "night" → drop CFG to 4 or bump denoise to 0.6
• A tag list contains 1girl → feed a different LoRA strength
• Whatever your captioning/LLM node returned contains a marker → use a different upscale factor

Classic "wire a text decision into the sampler" glue.

How it works

Under the hood it's one re.search() call from Python's stdlib - no dependencies, no model files. The three optional toggles map straight onto regex flags: case_insensitive sets IGNORECASE, multiline sets MULTILINE, dotall makes . match newlines too.

The one that will trip you up: case_insensitive defaults to True. Most regex tools are case-sensitive by default; this one isn't. If "1Girl" shouldn't match 1girl, open the advanced inputs and switch it off.

The inputs that matter

• string (multiline) - the text being tested. A prompt, a tag list, anything.
• regex_pattern (multiline) - the pattern to search for. Remember, this is a substring search: pattern fox matches the quick brown fox. It is not a full-string match. If you want the whole string, anchor it: ^fox$.
• output_on_match / output_on_non_match - the two FLOAT values it can return. These are just pass-through values; it never extracts or transforms anything from the string.
• The three advanced booleans - set them once and forget them.

Output: a single FLOAT, whatever value matched the branch. Wire it into any float input in your graph - CFG, denoise, a LoRA strength, an upscale factor, a KSampler step count. Because it's just a number, it composes with everything.

Installing it

ComfyUI Manager → Install Custom Nodes → search "comfyui-db0-pack" → install, restart. Or the manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/db0/comfyui-db0-pack

Restart ComfyUI. No extra Python packages, no model downloads - it's pure stdlib.

Gotchas

• Needs a recent ComfyUI. The pack is written against the newer declarative node API (comfy_api), not the classic decorator style. If it fails to import, update ComfyUI first - older installs can't load it.
• Bad regex = hard error. There's no try/except around re.search, so a malformed pattern throws re.error and kills the workflow. Double-check your pattern, or test it in a scratch text editor first.
• Case-insensitive by default (see above). It's the trap that wastes the most time.
• ^ and $ change meaning if you flip multiline on - they start matching at every line boundary instead of the whole string's start/end.

It's a tiny node, so if you only need one comparison, a String-to-boolean conditional might do. But the moment you want to steer numeric settings off a string, this is the direct path.