AGSoft Text Operation

If you've ever chained three text nodes just to uppercase a prompt, this is the node you were looking for. AGSoft Text Operation from the comfyui-AGSoft pack stuffs ten everyday string transforms into a single dropdown, and it throws in a couple of outputs that let the result feed logic downstream. It's the text-workflow equivalent of a leatherman: you won't use every blade, but the one you need is always there.

What it actually does

Two inputs, that's it: a text field (multiline, so paste whole paragraphs) and an operation dropdown. Under the hood it's plain Python string handling - no models, no deps, nothing to break:

• uppercase / lowercase / capitalize - the obvious ones. capitalize follows Python semantics: only the first character of the whole string gets capitalized, not the first letter of each word. If you want title case, this isn't it.
• invert_case - flips every character, lower becomes upper and vice versa.
• reverse - mirrors the whole string, so "abc" → "cba". Great for making your prompt unintelligible; occasionally useful for real.
• trim - strips leading/trailing whitespace.
• remove_spaces - nukes all whitespace, including tabs and newlines, not just spaces. Handy when a model's tokenizer chokes on stray whitespace.
• count_words - counts words via a regex over non-whitespace runs and writes the number to the count output.
• split_lines - normalizes line breaks (handles \r\n too) and puts the cleaned lines into the lines output.
• join_lines - flattens multiline text into a single line, which is what you want before feeding a prompt into a CLIP encoder.

The outputs that matter

The node always returns three sockets even though not every operation uses them:

• text (STRING) - the transformed string. For count_words and split_lines it passes the original text through untouched, so downstream wiring doesn't break.
• count (INT) - only meaningful after count_words. You can wire this into a comparison node to gate a workflow, e.g. skip sampling if the prompt is under a word count.
• lines (STRING) - only populated by split_lines.

Where you'd reach for it

The boring-but-real use: clean up a prompt before it hits the encoder. Some models respond better to specific casing, and batching a folder of prompts with inconsistent whitespace/case is exactly the fiddly job this kills. Pair it with a Show Text node to eyeball the result while you iterate. Because the input field supports dynamic prompts, it slots neatly into wildcard-heavy workflows too.

Installing it

Same story as the rest of the pack - it ships inside comfyui-AGSoft, a big Russian-authored utility pack with bilingual docs (you'll see the RU/EN tooltips). Install the pack once and every AGSoft node comes with it:

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

Or search comfyui-AGSoft in ComfyUI Manager and click install. No model downloads, no heavy dependencies - this node is pure standard library, so the pack's numpy/opencv-python/translators requirements don't even apply to it.

Gotchas

Honestly, there aren't many - it's stdlib string code, so there's basically nothing to break. The two traps are both "read the docs" traps: capitalize isn't title case, and remove_spaces is more aggressive than the name implies. If you pick count_words and see the count socket sitting at zero, check for an empty string - empty text returns 0 by design. Otherwise, this is the rare node that just works.

One more thing worth knowing about the pack in general: its loader scans every .py file and imports them all, logging per-module failures without killing ComfyUI. So if an AGSoft node you expected is missing from the menu, glance at the console log - a sibling module may have failed to import while this one loaded fine.