Show Text

ComfyUI graphs are notorious for hiding what's actually flowing down a wire. You build a wildcard picker, a captioning node, a prompt-combining chain - and unless something displays the string on-canvas, you're guessing whether it did what you think. Show Text is the plain, no-frills answer: feed it a string and it displays that text right on the node in the graph, so you can eyeball intermediate output without routing it anywhere else.

It's categorized under GR85/Logging for a reason - this isn't meant to feed the next stage of your pipeline, it's a debugging tap you drop mid-graph. It's marked as an output node, meaning ComfyUI treats it as a terminal point that's allowed to run and display without anything wired downstream of it.

The interface is about as simple as it gets. One required input, text (a STRING), and one output that mirrors it back out as a list of strings. That output existing at all is what makes this genuinely useful rather than just a dead end: you can splice it into the middle of a chain - prompt builder → Show Text → sampler - see what the prompt looked like at that point, and keep the graph running exactly as before. It doesn't consume the string, it just lets you look at it on the way through.

Installing it. Search "comfyui_gr85" in ComfyUI Manager, or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/veighnsche/comfyui_gr85

Restart ComfyUI. No models, no extra dependencies - it's a display node.

Where this actually earns its keep. Wildcard expansion is the classic case: you write a {red|blue|green} car into a wildcard node, and without something showing you the resolved string, you have no idea whether it picked "red car" or silently broke on a syntax typo. Drop a Show Text right after any node whose output is a string you can't otherwise see - an IntToString cast, a Florence-2 caption, a combined tag injector - and you turn an opaque pipeline into one you can actually debug. If you're chasing "why does my image not match my prompt," this is usually the fastest way to rule out "the prompt itself was never what I thought it was" before you go blaming the sampler or the model.

It's a cheap habit worth building generally: any time you add a node whose main job is transforming text - splitting, combining, casting, filtering - a Show Text tapped off its output costs you nothing and saves you from debugging blind. It's the string-graph equivalent of a print() statement, and just as disposable once you've confirmed the thing upstream is behaving.