llama.cpp Prompt Output

Every LLM workflow ends in the same problem: the text came out, but where do you look at it? ComfyUI is built to display images, not strings, so without a display node you're staring at a text output only through the node inspector. Prompt Output is the pack's answer - a terminal node that shows the LLM's text right on the canvas and passes it through so the workflow can still use it.

It's deliberately boring, and that's the compliment. One required input, one optional toggle, one output.

What it does

Wire any STRING into text and the node displays it and forwards it unchanged through the text output. The only knob is plaintext, which converts common Markdown and HTML markup into plain text before display. Turn it on and you get the raw content without the # headings, **bold**, `code` fences, and link syntax - which is usually what you want when the text is going to be pasted into a prompt or a file.

A useful detail: the conversion is conservative and applied only to common markup patterns (image/links to their text, code fences stripped, headers un-hashed), so you don't lose anything weird.

Wiring it

The minimal quick-start from the README is Start llama.cpp Server → Basic Prompt → this node. Connect the prompt node's response to text. Since the node passes text through, you can also put it in the middle of a chain - display the intermediate output of one LLM step while still feeding it forward to another node - though the straightforward use is as the end of the line, a way to actually read what the model produced.

The plaintext toggle is the thing to remember. If you're using a template like Prompt Enhancer whose system prompt says "do not use markdown formatting," you may not need it; if you're running a raw chat model that insists on formatting, it's what makes the output usable downstream.

Where it fits

In the wider picture this node is the display half of the LLM-in-the-graph pattern: a captioner or enhancer produces text, and Prompt Output is how you check that text before it feeds your encoder. It pairs naturally with the suite's other output-friendly nodes - Model Info, List Models, Token Count all emit strings, and this is the cheap way to actually see them. There's nothing to install separately (it's one of the 17 nodes the pack registers) and no settings that can hurt you. For a node this simple, the only "issue" is forgetting it exists and clicking around the workflow JSON to find your text - don't.