Prompt Stash Passthrough

Text in, text out, and while it's in between you get to grab it. That's the entire job of Prompt Stash Passthrough, and it's the fix for the most annoying failure mode of LLM-assisted ComfyUI workflows: the LLM node spits out a fantastic prompt, it goes straight into the encode node, and the generation is gone forever because nothing kept the text. This node sits in that gap and hands the prompt back to you.

The pack it ships in - phazei's Prompt Stash - is basically a pocket prompt library for ComfyUI. The Passthrough is the leaner of the two text nodes: no saving, no lists, just a way to watch, edit, and keep whatever text is flowing through your graph. Think of it as a tap on the pipe rather than a reservoir.

How it works

The node has a prompt_text field you can type into directly, plus a text input port. The use_input_text toggle decides which one wins. Flip it off and your typed text passes through untouched; flip it on and the incoming text takes over and the widget syncs to show you what came in. That input is lazy - ComfyUI won't even evaluate it unless the toggle is on, so you don't force your upstream LLM to re-run every time the graph executes.

The headline feature is pause_to_edit. Turn it on and execution halts at this node, the frontend pops a Continue button, and you get to rewrite the prompt live before the run resumes. Behind the scenes the server waits in a short polling loop, then reads your edited text back and - here's the clever bit - writes it into the saved workflow metadata. When you save the workflow and reopen it, the captured text is baked into the node, not lost like a transient connection value.

The only output is text (STRING). Wire it to whatever consumes a string - a CLIPTextEncode, an LLM node, or another text transformer.

The inputs that matter

• use_input_text - switch between your typed prompt and the incoming connection.
• text - the input port; only computed when use_input_text is on.
• pause_to_edit - halt the run so you can fix the prompt mid-flight.

That's genuinely it. This is a deliberately tiny node, and it's better for being tiny.

Installing it

Same routine as every ComfyUI custom node. Easiest path is ComfyUI Manager: search "Prompt Stash" and install. Or go manual:

cd ComfyUI/custom_nodes
git clone https://github.com/phazei/ComfyUI-Prompt-Stash

Then restart ComfyUI. The pack has no model downloads and no pip dependencies beyond ComfyUI's own aiohttp server - it's pure Python plus a couple of JS files.

Where people get burned

The pause feature is the one that bites. If you close a workflow while a run is paused, the node can stay "paused" server-side and stall later runs. The fix is built in: the pack's sibling Prompt Stash Manager node has a Clear All Paused button that resets every stuck pause state. Get into the habit of knowing it exists before you need it.

Also, lower your expectations about magic: like the rest of the pack, this node passes text through as-is. It will not evaluate {red|blue|green} wildcard syntax - that's a feature of dedicated wildcard node packs, and the author has confirmed it's not on the list here. If your prompts rely on dynamic wildcards, this isn't the node doing the substitution for you.

Install it when you start running local LLMs for prompt crafting. The first time a really good generation comes out of a Qwen or a Llama node and you actually keep the prompt that produced it, you'll get why this exists.