LLM (VLM) Prompt

LLM (VLM) Prompt is the unglamorous half of this two-node pack: a multiline text box that takes what you type and outputs it unchanged as a STRING. That's it. No API, no model, no key. But it exists for a real reason - its sibling node LLM (VLM) API forces its text and system prompt inputs to be sockets, not widgets, so you can't type into them directly. This node is the pack's answer to "where do I get the string from?" It's a tiny piece of plumbing, and plumbing is exactly what makes the graph readable.

It's the same idea as ComfyUI's built-in string primitives, by the way. If you're comfortable with those, you don't strictly need this node - wire any STRING source into the API node and it'll work. What this one buys you is a guaranteed-multiline text box that sits in the same pack as the thing it feeds, so the "Install a pack and use it" workflow stays self-contained.

How it works

It's a pass-through. You type a prompt into the prompt field (multiline, so paste in a paragraph), hit run, and it returns exactly what you typed as the text output. The code is a three-liner: def execute(self, prompt): return (prompt,). No transformation, no formatting, nothing to trip over.

Where it earns its keep: wire its text output into the LLM (VLM) API node's text input for the user message, or into system prompt to set the model's role ("You are a visual analysis expert…", like the pack's README demo). Because those inputs are forced sockets, the graph requires a separate text source - and this node is the one designed for the job.

Inputs and outputs

One input, one output:

• prompt (STRING, multiline) - what you type.
• text (STRING) - the same text, as a wire you can connect anywhere.

Useful trick: because it emits a normal STRING, you can also drop one mid-graph as a reusable prompt value. ComfyUI's primitive/string nodes do the same thing, so this is a convenience more than a capability - but when you're following a workflow that references "LLM (VLM) Prompt", this is the node it means.

Install

It ships in the same pack as the API node, so installing ComfyUI-LVLM-API gives you both:

cd ComfyUI/custom_nodes
git clone https://github.com/iwindy910/ComfyUI-LVLM-API

Or search "ComfyUI-LVLM-API" in ComfyUI Manager. Then install the requirements (just the OpenAI SDK, really - numpy and Pillow come with ComfyUI) and restart:

cd ComfyUI
.\python_embeded\python.exe -m pip install -r custom_nodes\ComfyUI-LVLM-API\requirements.txt

Gotchas

Not many, because there's not much node. The one thing people trip on: if you bypass this and type directly into the API node, you can't - text and system prompt are socket-only by design. If a workflow's LLM API node seems to have a blank text field with no way to edit it, this is why; add an LLM (VLM) Prompt (or any string node) and connect it.

Otherwise it's as close to a zero-footprint node as this ecosystem has. If you installed the pack just for the LLM/VLM API node, this one is its quiet, dependable sidekick - no GPU, no network, no surprises.