🫙 LLM Submit Input

Chatting with an LLM inside ComfyUI has a UX problem: the graph's normal "hit Queue" runs the whole workflow, which is not the same as "send this one message to the AI." LLMSubmitInput is the pack's answer - a multi-line input with a real Submit Prompt button on it, so dispatching a question feels like a chat app instead of a chore.

It's built to sit in front of the Gemini Persistent Chat node, and it's smarter than it looks at first glance.

How the trigger trick works

The clever bit is the trigger output and how it's meant to be wired. Connect LLMSubmitInput's trigger output to the chat node's enable_ai_processing input. Now two things can happen:

• You press Submit Prompt on the node. A JavaScript handler sets a hidden state flag and fires the ComfyUI queue. The workflow runs and the LLM actually gets your message.
• You hit Queue from the sidebar (or any other path that doesn't go through the button). The workflow runs fine - image generation still works - but the chat node sees AI processing disabled and just echoes back the last response. Your prompt is never sent.

That's the "selective execution" behavior in the README: global queue runs render the workflow, but only the button dispatches to the model. If you've ever hit run and watched a stray prompt burn API credits, you'll see the appeal immediately.

The inputs that matter

• text (multiline STRING) - your message.
• auto_clear (default on) - wipes the box after a successful submit, same behavior as Clearable Text Input. Turn it off if you're iterating on the same prompt.
• trigger_state - a hidden widget the button flips. You don't set it by hand; it's there so the node can report "triggered" to the UI.

Outputs are text (STRING, wired to the chat's user_prompt) and trigger (BOOLEAN, wired to enable_ai_processing).

Install and setup

Standard MoonNodes install - Manager search for ComfyUI-MoonNodes, or:

cd ComfyUI/custom_nodes
git clone https://github.com/m0rtus59/ComfyUI-MoonNodes.git
cd ComfyUI-MoonNodes && pip install -r requirements.txt

Restart, and the Submit button appears on the node once the pack's JavaScript loads.

Where people get burned

The trigger wiring is the whole point, so if you skip it - if you just connect text and run from the sidebar - the chat node will sit in its paused state and you'll see "⚠️ AI Processing is PAUSED" instead of an answer. That's not a bug; you forgot the other wire. It's the one setup step the README stresses, and it's the first thing to check when the LLM goes quiet.