Comfyg-Prompt

You've got five prompt ideas and you want five images, one per prompt, without rebuilding the graph five times or babysitting the queue. That's the entire pitch of Comfyg-Prompt: one node that holds a list of prompts, runs the first one, and queues the rest as separate jobs. You hit Run once and walk away.

It's a tiny, brand-new utility from Gabriel Guerreiro (guerreiro), and it has basically no community footprint yet - nobody's complaining about it because almost nobody's installed it. Don't read that as a red flag; the code is simple and it does exactly what it says. Just know you're an early adopter.

What it actually does

Open the node and you'll see a panel with numbered text boxes and a dashed "+ Add Prompt" button. Type one prompt per box. Below that sit two controls: a seed number and a seed_mode dropdown. That's the whole interface - there are no input sockets on this node at all, only the two outputs prompt (STRING) and seed (INT).

Wire prompt into your CLIP Text Encode's text input like any prompt string, and seed into your sampler's seed (convert the KSampler seed widget to an input if it's not already). Then run the workflow once. The node executes prompt 1 immediately, and if more prompts are waiting, it queues a copy of the whole workflow with its internal index bumped. That queued copy runs, queues the next one, and so on - a chain that drains your prompt list, one job each.

How the self-queueing works

This is the interesting bit, and it's worth knowing because it sets expectations. When the node needs to queue the next job, it deep-copies the API prompt it was given, increments its hidden _index counter, and POSTs the copy back to ComfyUI's own HTTP API at http://localhost:8188/prompt. The requests library is the pack's only dependency, and it's used for exactly that.

Your original node's _index never changes - only the queued copies' do - so re-running the workflow starts the chain over from prompt 1. The seed modes map directly: fixed uses the same seed for every job, increment adds the job index, decrement subtracts it (floored at 0), and random rolls a fresh value per job.

Where you'll trip

• It hardcodes localhost:8188. If your ComfyUI server runs on a different port - or you're doing the remote-server-with-tunnel thing - the queued jobs fail silently. The errors only show up in the console, so your first image works and the rest never arrive. Keep the default port or be ready to patch the URL.
• One run is the whole list. Hit Run five times and you'll get five chained jobs per prompt - 25 images. The node has no "just do the first one" mode.
• Random seeds come from Python's random, not from ComfyUI's sampler noise. That's fine - you're feeding an INT into the sampler's seed input either way - but don't expect the seed output to match whatever the KSampler would've rolled on its own.
• It's not a batch. Each prompt is a separate job with its own seed, its own output image. That's the feature if you're A/B testing prompts or farming variations; it's the wrong tool if you wanted all prompts in one image.

Install

Via ComfyUI Manager, search Comfyg-Prompt and install, then restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/guerreiro/comfyg-prompt

The README tells you to pip install requests - and it's true that requests is the only dependency, but every modern ComfyUI install already ships it, so you can almost certainly skip that step. No models, no checkpoints, no heavy anything. If you've been staring at a queue of prompts you keep re-typing into the same workflow, this is the closest thing to a free win.