Random Prompter JK🐉

Random Prompter JK is a categorized random-prompt generator with a truly absurd amount of content behind it - thousands of scene options, over thirteen thousand style entries (artists, painters, forms, genres), hundreds of expressions, facial actions, lighting and camera descriptions. The point isn't to make you pick from 13,000 artists. The point is that when you want variety, it picks for you - from a database that doesn't repeat, because it's that big.

How it works

Every category (scene, motion, facial_action, expression, audio, lighting, camera, style, description) has a dropdown with the same five-ish states: disable (exclude it), enable (use your custom value), random (surprise me), or a specific entry. Under the hood it reads from the pack's prompt_data folder - plain text files per category - and randomly assembles a prompt from whatever you've enabled. auto_fill (default on) writes the result into the editable fields so you can see and tweak it, and remove_prompt_emphasis strips stray (word:1.5) weight syntax from your subject so it doesn't leak into the final prompt.

The three inputs a beginner actually sets:

• custom_subject - the anchor. Put your main subject here (or pipe in a Portrait Master output, which the pack's own workflows do). Everything else dresses it up.
• prompt_priority - the ordering of the assembled prompt (subject + scene, description + style, etc.). This genuinely changes how a model reads the prompt, so it's worth playing with.
• exp_str - an expression-intensity modifier ("slightly," "quite," "hardly") that pairs with the expression category.

The seed input controls the randomness; 0 means "new random seed every run."

The output

One STRING, prompt, ready to feed a CLIP Text Encode. You'll notice the dropdowns also list use image 1/2/3 entries, but this node has no image input of its own - those options are plumbing for the pack's larger auto-prompt workflows where an image reference feeds the prompt generation. On this standalone node, the categories and your custom values are what do the work.

Where it shines (and where it doesn't)

It shines for character/portrait variation - same subject, endless scene/lighting/camera combos, which is exactly how Jake uses it in the Portrait Master + Random Prompter imgen workflow. It's less useful if you want a consistent style: random selection is random, so for reproducible batches you're better off locking the categories you care about and only randomizing the ones you don't. And 13,000 styles means you should curate - drop your own text files into prompt_data if you want to narrow the pool. The config (in config.ini, [RandomPrompterConfig]) controls which folders it reads and the probability knobs for how often it uses custom values versus empty.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/jakechai/ComfyUI-JakeUpgrade
pip install -r requirements.txt

Restart and it's under the pack's Prompt section. It shares its slot with a sibling: the pack's config.ini has a RANDOM_PROMPTER_ABC flag that swaps between this standard version and the ABC-strategy implementation - both named Random Prompter. If the node behaves differently than a tutorial describes, that flag is why.