AI4ArtsEd Random Instruction Generator

This node doesn't generate anything. It's a randomizer that hands you four creative instructions - prompts that tell an LLM how to transform some other text. "Write this as a fable with animals," "translate this into the dialect of a future underwater civilization," "speak this message as if through the fog of a fever dream." You get four of these from a pool of about fifty, sampled fresh on every run.

It sits at the front of a chain: Random Instruction Generator → an LLM node like ai4artsed_prompt_interception → your image prompt. The instruction becomes the "how to rewrite it" part, and the actual subject text goes in elsewhere. That's the whole point of the AI4ArtsEd approach - instead of you hand-writing one transformation, the node picks a weird lens and the LLM does the work.

What it actually does

The implementation is honest about being simple: it holds a hardcoded list of transformation phrases and uses random.sample() to return four distinct ones. The pool leans literary and surreal - Wittgenstein-style philosophy, "the internal monologue of a tree," "a recipe for emotional resilience," "a eulogy for a lost language" - so the results are more art-school than photography-studio. If you want variety for variety's sake in a cultural-education workflow, that's the vibe.

Inputs and outputs:

• Inputs: none. The README's mention of a text input is stale - the shipped code takes nothing.
• Outputs: instruction_1 through instruction_4, four STRINGs, guaranteed distinct.

There's no seed input and no IS_CHANGED flag, so the sample happens every time ComfyUI executes the node - which is every time any upstream input changes. Since it has no inputs, a fresh run needs some change elsewhere in the graph to trigger it.

Wiring it in

The natural place for these outputs is ai4artsed_prompt_interception's style_prompt field (the "task" part of the rewrite), or ai4artsed_clean_prompt_randomizer's context. Wire one of the four outputs in and leave the others hanging, or feed them to ai4artsed_text_remix in random mode so each run picks a different one.

Installing it

Standard for this pack:

cd ComfyUI/custom_nodes
git clone https://github.com/joeriben/ai4artsed_comfyui_nodes
# restart ComfyUI

Or search "AI4ArtsEd" in ComfyUI Manager. No dependencies, no models, no key.

Common issues

The same instructions keep coming back. The pool is ~50 entries and the sample is random, but you can get repeats across runs - that's just how random sampling works, not a cache bug. There's no seed to pin.

You expected a specific input. It has none. The whole "random instruction" job is stateless: feed the outputs somewhere downstream and let the LLM node worry about the subject.

It's the kind of node that looks pointless until you've spent an hour hand-writing five prompt variants and wish you could get fifty for free. For that, it's fine. For a tighter, more controllable workflow, you'll probably swap in your own lists - but as a source of prompt-variety noise for education and experimentation, it does the job.