AI4ArtsEd Random Artform Generator

AI4ArtsEd Random Artform Generator hands you four reinterpretation instructions drawn from a pool of roughly fifty art forms and cultural traditions, and it's the node in this pack with the clearest personality. One run might tell you to rewrite your prompt "in the style of Japanese Noh theatre," the next as "a Yoruba praise poem," the next "as instructions from a Bauhaus design manual." The list deliberately reaches beyond the usual western canon - Capoeira song, Sámi joik, Māori moko tattoo design, Quechua myth, shadow-puppet theatre from Southeast Asia - which is exactly the "cultural and stylistic translation" the AI4ArtsEd project is about.

It's a sibling of ai4artsed_random_instruction_generator (which does the same job with more literary/absurdist transformations). The difference is flavor: this one is the art-history sampling, that one is the surrealism sampler. Wire either into an LLM node's style_prompt and the LLM rewrites your text "as if" through that lens.

How it works

There's one input - random_seed, an INT - and the code is honest about it being decorative. The comment in the source calls it a "legacy input" that exists only so the node has something to connect, and the actual randomness is time-based (time.time_ns() seeds the RNG on every call). The class also sets ALWAYS_EXECUTE = True, which means ComfyUI re-runs it every queue pass regardless of whether inputs changed. So you get a genuinely fresh four-pick each time, unlike the other text nodes that depend on ComfyUI's change-detection.

Outputs are artform_1 through artform_4 - four STRINGs, guaranteed distinct, random order.

Wiring it in

The intended chain: Random Artform Generator → ai4artsed_prompt_interception (style_prompt) or another LLM node → the rewritten prompt → an encoder. Or feed all four into ai4artsed_text_remix in random mode so each run silently picks a different artform. Either way, you get prompt-variety for education and experimentation without writing a thing.

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 beyond Python's random and time, no models, no key.

Common issues

Outputs repeat across runs. With ~50 entries and sampling four, repeats are statistically inevitable - especially if you run it a lot. Not a bug, just small-pool sampling. There's no way to pin a seed for reproducibility (the seed input is ignored), so this is a "variety" node, not a "reproducible" node.

The seed input does nothing. That's by design, per the source, but it catches people who connect a randomizer expecting to control the draw. It's there for graph compatibility.

It's not a generator in the image sense. It produces instructions, not prompts - a step before an LLM does the actual rewriting. Expect to see the instruction verbatim in the LLM's output prompt, which you can then trim.

For what it is - a cheap, culturally-aware source of prompt-variety - it works fine. Just don't expect fine-grained control or reproducibility; that's not the job it was built for.