That was covered by the court's generation of more images using the same model and prompt. A poem is an entire work. It is the art work. If you type the poem twice you get the same poem:
See?
But if I write the same prompt into, say, Perplexity twice watch what happens: "Write me a jueju in Classical Chinese that criticises people who use AI slop and call it art. Make sure that the contrast point is in place. Then translate that jueju into English, focusing more on the poetic qualities than literal translation."
First run:
## 绝句：讥AI伪艺
画屏新巧出机声，
点墨无魂似鬼行。
人手一挥真意在，
虚光何敢唤丹青。
### English Poetic Translation
On screens, new wonders birthed by code appear,
Yet soulless strokes drift ghostlike, insincere.
A human hand—one gesture, spirit shown;
How dare mere light call itself art alone?
Second run:
**绝句**
画意千年笔下生，
今朝机器乱涂成。
人心未动空称艺，
一片虚声满世鸣。
**English Translation**
For ages, art was born from living hands—
Now, soulless engines smear the canvas dry.
No heart has stirred, yet “art” is what they claim;
Empty applause, a hollow world’s reply.
There's barely any resemblance between the two poems, despite the prompt being identical. And here's the thing: that prompt is approximately the same size as the poem. Could you get a very precise poem output that's a direct match for exactly what you intend to communicate? Maybe. (I personally doubt it, but I'm not going to come out and say it's not possible.) If you can, though, create a prompt that generates poems that are all very close to each other I suspect it's a lot more work than, you know, just writing a fucking poem would be.
Personally I think this is a really good acid test that the judges devised to decide how much "human input" there was in the work to see if it qualifies as "art". Kudos to them!
