9 Comments
User's avatar
Linda's avatar

Yes, the castle here is a nice touch. Perfect

Expand full comment
John E Simpson's avatar

It's always a mistake (for now!) to ascribe intentionality to AI-generated works. But if this were by a human artist -- or simply assumed to be by one, for that matter -- I'd really have to wonder what the artist meant by including the castle. Does it suggest the start of a journey, or the end, or simply a waystation? In the context of this "maxim" (which the AI didn't know about), is there just for the benefit of the boatman's passenger -- or is there regardless of who the passenger is?

Mysteries, mysteries...! 🤩

Expand full comment
John E Simpson's avatar

P.S. typos: "is there" = "is it there" [twice]!]

Expand full comment
Linda's avatar

It is a mystery! Perhaps something no longer obtainable. I do think it was for that passenger.

Expand full comment
Linda's avatar

I haven’t tried AI. But the image produced reminds me of the time artists incorporated real or imagined castles, or ruins into their paintings to give their landscapes more relevance or status.

Expand full comment
John E Simpson's avatar

Aha! So maybe there *was* something like intentionality at work. "Hmm, I've been prompted to produce an image of a river scene in the style of a Durer etching... Perhaps something like this?"

https://images.app.goo.gl/Zn1PA9iZshRsrJW9A

"...or this (although by Rembrandt)?"

https://images.app.goo.gl/AE7LUM4LtcHQgVLt5

Expand full comment
John E Simpson's avatar

BTW, I've been enjoying playing around with these AI tools. I have some photographic skills, in-camera and after-the-fact, but insufficient access to many subjects I'd like to photograph. The AI stuff is a workaround for now!

Expand full comment
Michael Simpson's avatar

I like this… a lot. It’s also the first time anything has made me curious about trying AI at all. In this case I was wondering if it’s possible to enter the entire text of the writing into the “prompt” in addition to the small Durer one provided, and to see how it might have rendered it differently.

Expand full comment
John E Simpson's avatar

Most (all?) of the AI image generators I've tried place limits on the length of the prompt. They (all?) also limit the number of images you can generate in a single 24-hour period. Their developers also are understandably nervous about the potential for misuse of the tools, so they might reject a prompt for "terms of service violation"-type errors. Consequently, there's a LOT of prompt editing that potentially has to happen, especially if you've got a fairly clear picture of what result you want.

The illustration for the installment of '23kpc' that went up yesterday is a case in point. As I said in the caption, I simply used as a first-cut prompt a couple sentences copied-and-pasted straight from the text of the post. One AI engine wouldn't accept the passage as written for one of those imagined TOS-violation reasons, and even after repeated rephrasing and eventually using up all the images it let me try out, I still had to take it to a different engine to finally halfway satisfy me.

The experience of illustrating next weekend's '23kpc' post was weird. I ran the prompts -- plural -- through 3-4 different engines; they all got it ALMOST-but-not-quite right immediately. (Probably because there's no human figure involved -- just an inanimate object.) So I had to tune, and retune. And then I wanted the general "look" right. Started out thinking I wanted a blueprint-style image, but none of the results looked right. Ended up choosing one sorta-kinda photo and a couple pencil-sketch variation -- three finalists -- and selected one "winner." But those three finalists were whittled down from probably close to 100 originals, 50 saves, and a dozen or so downloaded for side-by-side comparisons.

A predisposition to OCD perfectionism can be a real hindrance to the process! 😉

Expand full comment