D&D General DALL·E 3 does amazing D&D art

[overgeeked]

@overgeeked.
Bloody h... By Ningauble!

LOL. I didn't even come close to that back then!

This also works well. "Ligne claire comic art style of Hergé in Tintin. Clean lines. Minimalist background detail. Realistic proportions. Exaggerated facial expressions and body language. Limited use of shadows. A pale blue Renault 4 on a cobblestone street. One person is sitting in the car. One person is walking down the street."

Re: "AI" art fooling you. It's how the human brain works. You take in the big picture but gloss over the details or fill them in yourself. It's how comic art works, too. And fiction for that matter. Your brain just fills in the details.

The trick, I think, is to lean into the "AI's" strengths and away from its weaknesses. Don't try to replicate real places or situations. Lean into the fantastic. Lean away from realism. Pick art styles that don't rely on realism and subjects that don't exist. If you can't reference the thing in the real world, you won't know that something's off even if you inspect the hell out of it.

Just switching the art style can hide many of the flaws of asking an "AI" to create detailed images of real places.

For example. "Rough digital painting. A pale blue Renault 4 on a cobblestone street. One person is sitting in the car. One person is walking down the street."



Or. "Simple cartoon or comic book art style. A pale blue Renault 4 on a cobblestone street. One person is sitting in the car. One person is walking down the street."

 

[Reynard]

Perhaps they should learn to run a "final check" of sorts on the images they produce?

As I understand it, that is how they check of "unsafe content" such as nudity or whatever. Not sure if it could be implemented for actual quality, though.

 

[ilgatto]

As I understand it, that is how they check of "unsafe content" such as nudity or whatever. Not sure if it could be implemented for actual quality, though.

I suppose checking for such things would be a fairly easy process. Just give the things some data on things NSFW and let them compare their products to that in some way. But, even though they would be rather specific, these "data" could well amount to a huge amount, perhaps making it unfeasible to do that for the gazillions of images that could be generated.

And even with these limited checks strange things seem to happen I've always suspected that my attempts to make an image of a single red rose on an unmade bed sort of automatically fell into the NSFW category for some reason - the things just wouldn't do it. And what can be simpler?

 

[ilgatto]

This also works well. "Ligne claire comic art style of Hergé in Tintin. Clean lines. Minimalist background detail. Realistic proportions. Exaggerated facial expressions and body language. Limited use of shadows. A pale blue Renault 4 on a cobblestone street. One person is sitting in the car. One person is walking down the street."

Re: "AI" art fooling you. It's how the human brain works. You take in the big picture but gloss over the details or fill them in yourself. It's how comic art works, too. And fiction for that matter. Your brain just fills in the details.

Yeah, you're seeing, but not looking. It's also a biological/mechanical thing. You use different parts of your brain when you're visually interacting with your environment. One of these fills in details "from memory" as it were - a bit like an AI works, come to think of it.

The trick, I think, is to lean into the "AI's" strengths and away from its weaknesses. Don't try to replicate real places or situations. Lean into the fantastic. Lean away from realism. Pick art styles that don't rely on realism and subjects that don't exist. If you can't reference the thing in the real world, you won't know that something's off even if you inspect the hell out of it.

Yeah. You get nicer images as well, IMHO. I'm not a fan of these hyper-realistic things, although they can be bloody impressive.
Funny thing is, that these hyper-real images sort of all look the same in the end. I've been reading through this thread and found that I was starting to skip them as a matter of course. Another one, another one, ah, yes some more. Started skipping through them until something caught my eye. Could be a matter of taste, though.

For example. "Rough digital painting. A pale blue Renault 4 on a cobblestone street. One person is sitting in the car. One person is walking down the street."

Just switching the art style can hide many of the flaws of asking an "AI" to create detailed images of real places.

View attachment 357071View attachment 357072View attachment 357073View attachment 357074

Or. "Simple cartoon or comic book art style. A pale blue Renault 4 on a cobblestone street. One person is sitting in the car. One person is walking down the street."

View attachment 357079View attachment 357080View attachment 357081View attachment 357082

Heh. Just can't get over how you keep producing these R4s as if it's the easiest thing in the world.

 

[overgeeked]

Yeah, you're seeing, but not looking. It's also a biological/mechanical thing. You use different parts of your brain when you're visually interacting with your environment. One of these fills in details "from memory" as it were - a bit like an AI works, come to think of it.

Yeah. You get nicer images as well, IMHO. I'm not a fan of these hyper-realistic things, although they can be bloody impressive.
Funny thing is, that these hyper-real images sort of all look the same in the end. I've been reading through this thread and found that I was starting to skip them as a matter of course. Another one, another one, ah, yes some more. Started skipping through them until something caught my eye. Could be a matter of taste, though.

Yeah. There's something about infinitely generating "AI" art that makes you less and less impressed with it as time goes on. Not in a quality sense, that's steadily improving. But in a novelty sense and in an appreciation for "art" sense. It literally makes "art" disposable. You don't need to bother keeping or hording the "art" because you can generate an infinite amount of it. It becomes worthless, effectively. Unless there's a piece that really just pops. In a way, the prompts that generate the "art" become more important and "valuable" than the generated "art" itself.

Heh. Just can't get over how you keep producing these R4s as if it's the easiest thing in the world.

I mean...it is almost literally nothing. You type the prompt in and hit "create". There's effectively zero effort.

Some people out there really love their R4s and took a lot of pictures of them. Then the "AI" programmers came along and stole them. So now these "AI" art programs can generate R4s with ease. So too with the art and art styles.

 

[ilgatto]

Yeah. There's something about infinitely generating "AI" art that makes you less and less impressed with it as time goes on. Not in a quality sense, that's steadily improving. But in a novelty sense and in an appreciation for "art" sense. It literally makes "art" disposable. You don't need to bother keeping or hording the "art" because you can generate an infinite amount of it. It becomes worthless, effectively. Unless there's a piece that really just pops.

My thoughts exactly!

In a way, the prompts that generate the "art" become more important and "valuable" than the generated "art" itself.

Indeed, indeed. As I was trying to get the AIs to do what I wanted, I found myself thinking more and more about how humans communicate and how "language" is actually a construct that can be very clarifying and very misleading at the same time.
When we try to explain things to others, we use words in ways to make the people we talk to most likely to understand what we're trying to say. It's easy for me to talk with old friends about where we grew up. When I say such words as "pond", "lamp post", "the school", "that tree", everybody immediately understands what I'm on about because we were all there.

For an AI, these words can literally mean anything. First, they're just letters in a certain sequence (which actually isn't even important to AIs, as I understand it), then there's many meanings to the words they form. "Pond" is a pound in Dutch, "the school" can be a school of fish, a "lamp post" can look like anything.

So, like you say, mastering the art of "prompting", of "talking to the AI" so it will catch your drift, becomes paramount - and made me think a lot about how people can sometimes completely miss points you think you made glaringly obvious.

Actually, the whole concept of an AI is utterly fascinating when seen in this and other lights.

Unfortunately, humans will be humans.

I mean...it is almost literally nothing. You type the prompt in and hit "create". There's effectively zero effort.

Yeah, yeah.
Almost tempted to have another go at it myself!

 

[overgeeked]

Yeah, yeah.
Almost tempted to have another go at it myself!

 

[overgeeked]

For an AI, these words can literally mean anything. First, they're just letters in a certain sequence (which actually isn't even important to AIs, as I understand it), then there's many meanings to the words they form. "Pond" is a pound in Dutch, "the school" can be a school of fish, a "lamp post" can look like anything.

So, like you say, mastering the art of "prompting", of "talking to the AI" so it will catch your drift, becomes paramount - and made me think a lot about how people can sometimes completely miss points you think you made glaringly obvious.

Actually, the whole concept of an AI is utterly fascinating when seen in this and other lights.

I’m not sure there’s any “mastery” involved. From what I can tell, it’s basic descriptive language. “A pale blue Renault 4” is far more descriptive and precise than “a car”. Same goes for anything you want the “AI” to produce. If you want a swirl of energy around a tree, you have to describe it as precisely as you can. Abstract concepts don’t translate. You have to describe them. If you type in “love” you’re bound to get a heart emoji or two people kissing…or the prompt blocked for inappropriate content. But, weirdly, you can get it to produce abstract and psychedelic art.

 

[Bio-Karbon]

Starting to dabble with the Gnolls of Nost, coming out pretty darn good so far:

My personal favorite:

 

[Reynard]

My personal favorite:

That's really great. Which AI is that?