D&D General DALL·E 3 does amazing D&D art
- Thread starter M.T. Black
- Start date
Parmandur
Book-Friend, he/him
Oh, I'm sorry, I lost the exact prompt for that obe. But it was surprisingly basic, "wearing a green leather jacket over a floral dress". Guess that was good enough to get a consistent result.
J.Quondam
CR 1/8
Yes, I'm aware of that, of course. That's why I use the scare quotes.
Parmandur
Book-Friend, he/him
What were the exact style prompts to get that...?
Parmandur
Book-Friend, he/him
Sorry, pedantry on my part.
J.Quondam
CR 1/8
It's fairly simple, something like "grainy old west sepia photograph" or some variation on that.
Eta: Similarly, in the others I used things like "grainy black and white security camera footage" or "polaroid photo faded colors". Basically just a statement of the particular medium.
Parmandur
Book-Friend, he/him
The security camera footage of the 1E PHB is an inspired take.
J.Quondam
CR 1/8
Thanks! I tried several takes and wrestled with it for a long while, though, trying to get the right number of adventurers, different angles, the dead lizard, etc. But couldn't even get it to render someone climbing on the statue, much less prying out the eye as in the source image. Same issue with the dragon hunters, trying to get them lined up, dressed, and sized analogously to the Elmore painting just didn't work for me. Clearly I have much to learn about AI-whispering.
It's impressive what people have done for individual subjects and with crowds. But seems that when working with multiple subjects, it's just hit or miss, rinse and wash, over and over again, til you get kindasorta close to what you're aiming for.
Now when a user can select parts of an image for refinement, "freeze" particular subjects for repetition, specifically place things relative to one another, and so forth? That will be incredibly, scary powerful. I imagine that the high-end apps like Adobe's suite are working on just that sort of thing.
pukunui
Legend
I was just thinking that all those bearded dwarf ladies would fit nicely into Discworld.
Yes, I can already do that in Photoshop 2024. I can select a part of the image to change with its generative AI. You can even erase something and have it fill in the background. It's not perfect by any means but it's very impressive.
Parmandur
Book-Friend, he/him
I think it isneaaier to do crowd stuff when tapping into a huge database of sample images, such as with those religious images I did tapping into Christian and Hindubdevotiknal art.Thanks! I tried several takes and wrestled with it for a long while, though, trying to get the right number of adventurers, different angles, the dead lizard, etc. But couldn't even get it to render someone climbing on the statue, much less prying out the eye as in the source image. Same issue with the dragon hunters, trying to get them lined up, dressed, and sized analogously to the Elmore painting just didn't work for me. Clearly I have much to learn about AI-whispering.
It's impressive what people have done for individual subjects and with crowds. But seems that when working with multiple subjects, it's just hit or miss, rinse and wash, over and over again, til you get kindasorta close to what you're aiming for.
Now when a user can select parts of an image for refinement, "freeze" particular subjects for repetition, specifically place things relative to one another, and so forth? That will be incredibly, scary powerful. I imagine that the high-end apps like Adobe's suite are working on just that sort of thing.
