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

@Reynard, I took your prompt, made some minor mods, and generated this:

exterior, snowy rocky landscape, female warrior with bow is surprised when an armored skeleton with a huge battleaxe bursts up from the frozen ground, high detail digital painting

View attachment 296804
This is so close to an actual Dragon magazine cover it’s hard to imagine it wasn’t in the training data.
 

log in or register to remove this ad

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.
Yeah, that makes sense. Probably the same for sports crowds, classrooms, and so forth. (And also cattle stampedes, I've found!) A crowd is almost like a texture in a sense, fairly straightforward to grasp at a small scale then stamp out more in great quantity. And on the other hand, there are so many portraits of individuals with sufficient metadata to define a huge variety of poses, expressions, lighting conditions, hawtness, and whatnot.

But a small party of unique individuals seems to be tougher, at least with the limited prompt length available in the Bing's Dall-E interface. It feels obvious as a user to just ask the AI to "zoom in on each unique subject and do whatever you do for a portrait of an individual subject." But apparently it's not how it works. Maybe with a bigger context window and some ability to actually converse/iterate with the backend chat bot it might be doable, just like how you can direct an LLM to write a particular scene or whatnot.

In any event, it certainly is fascinating.
 

Any time I try to put in a crowd in a scene I have to be very specific on the make up said crowd or I will get a default to a bunch of dudes. If I had more of an art vocabulary I might be able to manipulate the composition a bit more.

I am still having a problem with stating human and getting pointed eared elves. I definitely have to work at getting ethnic diversity in my art with this thing because it does default to Caucasian folk quite rapidly.
 

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.
Ah, indeed... I guessed as much! I've been intentionally avoiding investigating higher end software because then I'd probably never leave the house. 😅
 


I am still having a problem with stating human and getting pointed eared elves. I definitely have to work at getting ethnic diversity in my art with this thing because it does default to Caucasian folk quite rapidly.
Not sure why, but when I switched from overtly fantasy scenes to the old west ones (ie, stated explicitly in the prompt), the ears became normal human ears and all the people became decidedly less beautiful. I assume that's because relevant data underlying the renders had shifted significantly to real world imagery/photos instead of ArtStation, etc?

Still, even in "old west mode" it couldn't quite get over the elf thing:

00000.jpg


I blame cosplayers.

.
 


I clearly need to work on prompts and use my boosts.....

Dont be afraid to use them all!

I got my first 100 boosts on Sunday of past week, then got 100 others the next day for the new week, then got a bunch of free ones that never went down because they were ''sorry the website was taking so long to generate a picture'', then this morning (I was still at 0 boost because I abuse the website) I got another 100 boosts after the site did not let me create pictures for a few hours.

...god...I generated more than 300 pictures (x4) in the past 2-3 days!
 



Remove ads

Top