• The VOIDRUNNER'S CODEX is LIVE! Explore new worlds, fight oppressive empires, fend off fearsome aliens, and wield deadly psionics with this comprehensive boxed set expansion for 5E and A5E!

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

Scribe

Legend
I wonder what makes it decide to do 1, 2 or 3 images instead of 4.

Yeah, I wish I could get back into but its not letting me create and now it says 30 min wait, but I did have a few prompts for my Genasi Monk that only came back with one response, and most came back with 3.
 

log in or register to remove this ad


IIRC, it's a confidence issue of there being sifficent information to make multiple images. Think more complex requests run the risk of the AI getting stuck on one or two.

It's quite complicated because what's under the hood isn't known.

But one can do some educated guess.

The happy few with dall-e 3 through chatGPT pro subscription have posted videos where they show some more information. They type a simple prompt and ChatGPT "enhances" the prompt with details to make 4 different images. Then, the graphic engines creates the images.

Given how AI generally works, it's possible the added details and words are those closely and often encountered together. If your prompt contains a word that happens to generate an "unsafe association" at this stage, it's possible the image gets cut off before generation (saving computing power).
 

Reynard

Legend
Supporter
Prompt: tall muscular powerful beautiful female orc fantasy warrior woman wearing leather armor and carrying a very large sword on her back, she leans down to smile at a cute friendly short thin male elf man wearing an embroidered jacket, he looks up at her lovingly, colorful high quality digital painting style
hawt.jpg

It is interesting how the gendered terms "muscular powerful" and "cute friendly" weigh so much more than actual gender identifiers "female" and "male."
 

Scribe

Legend
It is interesting how the gendered terms "muscular powerful" and "cute friendly" weigh so much more than actual gender identifiers "female" and "male."

Its the same with Elf ears, fit/athletic female (I spent most of my boosts trying to get her in more than work out attire!) and trying to push a different skin tone, like my onyx/coal skinned bald dwarf (who took the REST of my boosts).

The system produces what on what the expected default would be, because thats what populates the data set in a feedback loop.

Tropes, Stereotypes, and a general understanding based on weight of content.
 


J.Quondam

CR 1/8
Its the same with Elf ears, fit/athletic female (I spent most of my boosts trying to get her in more than work out attire!) and trying to push a different skin tone, like my onyx/coal skinned bald dwarf (who took the REST of my boosts).

The system produces what on what the expected default would be, because thats what populates the data set in a feedback loop.

Tropes, Stereotypes, and a general understanding based on weight of content.
I spent an entire resupply of boosts yesterday trying to manage over-zealous elf ears.
It did not go well.
 



Reynard

Legend
Supporter
I am legitimately surprised it was able to follow the view directions in my prompt.
a dark and misty swamp filled with blighted trees and stinking black water, a massive bloated zombie covered in mud and moss lumbers through the swamp, all viewed looking over the shoulder of a tiny slim fairy hiding from the zombie behind a tree branch [style]
swampy1.jpg
swampy2.jpg
swampy3.jpg
swampy4.jpg
 

Voidrunner's Codex

Remove ads

Top