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


log in or register to remove this ad

Huh. This is the first time I've been getting the blemishes.
I've definitely noticed them much more on head-shot type portraits. Just did some (unstyled) ones a little while ago that came out like photographs, and noticed the white gobs. Tbh I can't say they're more or less prominent lately, though. Hopefully it gets resolved.
 


I've definitely noticed them much more on head-shot type portraits. Just did some (unstyled) ones a little while ago that came out like photographs, and noticed the white gobs. Tbh I can't say they're more or less prominent lately, though. Hopefully it gets resolved.
Weird. Today is the first day I've noticed them - both in my own creations and those that other people have posted here. I do occasionally get little weird off-color bits around the edges of a figure (often their ears) but not these obvious white spots and faulty body parts where it looks like the algorithm failed to finish the image properly.
 

Been playing around with @Saracenus' art deco prompting:

Nissa Nackle, gnome vengeance paladin of the Silver Flame

_4bbca90f-5bf5-42be-a0e5-813184256fda.jpg
_94c83a75-ab47-4870-ae9c-116cf4869ae4.jpg
_867ea122-083e-4dba-9d88-ce62935befb4.jpg
_4964478e-895f-4c73-a6b0-fb1cef174cd8.jpg


The prompt indicates mithril chainmail with a white tabard bearing a stylized silver flame on it. As you can see, it doesn't always get the armor or the tabard right (sometimes it substitutes in a cloak). She's also meant to be holding a rapier and a shield. It mostly does OK with the shield, but it doesn't seem to know what a rapier is.
 

I have issues with AI-generated images, but I can't deny that it's producing some very striking results. The prompts here are variations on this: "a beautiful feminine sorceress with long rainbow hair wearing a white suit with pinstripes and a vest but no hat casting a spell standing in a circle of magic symbols on a crowded Victorian city" I added in various styles (these were mostly "construction paper collage") and for some, swapped out the white suit for a dress shirt with rolled-up sleeves and tattoos.
_50ac4039-cb91-4ce1-8fb3-0fc549ebe28a.jpg

_3222bf67-65a9-478c-86b6-48519f153668.jpg

_44015d6b-bc61-41ef-9f2d-4d11c474638c.jpg

_793615d1-cadc-49f2-8ffb-55cfb04389f6.jpg


Given what I've been going for, the fact that the AI got weird with the arms and fingers here actually works well.
_2b459656-e54a-4ee9-9c3b-ac5be1c42541.jpg


My main complaint (aside from ethical issues about the underlying technology, which I don't want to discuss here) is that the system really wants to put hats on these images. Almost every run has at least one hat; if I don't say "with no hat" then they're basically all be-hatted. "Bareheaded" works sometimes, but it also makes the focal character bald quite often.
 




Concept bleeding with "fairies."
Maybe, but it often gives her pointy ears even when I don't include the fairies. It might be that I'm specifying that she's a sorceress or that I'm mentioning that she's in an enchanted setting. (Because of her background as a "feylost" and the fact that the bulk of the Witchlight adventure takes place in the Feywild, I've been trying to produce images of her in fey-like settings.)

OK, this time I got some without pointy ears!

_6abb8abd-048b-4d02-8684-f80296274291.jpg
_d9cc4899-f909-4440-af42-c2ae5d9522d3.jpg


Now I need to see if I can add in her friendly but still nightmarish-looking quori spirit somehow.
 

Remove ads

Top