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

[Saracenus]

This is an example of an evolving prompt. I am finding that describing an action with a person or object will get more dynamic images... I am sure if I played with it I could have the knife in the cheese wheel...

Click on a thumbnail to see a larger image.

PROMPT: Olive skinned human woman with asymmetrical features chubby face broken nose ear piercings in a chainmail shirt dagger in hand, various styles

PROMPT: Olive skinned human woman with asymmetrical features chubby face broken nose ear piercings in a chainmail shirt dagger in hand in a tavern room, various styles

PROMPT: Olive skinned human woman with asymmetrical features chubby face broken nose ear piercings in a chainmail shirt using a dagger to slice into a cheese wheel in a tavern room, various styles

 

[Parmandur]

five red-skinned strong agile lithe graceful beautiful robust large muscular Elven lady rangers dressed in leathers covered in buckles and straps laughing uproariously together and drinking wine, tavern, high quality digital painting

five red-skinned strong agile lithe graceful beautiful robust large muscular Elven ladies wearing blue leather jackets over a yellow blouses and green skinny jeans with fashionable purses and gawdy jewelry and bejewled swords on their back chatting and smiling while drinking coffee, magic cafe, high quality digital painting

 

[Saracenus]

Refining my prompt from my previous post, mainly changing the order and some of the wording to see what it will generate. Aside from the knife being held weirdly this came out striking...

PROMPT: female human with olive complexation and asymmetrical face with a broken nose in a chainmail shirt slicing a cheese wheel with a knife in a fantasy tavern room, fine art

Illustration version

Line art version

Woodblock print

Oil and Canvas

Cartoon style

Anime style

Water color

Fibonacci sequence

dia de los muertos style

 

[pukunui]

A city of fire
View attachment 298612View attachment 298613View attachment 298614View attachment 298615

What are the different styles you’ve been using?

 

[Jfdlsjfd]

How would be a D&D version of "Jem and the Holograms"?

I see the AI is not ready to create the exact type of creature we really want. For example a goblin from Harry Potter saga can't be the same like the one from Warhammer, Warcraft, D&D 3.5 o D&D 5ed. It is not its fault. The AI doesn't know yet the D&D creatures.

A real person would do the same. If you speak of an Elf, I have no way to know if it is an House Elf or a Noldor you have in mind. Getting longer prompts, where the distinctive traits are explicitely given to the AI, increases your chance of getting it right. I guess a D&D-only creature like a bulette will be difficult to draw with a general model unless it happened to have been trained to recognize the concept by chance, and in this case it might be only able to draw very similar bulettes...

Other point is the characters always with the same static poses

You can prompt specific poses. I think I posted Hermione doing a handstand earlier in this thread, it is one of the strong point of coupling a LLM to the image generator.

A beautiful elf lady, slouching on a couch, eating pop-corn from a wooden ball. She's facing the viewer. Her feet are crossed and laying on a table In front of her. On the table, a bowl of guacamole is visible. On the top of the image is the quote "Galadriel on the dole".

(Several image were blocked in the display phase, so I guess this prompt is prone to generating NSFW results, and a few other, while child-friendly enough for Microsoft I don't feel safe posting here).

A beautiful elf female warrior in full armour sommersaulting in mid-air, wielding two intricate swords. She's fighting over a group of three goblins warriors who look at her in awe.

Maybe I didn't get what you said with "static poses" but I think you can achieve something very different from the "default" standing up, looking at the viewer.

 

[Parmandur]

A real person would do the same. If you speak of an Elf, I have no way to know if it is an House Elf or a Noldor you have in mind. Getting longer prompts, where the distinctive traits are explicitely given to the AI, increases your chance of getting it right. I guess a D&D-only creature like a bulette will be difficult to draw with a general model unless it happened to have been trained to recognize the concept by chance, and in this case it might be only able to draw very similar bulettes...



You can prompt specific poses. I think I posted Hermione doing a handstand earlier in this thread, it is one of the strong point of coupling a LLM to the image generator.

View attachment 298885View attachment 298886
A beautiful elf lady, slouching on a couch, eating pop-corn from a wooden ball. She's facing the viewer. Her feet are crossed and laying on a table In front of her. On the table, a bowl of guacamole is visible. On the top of the image is the quote "Galadriel on the dole".

(Several image were blocked in the display phase, so I guess this prompt is prone to generating NSFW results, and a few other, while child-friendly enough for Microsoft I don't feel safe posting here).
View attachment 298889View attachment 298890

A beautiful elf female warrior in full armour sommersaulting in mid-air, wielding two intricate swords. She's fighting over a group of three goblins warriors who look at her in awe.

Yeah, you can get better results by being specific on pose, facial expression, etc. And adjectives, adjectives and adverbs.

strong agile lithe beautiful robust Elven lady wearing a gown and sunglasses smiling while smoking and drinking wine, tea garden, high quality digital painting

 

[pukunui]

More Orc-Elf courting:

View attachment 298636View attachment 298637View attachment 298638

How did you get it to give the elf girl the same exact outfit in every iteration?

I see the AI is not ready to create the exact type of creature we really want. For example a goblin from Harry Potter saga can't be the same like the one from Warhammer, Warcraft, D&D 3.5 o D&D 5ed. It is not its fault. The AI doesn't know yet the D&D creatures.

I was thinking last night that it would be cool if WotC came up with their own art generator trained on official 5e art, so that you could make your own art of goblins that look like 5e goblins and so on. That's unlikely to ever happen though, especially given their stance on AI art after the Bigby's issue.

 

[Reynard]

What are the different styles you’ve been using?

"High resolution digital art"
"Moody concept art"
"Heavy ink sketch"
"Gritty comic book"

 

[Parmandur]

How did you get it to give the elf girl the same exact outfit in every iteration?

100% did not.intend that: I assume that it is an algorithmic quirk when requesting a group of people with the same description, cut and paste. It was notable that I got diverse clothing when I stipulated "rainbow."

 

[Parmandur]

How did you get it to give the elf girl the same exact outfit in every iteration?

Oh, !I see you were talking about thw green leather jacket and floral dress. They are not the same, each is diatinctly different in the details.