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


log in or register to remove this ad


My wife's Squishmallow prompt design was for a Beholder (see the previous Teifling), but it wasn't coming out right. Came back to it last night and cracked the code to get it right...

PROMPT: Photo of a soft and plush Sqishmallow designed after the D&D beholder. It has a prominent central eye on its body and multiple eye stalks with eyes at their tips. The toy boasts pastel colors and a joyful smile.

1697745698463.png
 

Showed it to the wife, she nixed the pastels and wanted it more "traditional..." Well, I want one of these...

PROMPT: Photo of a plush Sqishmallow toy designed after the D&D beholder with a prominent central eye on its body. The toy has multiple eye stalks with eyes at their tips, colored in rich red hues, and showcasing a joyful smile.

1697746078353.png


Here is the version Bing created with the same prompt. (the above was done in GPT-4).

1697747139419.png
 
Last edited:



Whats the prompt for that?

PROMPT: Full-color character reference sheet capturing a Lizardfolk warrior from the D&D realm, with features and skin texture echoing that of a Komodo Dragon. This muscular reptilian figure is armed with a reflective shield and a pointed spear, poised for any adversary. The image is widened and lengthened to provide a more expansive view of the character and its surroundings.

Just remember that GPT-4 can use prompts longer than 400 characters, which is a limitation of the Bing version of DALL-E. I think this prompt should work as is.

EDIT: It does work, here is example it did in Bing:

1697747404954.png
 
Last edited:

To be fair, that's probably related to the training data. Red hair and dark skin are not a naturally occurring phenomenon, so the algorithm will avoid it unless forced most likely.
That isn’t quite accurate, but besidea that, it’s weird the program can’t just parse “hair color red” as overriding source hair color.

It has an easier time doing that for blue hair than red.

But seriously red hair appears naturally on every continent, very much including amongst Polynesians with dark skin. It’s 100% just the biases of the people feeding the program training data, and of western (especially American) popular culture.

Here are some that did alright.
 

Attachments

  • IMG_2168.jpeg
    IMG_2168.jpeg
    282.9 KB · Views: 57
  • IMG_2169.jpeg
    IMG_2169.jpeg
    256.2 KB · Views: 55
  • IMG_2170.jpeg
    IMG_2170.jpeg
    284.4 KB · Views: 57


So, I played around with the base image in GPT-4 to iterate on it... There are all kinds of wrong in this image, it seems the more I attempt to play, the more artifacts it introduces. :ROFLMAO: The lizard nips and the extra tail growing out of its arm are precious. The good news is I can slam this into Photoshop and remove these details pretty quickly with the newly introduced Generative AI, meaning I don't have to erase things and then touch them up by hand, which will take minutes instead of an hour or two.

1697748007674.png
 

Remove ads

Top