• Light/Dark
  • Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.

Share Your Generative Art |OT| D̵A̵L̵L̵-̵E̵ ̵M̵i̵n̵i̵ MidJourney is pretty damn impressive and can generate almost anything [Read Staff Post]

AI Generative Art is

  • Black magic

    Votes: 110 18.7%

  • Nightmarefuel

    Votes: 154 26.1%

  • Pretty fucking impressive

    Votes: 217 36.8%

  • Overblown

    Votes: 52 8.8%

  • Merely the beginning

    Votes: 271 46.0%

  • Just a computer doing computer stuff

    Votes: 114 19.4%

  • THOR-E: The Generative World

    Votes: 95 16.1%
  • Total voters
    589
BeeDog

BeeDog

Member
Oct 26, 2017
4,549
Ehhh, not the most successful one:

gvttJcD.png
 
plagiarize

plagiarize

Eating crackers
Moderator
Oct 25, 2017
27,511
Cape Cod, MA
Picard prompts are terrifying as I found out when trying to get Dall-E to do it's own version of 'What if everyone was wearing Deanna Troi's outfit.


microsoft_edge_screenpxjlc.png


picardongrassfvjbb.png
 
Nikus

Nikus

Avenger
Oct 25, 2017
10,362
"Too much traffic, please try again."
Haven't been able to use it at all today
 
Green

Green

Member
Oct 27, 2017
7,410
yTs5wMd.png



I also really like how this one turned out. I'd probably have this on my wall tbh:
BXnTy8K.png
 
Last edited:
The Argus

The Argus

Member
Oct 28, 2017
2,291
I haven't found a way to get it to write text on things. Like: The Word "Focus" on a rocket. Gives me a warped rocket. It's super impressive; but everything generated is so warped, especially eyes.
 
underFlorence

underFlorence

Member
May 19, 2019
1,627
Germany
The Argus said:
I haven't found a way to get it to write text on things. Like: The Word "Focus" on a rocket. Gives me a warped rocket. It's super impressive; but everything generated is so warped, especially eyes.
Click to expand...
Click to shrink...
Getting AI to actually generate correct text on images is something that, to my knowledge and understanding, is hard to do within the context of generative neural networks as seen here. That's because as with any other characteristic, neural networks don't really recognize letters as such. They recognize general aspects of the shapes, but they don't really "understand" how it comes together. This could be remedied (and I'm sure there are neural nets that make a use of it) by using an adversarial neural network that also implements text recognition, but it would of course also increase complexity.

With Dall-e 2, it seems that it manages to at generate actual characters, specifically characters that appear on the logo, though the ordering is still messed up:

 
You must log in or register to reply here.