Yeah, yeah, I know, Google, but it’s pretty cool and I’ve made a ton of fast, easy, trippy, fun little artworks with it.

  • @Kind_Stone
    link
    22 years ago

    For those who like to mess with stuff like that - there’s also ArtBreeder which can generate different pics using genes.

    • KiG V2OP
      link
      12 years ago

      Woah. What the fuck. This shit is wild. I’m gonna have to make an account tomorrow. Thanks for the plug!!

  • KiG V2OP
    link
    12 years ago

    For those not in the know, Google Deep Dream is an A.I. that takes an image you upload and recreates it using a “style image” as a sort of “palette.” This was the original image I used:

    And this was the style image I fed it to use to recreate the first image:

    To add a little depth to my input, I make little collage boards like that, which often produces cooler results than simply adding a single picture (although I’ve made plenty of awesome ones just using single images). Since Google sucks you have to use the site for a bit before you can have access to higher quality images, and it has a cap on how many you can make in a short period of time, but idk I really like it, I have used some for inspiration for concepts or ways to approach art and have even painted a couple with the intention of selling. Plus it’s just trippy eye candy, can’t go wrong with a lil of that in your life.

    • @KommandoGZD
      link
      22 years ago

      Got intrigued by your post, because that image is mean. How did you manage to preserve the shape and features of his face so well? I tried a couple of photos with different stock style images (single and collage) with different weights and they all washed out the shapes of people in the photos massively. Backgrounds tended to look pretty good, but faces etc were hardly recognizable. Is that more on the type of photo (eg close-up on the one you used) or the style image and settings? Or just a combination of all of them?

      • KiG V2OP
        link
        22 years ago

        To be honest sometimes I’ve found little rhyme or rhythm to it. I used to experiment with the settings a lot but I couldn’t find a “secret sauce” that way, I tend to always stick these days to 100% scale and 50-60% weight. I’ve never once used “enhance.”

        Like I said, Google sucks and part of the problem is you need to make a certain number of Dreams before you have access to higher iterations and higher resolutions–THOSE two are definitely the most important.

        I tend to have the most problem with people/faces as well, while the “popular” page is chock full of nothing BUT faces. Believe me, I have tried doing faces dozens of times before that turned out really bad looking, either way too abstract or barely different from the original image.

        I will say that sometimes a style image/dream image will just, for whatever reason, NOT want to do anything fun, no matter how you try it.

        Good luck comrade!