• Seralth@piefed.seralth.com
    link
    fedilink
    English
    arrow-up
    1
    arrow-down
    3
    ·
    18 days ago

    Modern models don’t really have tells anymore. They either explode really obviously or good enough that they just look like a normal casual artist.

    The small hints and tells just don’t really happen anymore. Doubly so if words are added by hand :/

    • ThunderComplex@lemmy.today
      link
      fedilink
      English
      arrow-up
      4
      arrow-down
      1
      ·
      18 days ago

      You are absolutely wrong. Just go look at the recent AI flyer trend. AI has and will always have tells (as long as we talk about the traditional DCGAN distill noise to image concept that’s the norm right now). You can minimize them with more effort, but a lot of people don’t even go thru that effort or know that AI image gen extends far past the “ChatGPT make me a image” paradigm.

      • n0xy (it/its)@lemmy.dbzer0.com
        link
        fedilink
        English
        arrow-up
        3
        ·
        15 days ago

        The most obviouo ones are oneshot generations who don‘t even care about issues. But with own pestprocessing you can do a lot better. Or just generating parts of the image and arrange them freely, including covering the worst issues. Maybe I should try how good that can work with local AI and Krita/GIMP

        • ThunderComplex@lemmy.today
          link
          fedilink
          English
          arrow-up
          2
          ·
          15 days ago

          Yip. Considering how good AI image gen can actually be, those oneshots are just a spit in my face that very obviously say “I do not care how this looks, fuck you for complaining”.
          Also idk if “pestprocessing” was a typo but it fits so well here xD