Based on the popular book

  • Mr Fish@lemmy.world
    link
    fedilink
    arrow-up
    6
    ·
    11 months ago

    Yeah, it’s pretty normal for image generation. It has no real understanding of language. All it knows is, “This shape is often near this shape.” Honestly, it’s surprising that nearly all the shapes in these are actually letters, even if they’re in the wrong order. You’ll often see half of a letter missing, two letters merged, or other weird letter-like shapes.

    • Ignotum@lemmy.world
      link
      fedilink
      arrow-up
      2
      ·
      11 months ago

      Thus is the best text over seen a model generate thus far, the model i usually use generates text like this:

      And the earlier versions of the models were even worse :P

      • 0x4E4F
        link
        fedilink
        English
        arrow-up
        2
        ·
        11 months ago

        Why not just make it generate the image and you add in the text. It’s far simpler than generating the entire image yourself, the text alone is like an hour, tops, in Photoshop.

        • Ignotum@lemmy.world
          link
          fedilink
          arrow-up
          2
          ·
          11 months ago

          Yeah if i was going to make something like this i’d just add the text myself, would be convenient if the models eventually got good enough to make the finished thing in a single step though

    • 0x4E4F
      link
      fedilink
      English
      arrow-up
      1
      ·
      11 months ago

      Why not just make it generate the image and you add in the text. It’s far simpler than generating the entire image yourself, the text alone is like an hour, tops, in Photoshop.