I’ve posted 2 images over the past month that turned out to be AI and I’m getting angry at myself for not crabbing them. Does anyone have a good resource or way of determining AI generated images?

  • Deestan@lemmy.world
    link
    fedilink
    arrow-up
    18
    ·
    edit-2
    9 months ago

    Most of the time, it helps to zoom in and see if things are in a “wrong” place or “melt” into each other, or something that should be symmetric isn’t.

    Finding letters and words helps, too. They’re often nonsense words or unclear letters.

    E.g. https://lemmy.world/post/10202384?scrollToComments=true if that’s one you were referring to, you can see symmetry problems in the support beams, floors. Letters and writing is wrong. Back of truck has stuff “melding”.

    • Stovetop@lemmy.world
      link
      fedilink
      arrow-up
      3
      ·
      9 months ago

      This might work for now, but there’s likely gonna be no way to tell someday soon. What do we do for the long term?

      • Deestan@lemmy.world
        link
        fedilink
        arrow-up
        6
        ·
        edit-2
        9 months ago

        Yeah, long term it will be near impossible to tell by inspecting the image.

        We will end up with a system of identifying and trusting the image creator to be truthful, and tracking which/if modifications have been done to the picture.

        We have cryptography tools that can be used for most of that, but how it will work as a system for all the weird problems people have (like dying, grifting, stealing, selling access, lying, infringing) is not something we can just throw some tech on and assume it solved.