• AFK BRB Chocolate@lemmy.world
    link
    fedilink
    English
    arrow-up
    8
    ·
    1 day ago

    This thread makes me so sad about the future. Here we have a bunch of folks who are absolutely convinced that a real image is AI generated, some even after being shown that it predates AI image generators. But we live in a time when there are a giant number of real images online, the only AI generated images are very recent, and most of those have things like extra fingers. What’s it going to be like in twenty years, when image generators are so much better and there are so many more generated images? Having a photo of something will prove nothing.

    • tfw_no_toiletpaper@lemmy.world
      link
      fedilink
      English
      arrow-up
      3
      ·
      edit-2
      20 hours ago

      Extra fingers etc are gone except if the prompt monkey missed it. Current ones you can spot by them being off / feeling off or zooming very close into the picture to spot weird transitions. It’s not “20 years into the future”, it’s now.

      I was also fooled, for the past idk 6 months social media is flooded with AI generated cat pics (and videos) and they all look pretty much like the picture. I think it was just in its training data and feels so similar because of that.

      Example: Go ahead, try to spot anything besides the offness

      • AFK BRB Chocolate@lemmy.world
        link
        fedilink
        English
        arrow-up
        1
        ·
        17 hours ago

        There’s for sure the “offness,” but there are also things like the lack of veins in the cat’s ear, which we’d see with the like coming through like that. But I agree, they’re getting very good.

    • floofloof@lemmy.ca
      link
      fedilink
      English
      arrow-up
      6
      ·
      edit-2
      1 day ago

      I doubt it’s real. The foot placement is a bit implausible and what is that thing behind it? Feels AI-ish.

      Edit: I think it’s probably real after all, given the age of this reddit post and the existence of other photos that seem to be from the same shoot (Pinterest link).

      Edit: Someone else found the Instagram page and the image was posted there 5 years ago.

    • TheFriar@lemm.ee
      link
      fedilink
      English
      arrow-up
      7
      arrow-down
      1
      ·
      1 day ago

      It’s a cartoon-level of cuteness. The one in the bunch that goes to the door to put on the cute while the others sneak in through the open door to raid the kitchen.

      • It’s not AI. The oldest version of this image is from May 11th 2020, and the first widespread text-to-image model DALL-E came out in January 2021. That makes this image too old to be AI-generated.

        It also doesn’t show any strange AI defects the early models had, in case you may think it’s from a very early proprietary unreleased model or something.

      • ReallyActuallyFrankenstein@lemmynsfw.com
        link
        fedilink
        English
        arrow-up
        4
        ·
        edit-2
        24 hours ago

        Oh, wow…I let my guard down for one second. Ugh.

        Edit: I’ve gone back and looked at it very closely. I still don’t think it’s AI, on balance. This seems like a still from a video clip that has compression artifacts. My clues:

        Against AI:

        • The fine details like whiskers have enough random but consistent imperfection that it doesn’t match what AI normally outputs (either consistent perfection, or inconsistent imperfection - too-perfect whiskers or whiskers that blend into each other is what I’d expect from AI). There is some whisker-blending, but that seems more consistent with video compression blur than AI.
        • The couch seems to have post-compression remnants of hairs that aren’t likely from AI.
        • The eyes are one area where AI is very inconsistent, but the irises (for dilated eyes) are very consistent. Though there is heterochromia, I’d expect to see some AI mess around the eyes.

        For AI:

        • I agree the chew toy looks off. I don’t know that that counteracts the above.
        • It could be cropped and compressed explicitly to obscure AI giveaway details.

        I’m not 100% sure, but I suspect it’s real.

        Edit 2: As others have noted, it’s real. My AI detector isn’t obsolete, for a few more months/weeks/days at least.

  • tfw_no_toiletpaper@lemmy.world
    link
    fedilink
    English
    arrow-up
    4
    arrow-down
    4
    ·
    edit-2
    1 day ago

    Edit: Not AI since the og image is too old for that

    ~~From the c/aww sidebar:

    AI posts must be labeled [AI] in the title and are limited to one per week

    This is not the case OP, either change the title or remove the post~~

    • It’s not AI. The oldest version of this image is from May 11th 2020, and the first widespread text-to-image model DALL-E came out in January 2021. That makes this image too old to be AI-generated.

      It also doesn’t show any strange AI defects the early models had, in case you may think it’s from a very early proprietary unreleased model or something.

      • Boomkop3@reddthat.com
        link
        fedilink
        English
        arrow-up
        1
        arrow-down
        6
        ·
        1 day ago

        Ah darn, because a decade old technology is only available to the first company to sell it