Skip Navigation
79 comments

  • This doesn't surprise me, given how messy Facebook has become. What does disturb me is people not being able to recognize that they are AI-generated. Now, this could be due to the AI becoming so sophisticated that it can actually generate life-like images, or it could be due to humanity's inability to question what they're viewing and whether it is true or not. Either way, this is very concerning, and if it can happen on Facebook, I'm sure it's also happening on other social media sites as well.

    Speaking of which, how can we stop something like this from happening on Lemmy and other federated sites?

    • This is just the next wave to me. I wish we had better automation around helping people find the sources of an image or detect where it could from. I've showed people reverse image searching a little at least, hopefully we can see it improve to handle new forms of copies like this going forward.

    • It really depends on the image. Some AI generated images of people really seem indistinguishable from a photograph. Not all, but it's definitely going to become more common. At this point it seems like every month has a new breakthrough for some aspect AI regarding consistency and realism. As people work to bring stable diffusion video to a more reliable state the images themselves are only going to get better.

      There's a number of models out there right now that don't have hand issues as much. There are other methods like using a pose to force hide the hands. We're getting to the point where we just need metadata viewers on every image because the eye alone isn't reliable for this. How many artists have already been accused of being "AI art like" and they were not.

      Stopping it? I don't know if you can stop Pandora's box. Personally, I'm of the opinion that flooding AI images is the only way to "stop" it, simply by making people not care about it. At a certain point, you can only make so many variations of gollum as a playing card /ninja turtle/movie star before they get boring to the public. I also have doubts that the people using this are overall the type of people who would have paid artists for commissions, or that that would even stop a sale from happening in the first place (my partner makes mushroom forest scenes with Stable Diffusion but she also buys about $300 of art from our friends and local artists through the year). Like, Stable Diffusion doesn't make oil paintings.

      Little Jimmy in his room or the 60 hour week worker? They pose no harm making these images and it makes them happy, so why take that away from them? And regarding the nefarious aspect of it - that one is much harder but it's in part societal shame as well. Explicit images made by AI may make the process easier, but if it's something nefarious the people will find a way to do it regardless. Video sex scams and catfish were around long before AI. I'm not certain that these will inherently become more prevalent as the tech becomes more accessible. It's the people posting them. Which to me comes down to more of a societal issue than a technological one.

      However, in terms of trying to stop it maybe there could be a hash similar to the CASM that gets used here. Maybe the image uploader can look for metadata markers that come with generated images and deny those ones? That's an easy workaround though, since metadata can be stripped.

      I dunno. I think we've gotta EEE AI. Embrace AI images by flooding society with their presence, Extend AI images by getting the ability to use it into the hands of everyone, and then Extinguish the power that generated images hold over people because fake pictures don't matter in the slightest.

  • Especially egregious because the images are basically the same as the original photos, they just used controlnet to alter the details. It shouldn't be that difficult to stop this kind of thing, assuming Facebook even wanted to; looks like it is entirely possible to find the original popular post the AI posts are trying to copy to prove this is going on, and people are already doing the volunteer work of tracking it all, there would just need to be a way to report this stuff and confirm it. I doubt Facebook wants to do this though since engagement is engagement.

    • Couple problems:

      • the availability of the tools makes the potential scale of the problem pretty drastic. It may take teams of people to track this stuff down, and there's no guarantee that you'll catch all of it or that you won't have false positives (which would really piss people off)
      • in a culture that seems obsessed with 'free speech absolutism', I imagine the Facebook execs would need to have a solid rationale to ban 'AI generated content', especially given how hard it would be to enforce.

      That said, Facebook does need to tamp this down, because engagement isn't engagement when it's taken over by AI, because AI isn't compelled to buy shit from advertising, and it can create enough noise to make ad targeting less useful for real people.

      I personally think people will need to adapt to smaller, more familiar networks that they can trust, rather than trying to play whack-a-mole with AI content that continues to get better.

      We're overdo for some degrowth, especially when it comes to social media.

      • there’s no guarantee that you’ll catch all of it or that you won’t have false positives

        You wouldn't need to catch all of it, the more popular a post gets the more likely at least one person notices it's an AI laundered repost. As for false positives, the examples in the article are really obviously AI adjusted copies of the original images, everything is the same except the small details, there's no mistaking that.

        in a culture that seems obsessed with ‘free speech absolutism’, I imagine the Facebook execs would need to have a solid rationale to ban ‘AI generated content’, especially given how hard it would be to enforce.

        Personally I think people seem to hate free speech now compared to how it used to be online, and are unfortunately much more accepting of censorship. I don't think AI generated content should be banned as a whole, just ban this sort of AI powered hoax, who would complain about that?

  • I have a few relatives who seem incapable of understanding that miraculously high-definition photos from the 1800s containing never-before-seen imagery of lumberjacks posing with 12ft. tall sasquatches could possibly be inauthentic. .

  • I'm doing a fun art project now, I'm feeling like from the conversations I've been having in this thread about AI, that I'm totally okay with selling these logos as is. I put in inspiration prompts and "in the style of" so I'm sure I'm good.

    https://imgur.com/a/Rhgi0OC

79 comments