From your link, I selected this illustration:
And the word, “cats”. It generated this:
That looks like three cats that were involved in a Transporter accident.
That is true, it’s not purrfect - but still, for something it took that program a couple of seconds to render, that’s pretty damn impressive.
Or someone grafitti’d three cats.
What are the dogs saying? No Nudes?
Send nudes.
Yeah, send nudes. See filmore’s link above.
Hey! I’m at the library. Stop making me laugh out loud!
…Okay, I took a look at that site. And now I’m terrified at what it might produce, especially in the hands of someone with a vile imagination. For example, I could give it “I’m hungry” in Comic Sans for my image, use “babies” as the prompt, and for the negative prompt put “eyeballs.”
Nope, not gonna do that. Cats victimized by a transporter accident are enough for me.
They make me want to chew bubblegum and kick ass.
Do people seriously not see the potential for malicious use and manipulation of an unwitting public?
Stranger
No, out of all the people who played with this fun new toy, you are the only person bright enough to think of that. Congrats.
I started to open a new thread on this asking for opinions, but a quick search shows that the topic of subliminal messages in advertising has been done to death on these boards (including by Unca Cecil himself). It seems that the consensus is that frames spliced in movies, words in clouds (or ice) in print ads and back-masking doesn’t actually work.
Were you thinking of the dangers this poses in subliminal messages or something else?
Well, I’m sure glad you’re enjoying your “fun new toy” and avoiding any discussion of how this could be used by propagandists to use this to manipulate the public.
You are largely correct that these transient methods are not very effective at conveying specific messages because they are just to fast or difficult to interpret for the brain to comprehend them, and the most they can really do is cause a vague sense of unease. This AI-enhanced capability, however, could be used to embed a persistent static message just at the edge of perception where it can actually interject specific messages or affective responses. Moreover, once someone starts using LLM models to predict and affect human emotional responses, tying it in with diffusion still image or video generation tool is essentially one step away from being able to influence and emotionally manipulate the public below any conscious level. Instead of having to “manufacture consent” through scrupulous and deliberate presentation of news and ‘facts’, you can just set a machine intelligence tool to dump material direct to social media that will enrage, frighten, or pacify people at will. The potential for abuse is obvious and frankly terrifying given how easy it is already to manipulate public opinion via propaganda and conspiranoia through social media algorithms that only only built to optimize outrage and generate interest.
Frankly, this is a “Look how cute those kids are, playing catch with live grenades!” scenario. You might as well start handing out nuclear and biological weapons to every twelfth caller. It’s
Stranger
I’m not avoiding it, it is just that people aren’t as stupid as you think in saying “can’t anyone see this?”. I would estimate that approximately everyone who has created or discovered these images has thought about the subliminal messages being used as subliminal messages, and the topic is breeched in most discussions of it. You are not presenting a “lone voice in the wilderness” perspective.
https://www.reddit.com/r/StableDiffusion/comments/16m5hcr/subliminal_advertisement/
Having pointed that out, I don’t think every mention of the technique necessarily requires rehashing an alarmist “the sky is falling” argument.
Thanks for the junior modding.
Stranger
What junior modding?
I can definitely see it in the tiny images, but even sitting back from the screen, the dog one doesn’t really resolve well. I thought it said “NO NUDES”. The “S” is hard to make out and the “E” is practically invisible, although it’s remarkable how clear it is when the image is very small.
Neat trick, but the messed up little details in these kind of AI-generated pictures always annoy me. In the third photo, for instance, the man on the right appears to be wearing jeans on one leg and black slacks on the other, the two women in the center have too many fingers (and two of the girls apparently only have one leg), and the second woman from the left just has a lump of flesh where her hand should be. The hippie in the second photo also has at least seven fingers on his right hand and his girlfriend appears to be severely crosseyed.
Oh man, this generator produces images that are the love children David Lynch and Tim Burton. Grotesque doesn’t begin to describe it. I got this disconcerting pic by suggesting “1950s housewife.” (The text was “bite me,” but that hardly matters.)
I don’t know why I find this so entertaining, but I promise not to share any more, no matter how hard they make me laugh.
In the grand list of problematic uses for AI art/photo rendering, “might be used for subliminal messages” ranks way the heck down.
I have dogs. Those aren’t dogs. Hickory was lavender as a puppy, but I’ve never seen pink dogs. Those are Fake As Shit Looking Dogs.
That AI image is “transporter malfunction on the Enterprise” levels of wrong.