I’ve seen a few videos/articles about those, mostly “lol look at what dumb people are doing/commenting on” although, as briefly alluded to in the article, many of the comments are likely from bots and not real 70 year old grandmas saying “God is great” after seeing Jesus made out of snails.
I agree with the person towards the end of the article calling it sort of pedestrian. I think that you CAN make legitimately surreal and interesting images via AI, it’s just not the goal of people posting shit on Facebook to gain bot-driven engagement for whatever social media garbage reason people want accounts with lots of engagement. Like these people aren’t trying to make “art” regardless of whether or not you consider AI imagery to be art. They’re just doing some combination of deceiving and/or trolling so they’re not putting forth maximum effort.
The “legitimate” art is a little more interesting although the few examples images were kind of uninspired in my opinion. “Ice t-shirt on a clothesline” is the sort of thing my Midjourney friends’ Discord channel shit-prompt at night when we’re bored and just chewing through prompt ideas. Also, not to take away from her efforts, but Midjourney does a lot of the heavy lifting these days and tons of photographic jargon don’t really mean better renders. A large part of why her images are better than the stiff on Facebook is because she’s using a paid AI service with a proprietary models and the Facebook “slop posters” are likely using some free AI generator using an old model like SD 1.5 to save on render time/cost because, again, they’re just trolling with it and not trying to sell it.
I posted several similar articles in a different thread, starting here:
The weird stuff is usually memes that start off simple and become more complex as more people have fun with it. And they often mock real trends and memes. Like sob stories about a disabled soldier or a dying child who is very sad and lonely and you must “like” and share the post to make them feel appreciated. And it isn’t just bots, I have one Boomer neighbor who is a FB friend that posts these types of memes all the time. I just looked over his feed to pick a few gems:
(The other type of meme he constantly posts are “like if you recognize this, share if you have used it yourself”, and usually it is a photo of some 1950s to 1970s era artifact or situation, often implying that the Boomer generation is the only one that will understand the photo. Two recent photos are a potato masher and a Coke with peanuts in it.)
Sure, but none of those are a giant sculpture of Jesus made out of radishes or a baby morphing into a dog or similar nonsense.
Edit: To be clear, I don’t mean that ONLY bots are responding to these photos. Even the stupidest pics are probably getting some (sincere) human replies. But the patriotic/religious themed stuff seems to gather a lot of near identical responses very quickly regardless of how inane the content.
In the past few days Bing added a limited “expand” option for its Dall-E 3 generations. It was interesting when I first saw it, but it turns out that it doesn’t simply expand the borders of the image like outpainting or uncrop, but produces a completely new image only similar to the original. Which can be okay, but not of course if you are sold on the details of the original. The one aspect ratio available is 1792x1024, up from 1024x1024. Here are three examples.
The new AI engine Flux has been getting a lot of hype and Night Cafe has been very quick to make it available. I’ve given it a few tries and it can be pretty impressive, including in it’s ability to understand a prompt (it definitely understands better than SDXL, and in my very limited testing was better than Dall-E 3).
Here is a panda riding an elephant in Akihabara
And an elephant riding a panda.
Dall-E 3 understood a panda riding an elephant pretty well, but got more confused on an elephant riding a panda.
The one thing about NightCafe - and I suppose other sites, I’m not sure - is the “enhanced prompt” thing (or whatever they call it) that is now offered for every creation. It’s basically ChatGPT writing out a paragraph based on your prompt. It just makes things too busy when you don’t want it to be sometimes. I’m on the fence whether I like it at all or not.
Dall-E 3 has done that the whole time, without disclosing it or allowing you to turn it off. Dall-E 2 (at least on Bing) did it to a lesser extent, too, adding “ethnically ambiguous” and stuff.
Seven months after the original creation, I’m scrolling through Facebook and see my Jonsey image again. That’s a fresh 50 thousand likes, 1500 comments and 8400 shares.
Holy Schnikes … I just earned a fourth place finish at Night Cafe in the Pixel Art challenge. I wasn’t even trying that hard. I used some pixel art prompts from a different site and just poured in a title. Damn, third place would have been thirty more credits.
Interesting, in Bing/DE3 I ask for a sparkler, party hat, and the letters 2025 without mentioning anything about a holiday but it keeps figuring out that I’m going for a New Yew image on its own.
I dropped a few personal pics into this AI site that makes movies – you can make three per day for free, after that you gotta subscribe, I digress. The results are hit or miss. You can create just off text prompts if you want as well. I digress again.
The first one I did was to turn Bob into a vicious snarling dog (which is the complete opposite of) and got this:
So I wanted to try to make one where he ate my head, but it didn’t quite work out that way:
(those are videos so I’m not entirely sure how this will post, but it should link to my Flickr account anyway)
Bing gives and takes away on how weird it will let you get with prompts. They currently seem to be in a more generous period and I have been able to generate lots of beautifully weird stuff
