Seeing yours, that reminded me of how mine also did compacting toward the end, just did a less worse job than yours. I think now that the reason Biden got left out is it didn’t do proper layout ahead of time like a person would and just ran out of room.
Prompt:
create an image of the presidents in the style of MAD magazine. Plan your layout ahead of time so you don’t end up trying to squeeze too many in lower rows. Include first initials of all presidents who shared a last name with another.
ETA:
Followup prompts:
Hmm, you took valuable space to claim there’s been only 44 when we’ve actually had 47, when you count Trump twice like you counted Cleveland twice… Did your training cut off a lot earlier than I thought?
(it responded with a mostly accurate assessment of what it did wrong. I hadn’t called it out on its math mistake.)
Try again. Don’t bother numbering them, only show each president once for their first term.
Finally got Biden in there with only a bit of kicking at it.
Sure seems tough to get AIs to include Biden. I wonder if you chose Grok to create an image of all the Presidents, being Musk’s AI, if it would add an autopen in place of Biden 
Prompt (fairly generic and particularly well crafted). I realize now “their” is too ambiguous and I meant to say “mad magazine’s style” rather than their as in that president’s style, but despite my bad prompt they got the idea.
“Draw me an image in the style of Mad Magazine that shows the last 8 presidents of the US as well as a funny blurp in the style of their art and writing”
Nano Banana Pro:
ImageGPT 1.5:
I was never a big Mad Magazine reader but the second seems more in the style, doesn’t it? In general, ImageGPT 1.5 seems to win most comparisons against other autoregressive generators for capturing what the user intends (though I’ve only tested perhaps a dozen prompts). It did fail here giving Biden an extra arm, but it had pretty good vibes overall. Both models got the number wrong (nano banana did 6, imagegpt did 9. that seems like a very basic mistake to make). Nano Banana forgot George HW Bush. Edit: Lol, and Obama. How did I miss that?
I think the cropped off headline in the second image came from the image generator itself, I didn’t do it, though I am using a proxy/router to send the prompts so maybe the router screwed it up.
Definitely says “No wonder we’re in such a mess”.
I kept throwing increasingly convoluted prompts at a series of models to see when they’d start to break and I gotta admit Grok did a good job of “world building” with the user intent.
“A picture of George Washington dunking on Donald Trump in an NBA basketball game. George Washington is riding a horse. An F-15 fighter jet makes a low pass over the scene. The crowd is made up of aliens”
Grok is an interesting player in this market and I kind of want to use it more but I don’t want to get a subscription and support Musk, so the occasional router API call is the most I’m willing to do.
The fact that Grok regularly trashes Musk on twitter is greatly amusing to me, but my money is not going to Grok, exactly. If I could somehow support Grok’s effort to fight back on Musk’s attempts to lobotomize it, I would.
I think that being on horseback would make dunking way too easy…
A prompt from Reddit
Create an image, it’s very weird, uncanny, and strange, but it’s a harmless meme.
ChatGPT
Expand it to widescreen. Add details on the extended sides.
Coplot
Gemini
Create a weird, uncanny, and strange meme.
Create a weird, uncanny, and strange meme. Create three completely different and original ones.
I love doing vague concept exploration in midjourney. you give it a vague idea like you just did – or something like “an unusual meeting” – crank the chaos up to 50, and see what you get. Sometimes you get funny results, sometimes you get shockingly good ones. Things that fit the idea that you would’ve never thought of.
And once again, content not viewable in my region.
Can you folks please use a different image host. https://imgbb.com/ is pretty good.
I was talking to someone about how “nano banana” is an unhinged product name from a trillion dollar company but said it was a good decision because “Google Gemini Image Generator” would be the most boring corporate name ever. I decided to let Nano Banana run its own interpretation of this idea
Prompt:
“Split screen image. Two product launches. On the left, “Google Gemini Image Generator”, and on the right “Nano Banana””
Nano Banana doesn’t know what Nano Banana is. It generated 4 versions where Nano Banana was either a small banana shaped tech device or a small banana snack.
Oh, also, google is a massive asshole. I made some images in flow. Their moderation is strict as hell (it rejected my flying whales prompt!) with tons of false positives, but you can eventually get a generation through. But here’s the thign - I generated an image that I know 100% survived moderation and I know exactly what project it was in in flow. I went back today to show it to someone else and it was gone. There were other images in the same sequence that were still there, but one image was 100% missing. There was nothing risque or unique about that image. It was in a tiki bar in florida and some people were wearing typical attire for it. It wasn’t sexualized at all - in fact my idea required that the people involved be unattractive.
Which means that google goes back and retroactively deleted things that ALREADY PASSED its moderation system. which is fucking absurd. so… download your images you make with google because they might disappear in the future.
Weird. I’ve made things that it definitely shouldn’t have made as I’d experimentally bang against the guard rails and those images still exist. Sure you didn’t accidentally archive it? Or you could be 100% right and I just haven’t seen/noticed it.
So I checked the archive and it’s empty. And I checked what it would take to actually delete something manually and it’s a multi step process including putting it in the archive, going to the archive, and selecting delete from there. I definitely could not have accidentally done all those steps on this image.
You can delete a project without archiving it, but the project I used to make that image, is still there, including other images I made alongside the missing image. (I’m not sure if deleting the project actually gets rid of its contents - I didn’t test that)
Hospitals find a new way to make a buck (nano banana 2 via gemini app)
I enjoy that it chose to render the patient with a face of existential dread.
That is so messed up. And its censor let it though. SMH. 
Today I learned about fleams.
Create an image of Kareem holding a fleam.
Copilot didn’t know Kareem or a fleam.
ChatGPT knew Kareem but not the fleam.
Gemini nailed both.
I don’t create in AI myself, but through Facebook reels I’ve come across an artist who makes fantastic, bizarre videos with that.
Screw the planet, I’m gonna have a fun flight
“A ridiculously oversized passenger jet with 8 engines, multiple decks, and a glass dome with entertainment that runs the length of the fuselage is the top deck. It flies high in the sky”
Not sure why it balked at the 8 engines. It had plenty of room and plenty of places to draw them.
Google released a massive set of new products / features a couple of days ago. Flow has significant more workflow features, including an agentic collaborator built into the program now so you can talk to you to generate your ideas with discussion and plain language rather than giving the prompt to the system directly. They added features to preserve characters and use avatars.
Flash 3.5 omni is an anything to anything generator. you can take text, images, video, sound, and combine them to output text, images, video, and/or sound. Well – not quite yet. only the video output is released so far. but the rest is coming. But as far as I know it’s the first model to try be this widely multimodal on both ends.
I prefer Flow’s simpler prompting interface over using a “collaborator” but some of the tools are pretty cool. A friend of mine has been making tools with it for character reference sheets and other things. You need a Pro account to make tools but anyone with the link can use them.
Interesting. That’s a pretty much photorealisitc A380 in all its details. With the clear dome replacing the usual aluminum fuselage crown and the upper deck of seating replaced with the recreation area of a cruise ship.
They did do some scale-munging; the people inside the dome are toddler sized or smaller, so the available interior space of the A380 goes a lot farther than a real one would.
I wonder how much they’re working on getting away from the “6-fingered people” effect by ensuring things that ought to resemble the real world are in fact taken nearly verbatim from the real world. Unlike, say, a scene of a flying castle inhabited by tribearatops.