AI picture generators

Lots to choose from

My cousin Tim has been going nuts posting AI generated pictures with a holiday theme, displaying movie characters opening presents on Christmas morning.

This is an interesting example of how far AI is coming in such a short period of time.

Notice the tiny details, such as the lamb clock, and night vison goggles, or the robe Buffalo Bill is wearing. The only issue I would have with this is that the background appears to be more 1970s than 1990s (when the film came out). But amazing nonetheless.

I am not sure which online AI tool is being used, so as a test, I decided to look at a few of the more popular online applications, and see what results I can get.

The key is the information you feed into the chat program. The more details you can give it, the more interesting the picture. With that I input the same description into a number of AI prompts, to see what they came up with:

“create a 1990s holiday dinner setting, with Jodie foster (as Clarice starling) and Anthony Hopkins (as hannible lecter) sitting at a dinner table smiling at the camera, with plates of liver and onions and fava beans in front of them, and a bottle of chianti on the table, with filled wine glasses. The table cloth should feature cute cartoon lambs and a girl shepherd design”

Let’s see how each one did…

Google Gemini:

OK, first off obviously that’s not Clarice or Hannibal. Some algorithms won’t allow copyright content, like actual likenesses of real people (or other objections as we shall see). I can only assume that’s what is going on here. But at least the dinner plates had the items as well as the chianti and tablecloth right.

Next, I went to Microsoft’s Copilot:

OK, much much better. Every detail I asked for is presented in this picture. I even love the tablecloth design, was exactly what I was thinking of. Good job!

Next, I went big, with ChatGPT v 5.2:

I also love this one. Spot on with all the details, even adding a bread bowl. My only quibble is that the table is really small, but that’s the fault of my input, I could have been more specific with the table, seating, etc. But otherwise, great job!

But, that was with the latest non-free version of ChatGPT. I used an incognito tab and posted the same exact quote, and I got the following response:

“This image generation request did not follow our content policy.”

Digging further, it listed a number of reasons why it refused to generate the content and provided a number of suggested edits:

• Replace liver and fava beans with:

  • “an elegant gourmet dinner”

  • “a darkly humorous upscale meal”

  • “a stylized, symbolic entrée (non-specific food)”

• OR make the characters clearly fictional look-alikes, e.g.:

  • “a 1990s thriller-inspired FBI agent and a refined psychiatrist, not resembling real actors”

• OR switch to illustrated / cartoon style explicitly:

  • “stylized illustrated characters inspired by 1990s thrillers”

Funnily enough, it was the word “liver” that was the most troublesome, because although it is a pop culture reference (it’s words) it infers cannibalism. Well, yeah. So I changed the description slightly, to replace “liver and onions” with "“an elegant gourmet dinner”, but kept everything else, because it had to have Clarice and Hannibal in it, not just fictionalized characters, because then it wouldn’t make any sense:

Nearly the same, with a slightly different tablecloth (still nice) changed the meal to what appears to be pot roast with green beans, and a nice added touch of a candle. So, almost ok, just not quite as funny.

OK, those are the big three AI tools fighting one another, with clearly gemini coming in dead last. The only redeeming quality with gemini is I loved the choice of seating and table, as they aren’t next to each other but across the table from one another, the only AI to pick that layout.

Desktophut:

Again, not even close. I have no idea who those people are, why the guy’s plate is just green beans, she’s a double fisted drinker and instead of lambs in the tablecloth, they appear to be hand puppets? Can’t even read the label on the bottle. Bizarre AI choices all around.

And finally, Perchance:

WTF??? There is nothing…nothing remotely right. I can’t tell even what the tablecloth design is. I suppose they are going to forage their food with their hands, they appear to be Sears catalog models from the 60s, and the wine bottle….ugh. It’s not even holiday festive for crying out loud.

To be fair, this last one is an open source guy that made this on his own, but he has a ways to go. But these last two were recommended by reddit users, not sure what they were smoking when they linked those.

Now, there are a lot of other more commercial applications you can try that may cost some money, but no guarantees with the results, as we saw here with google’s offering.

So, there you have it, some holiday AI fun! Thanks to Timmah for the idea. Has anyone delved into AI photo generation, or AI in general?