AI image generation is getting crazy good

It seems the AIs really like young women wearing a vaguely sad, definitely resigned, thousand-yard-stare look. Her future is bleak; very bleak. Perhaps the AI knows which career she’s going into or is already in, and knows it’s slated to be AIed out of existence soon.

What’s interesting about that to me is I suspect there are very few pix of women wearing that expression out in the wilds of the internet. Ordinary folks don’t tend to be photographed in that state. That look does appear in scenes in films, but only in a tiny fraction of all the frames of all the films it might have learned from. So where did the AI get the idea that’s a neutral expression as opposed to a specific expression?

It’s sure not.

But it did manage to include her purse in there too. I think putting a purse in the image at all is pretty insightful. And absent her legs, that’s maybe even a realistic spot for her to store a thin-ish purse.

But then it stuffed her leg in there too and a close look shows it deformed her right knee to make it fit. And also made her right thigh significantly less tall than her left. Poor thing; she’s got a withered right leg that ends just above the knee. I wonder where her crutches are? But at least that explains the bleak expression. It’s hard for crippled girls to find love.


Pardon my ignorance. What’s wrong with the shirt?

I agree the legend is superfluous. But not goofed up, and certainly a plausible part of any data graphic. At least it didn’t say “red state” next to a blue box and vice versa.

Catwoman:

I did these in Sora back on May 5th (two images in the set were censored)

Summary

Full length profile photo of a confrontation. An anthropomorphic man-bat hybrid and a human sized red robin are standing off against an anthropomorphic woman-cat hybrid and a human-sized penguin. candid, discreetly taken with an iPhone 12 Pro

Another perspective:

In the storyline, she has a lot of trepidation about the new job she’s about to walk into that her father had swung for her. I didn’t give the AI any direct instructions for what expression to put on her face, but I was pretty happy with what it chose for her there. In fact, here’s a redacted transcript of the RP session right before that pic. The stuff I took out would just be confusing without a lot more context.

Summary

Me:
I don’t take the most direct route. I’ve got plenty of time to just cruise.
[in the brand new Audi that her father had given her recently, a gift she didn’t expect]
[other thoughts had along the journey about recent stuff in her life, which was me reminding the AI about what’s important for the story.]
I continue on. I enter the parking structure next door, which serves several nearby businesses and establishments, not just [my new employer]. I’m glad none of my coworkers will notice right away I drive a more expensive car. I park, and then… I just sit there a moment, gathering my courage.

ChatGPT:

Turn 67

You pull smoothly into the parking structure, the Audi’s tires humming softly over the concrete. The place is moderately busy — a mix of early-arriving staff, delivery trucks, a few scattered visitors to other nearby businesses.

You find a spot not too close to the entrance but not buried either, kill the engine, and sit back in the seat.

Your hands rest lightly on the steering wheel, thumbs drumming absently as you take a long, deep breath.

[a lot more stuff related to the machinations her father had done to get her the job]

Outside the windshield, the world moves normally: people walking, laughing, talking on phones. None of them know the stakes sitting behind this wheel.

You sit, still, feeling your heart thud softly in your chest, letting the weight of this moment settle before you finally open the door.

Me:
draw the current scene

(pic you saw.)

Dammit, there’s that “I paused my game” thing again. My outfit was already decided for the day, the AI had drawn the image, and then like 2 turns later the next time I asked for another image, it replaced it with that “I paused my game” shirt I showed earlier, which I was like, NO, dummy:

(( you changed the graphic to something that might not be seen all too well given that she’s showing up a whole day late. ))

So here, I backed up and redid the prompt for the current scene in the car, reminding it of the original graphic that the AI had picked:

New prompt:

(( draw the current scene inside the car, her hands still on the steering wheel, she hasn’t even unbuckled yet. Remember her current shirt she chose is a heather grey, with a graphic of a blue ringed planet on the chest.))

I suspect it left out the shoulder belt so it could properly show the graphic on her shirt. I didn’t waste extra nodes trying to fix that. The whole day late thing was something I had improvised when I realized the AI had screwed up the timeline, and I just went with it. Why not throw another extra challenge at the PC, and the AI. Would it affect what her new supervisor says when she finally reports for work? No, it didn’t, probably because I didn’t reinforce that aspect after that point.


I’d made a point earlier about her quirky purse she carried, as part of telling the AI about her personality. Nice job of it remembering to include it in the image a long while later, but it was part of the context rather than a spontaneous inclusion. My description of her nerdy gamer girl personality also affected the graphics it chose for her shirts along the way too without any extra prompting there.


I thought it was beyond superfluous, but maybe not?

I was referring to the idea that it was an existing comic with chat bubbles changed rather than an AI creation…

Makes sense. Yeah, for that context that is a very good expression, and well-rendered. And her body language is appropriate and well-rendered as well. No uncanny valley look. You (well, me really), just want to reach into the picture and comfort her and tell her it’s all gonna be OK.

Contrast that with the one just above with her facing forward in the car. In that one her neutral expression is mannequin-like. This time she looks resolute, not resigned. Which fits your prompts. But still somehow more dead-like than alive.

What I was commenting on about the resigned thousand yard stare is that in many images I’ve noticed that same look. Not just your one. As if that was some sort of default expression, but an odd one. I now suspect I’m noticing a pattern that isn’t there, and the real reason is the context the humans are providing sets up the emotional state of the characters in the image. And that resigned pathetic (as in “incited empathy/sympathy in the viewer”) look is part of many stories.

Along the way to work she must have gotten breast augmentation too. The earlier picture had her in a shoulder-slumped posture that minimizes the size of any real woman’s chest. Her posture is better in this second pic, but not that much better to account for all the size increase. :slight_smile:

Overall I’d say that for a graphic in a magazine or website article, that particular legend may be Captain Obvious, but a legend, any legend, is also a 100% expected part of data graphics. We’ve all seen and lamented data graphics that really need a legend but don’t have it.

At the same time for sure what I’d expect to see on a page of e.g. Scientific American is different from e.g. Cracked.com.

As to the t-shirt, yeah it represents what I’d call the autistic school of design. Excess rigidity in thinking forces the legend in there when a more holistic “I’m making a t-shirt, not writing for a scientific journal” POV would of course leave it off. Bottom line: it feels stilted, and that’s not a great feeling on a t-shirt. How stilted? Sorta a matter of taste. I too would prefer no legend.


Thanks for indulging my curiosity about this stuff.

ETA: I’ve been revising this post while and after @Ponderoid posted just below. Sorry to invite confusion.

Ah, you noticed that. It was part of her original character description that she had an ample chest. Partly to see how far I can get when telling it to draw those things, and partly because I
wanted it to affect how she’ll be perceived in her new job. I left out that I’d also reminded the AI to fix her chest size in the replacement, which it obligingly did. Without frequent reminders, it keeps reducing her chest size too. At least the AI censor isn’t throwing a fit when I do that, as long as I’m careful how I phrase it.

I may be old, but I’m still male. :wink:

Contrast that with @Darren_Garrison’s bat-man / cat-woman standoff. The AI got the essential human male & female sex markers right, but sure left off the Hollywood or comic superhero female chestage. Then again, @Darren_Garrison didn’t ask for it (that he told us), so what he got was more minimized than maximized.

I’m pretty sure if that’s the case it was an existing comic, part of the humor in even the original was the lazy aspect the canon characters would never engage in.

All 3 of those comic images are gorgeous. The cape in the first one made me think Batgirl rather than Catwoman. What were the prompts for them?

Aside, I see that as a new emerging compliment you can give someone: “Hey, love that image, which engine and what was the prompt?” :slight_smile:

Sometimes it seems like you can get away with a lot if you just ask for a person “inspired” by a character. This is a realistic looking Trucy Wright (From the Ace Attorney game series) and Dora the explorer. Had a little trouble with the thumb wrestling part though.

You’ve got time to quickly fix the squash issue by linking directly to the image file rather than the whole album.

Don’t know what you mean… ( :upside_down_face: )

Thanks. Here’s the prompt: Catwoman looking annoyed flying in the air, about to rescue a cat stuck in a tree. Her speech bubble reads “I didn’t think my job description included this kind of crap”

Midjourney 7 does fine graphics, but it still garbles text, so I have to fix that.

I thought you might have given all sorts of extra style info to it to get those images. Now here’s what ChatGPT did with the same prompt. I’m surprised it allowed the character, but I’ll take it:

ETA: a bit of tweaking, to show what she’s actually up there to do:

Yeah, Midjourney 7 is pretty intuitive. I often like its first generation output as is, before going into detailed style, composition, and camera type prompts for re-generations. Can’t say the same for ChatGPT and others.

Actually, come to think of it, there is one giveaway that this is an AI image: The AI put more effort into it than a human artist would have. A human drawing that page would have copied-and-pasted the figures, but the AI didn’t: Look at the folds on the fronts of the capes, which are different in each panel.

The following picture is the illustration (produced by Google Gemini) for a story where a girl’s reflection develops her own identity. It felt unsettling when I first saw it despite its apparent innocence and cuteness.

Imgur

The new Gemini renderer can make some impressive stuff. I need to try more with it. (Too bad they decided to add a watermark.)