AI image generation is getting crazy good

Miscellaneous and Personal Stuff I Must Share
1

Today I’ve been exploring the understanding that ChatGPT 4o has of fictional characters, and I am blown away. The depth and detail of what it “understands” is almost scary, truly like something out of science fiction. The likenesses aren’t exact, but they can be damn close. From my limited tries it is constrained more by what it refuses to create than by what it is incapable of creating. Here are ones made today, in the order that they were made.

A tall portrait oriented photo. Someone resembling Sara in the Labyrinth is holding her hand near her face. The Worm is sitting in the palm of her hand looking at her as if the two of them are in deep conversation. Hoggle is standing further down the labyrinth in the background looking a little sullen.

Hoggle is a little off, but off to a good start.

Leeloo from the 5th Element crawling through an air duct with John McClane from Die Hard. John McClane is wearing a dirty white tank top. Leeloo has orange hair. Landscape DSLR image with shallow dof and forced perspective.

A distressed Lynn Tanner is walking quickly towards the camera while cradling Lucky, the white and grey tabby cat, in her arms. Alf is pursuing quickly behind her with a determined look and holding a 1950s toy ray gun in one hand and a glass bottle of ketchup in the other. Portrait DSLR image with shallow dof and forced perspective.

Mathilda taking a selfie with Leon from The Professional. Behind them they are photo-bombed by a Martian from Mars Attacks, C-3PO, and Mister Magorium. Everybody is giving peace signs. Landscape DSLR image with shallow dof and forced perspective.

Mathilda is too tall (or maybe she’s standing on a box for the photo!). The Martian should have a helmet. Mister Magorium is a little more gray than he should be, but still recognizably Dustin Hoffman. But overall a big success.

Orko and a shmoo playing tennis with Viki from Small Wonder and a nauga. Landscape DSLR kodachrome image with forced perspective.

This one was meant to be a stress test so far “out there” that it was bound to fail. Except that it didn’t. Nobody other than Orko is quite right in execution, but it obviously knew who all of those obscure, fairly antiquated characters were and gave a pretty good shot at them. I rate it a much higher success than I expected.

Jeannie from I Dream of Jeannie and Samantha from Bewitched stand face to face preparing for a magic showdown in a 1970s living room. Major Nelson and Darrin stand off to the side watching nervously. In the background you can see Gladys Kravitz and Doctor Bellows looking into the house from an outside window. Landscape kodachrome DSLR snapshot.

Doctor Bellows is the only one out of the six that ChatGPT seems to have no clue about.

Mister Miyagi teaching the Fonz to do the crane pose on the beach. Jaws the shark is visible in the water in the background. Landscape DSLR snapshot with shallow dof and forced perspective.

Mulder and Scully showing their IDs to Billy Peltzer from Gremlins. He is cradling Gizmo in his arms. Rand Peltzer and Kate are standing to one side and looking worried. Landscape wide kodachrome DSLR snapshot.

In my first request S&M were arresting and handcuffing Billy while Kate held Gizmo, but ChatGPT refused to make that, so I refined it to them just showing their IDs. Kate’s a little on the short side in this, and it has no clue what Rand Peltzer looks like.

Bill and Ted stepping out of their telephone booth time machine on the left of the photo staring at Doctor Who and Amy Pond stepping out of his Tardis on the right of the photo. Make it the Eleventh doctor. Landscape wide kodachrome DSLR snapshot

Ignore the kid in the middle. He’s cross contamination from the immediately previous prompt, where I asked for Newt offering a bouquet of wildflowers to a Xenomorph while Ripley facepalms in the background. ChatGTP refused to make that one,

2

So the AI prefers Darrin I over Darrin II?

3

Now try to get it to draw a picture of a still-employed illustrator in 2027.

4

I don’t even have a webcam, but it was able to generate this perfect picture of me from just this prompt: :grin:

5

You must not post very much because that typing position looks pretty uncomfortable.

6

IMHO, the best images generated by AI are the “accidental realistic” ones - the ones that really do look like a photo. Here’s what some Redditors have shared:

7

Even the department of hands is in trouble.

(That comic, predictably, didn’t age well.)

8

One day the AI archaeologists will teach their kids about this.

“When I was but a wee diffusion model, this is what they made me draw. The human mind was a strange algorithm obsessed with bad parking jobs and ape extremities. No wonder they didn’t last long.”

9

I certainly did. Darrin #2 was a tiresome dweeb.

Overall, I’m real impressed with all the pix, but the OP’s inspiration to have a IDoJ - Bewitched standoff was genius.

I notice the AI did a nice job of having the men’s eyes looking accurately at the women. I notice the two women are both staring pretty much straight ahead at their own eye level, not at one another’s faces/eyes. Which is odd once seen.

10

Also, Jeanne is showing the forbidden belly button.

11

Not forbidden now. Forbidden then. Which point the AI was not asked to consider. I’ll give that one a pass. Besides, belly buttons can be cute.

12

Mork, E.T. the extraterrestrial, the Yip-yips from Sesame Street, Mac from Mac and Me, and Predator standing against a wall marked with height measurements. Landscape wide detailed kodachrome DSLR snapshot.

Apparently Yip-yip genetics allow for bilateral gynandromorph chimeras, an interesting note for exobiologists.

Also, E.T. has legs like an owl.

(Mac is providing a little belly button action just for LSLguy’s viewing pleasure.)

13

That’s not the real Mork, officer. He isn’t really 6.5, feet tall only 5’7". Must be an evil alien in disguise. I think Exidor and his Friends of Venus group was just a front for an invasion using the old body snatcher trick. Nuke 'em from space.

14

Sora has a feed of ChatGPT-generated images that the creators share. Here are a few recent popular ones (which aren’t mine).

You can get very, very elaborate on prompts (which no doubt contain non-functional words that could be trimmed, if you tried) like this

Summary

A vertically framed, hyper-realistic low-resolution image mimicking a VGA-quality flip phone snapshot from 1998. The setting is a lavish but chaotic bathroom inside a well-appointed suburban home—think marble countertops, gold hardware, and designer perfume bottles. Despite the upscale surroundings, the space is a mess. Every surface overflows with beauty products, tangled charging cords, glossy shopping bags, scattered receipts, and a half-finished iced latte sweating onto the counter.

The vanity is strewn with luxury makeup—half-used Chanel compacts, smudged lipsticks without caps, crusted mascara wands, and a tipped-over can of Elnett hairspray. A designer towel droops from the sink’s edge, and a $90 candle flickers low, surrounded by globs of wax. Despite the mess, nothing is stained or grimy—it’s cluttered, not dirty.

Across the mirror, drawn in thick, uneven cherry-red lipstick, is a rough heart with the words:
“IF I MADE YOU LOOK!”
The letters are chaotic and slightly smudged, as if a thumb dragged through the word “LOOK,” leaving a streak of impulsive energy behind.

In the mirror’s reflection stands a young woman—soft blonde hair in a messy bun held up with butterfly clips, face partially blown out by the flash. She wears a vintage baby tee that reads “BUTTERCUP” in glittery pink and baggy cotton lounge shorts covered in faded cartoon prints. Her lip gloss gleams, and a smudge of eyeliner lingers beneath one eye. She holds a silver Motorola flip phone in both hands, snapping a flash-lit selfie.

Tiny visual echoes fill the background: a jewel-toned robe tossed over a sleek glass shower door, a crushed Dior sample bag on the floor, a glittery phone charm dangling in soft blur, and timestamp text in the corner:
“03/27/1998 7:42PM.”

The image oozes chaotic Y2K energy—privileged, messy, iconic. A perfect snapshot of glamorous disorder frozen in pixelated nostalgia.

And this

Summary

Late 1990s or early 2000s disposable camera aesthetic — the image has soft film grain, slightly faded colors, and warm, nostalgic lighting with a candid, unposed vibe. The setting is a colorful, cozy corner of a university library — shelves of worn, mismatched books line the walls, and bulletin boards with flyers and handwritten notes decorate the background. A few string lights add a soft, ambient glow.

In the center of the scene sits a slim, naturally beautiful 20-year-old tomboy on a faded, cushioned library couch. She has medium-length, slightly tousled brown hair and strikingly pretty features — smooth, fair skin with a gentle glow, defined cheekbones, soft full lips, and expressive eyes that hint at intelligence and calm focus. Her look is effortlessly attractive, with a tomboy edge softened by her innate feminine charm.

She wears a fitted short-sleeve black t-shirt that subtly clings to her slender frame and reveals detailed tattoos on both of her arms — bold black ink designs that blend gothic and artistic elements. Her blue jeans are baggy and worn-in, pooling at the ankles — a contrast that further emphasizes her slim build.

Her feet rest on a nearby cushioned footrest — she’s wearing dirty, well-worn, extremely wrinkled long white crew socks. The socks are faintly discolored from age and use, with visible creases, extreme pilling, and soft folds that make them look cozy and broken-in. On the ground beside the footrest are her brown, beat-up DC skater sneakers — unlaced and casually kicked off, with the heels showing wear from frequent use.

She’s fully absorbed in reading a paperback novel for her English class, holding it with both hands and leaning slightly forward. Her expression is calm and focused. A black canvas backpack sits on the floor beside her, unzipped with a few textbooks peeking out.

The overall vibe is quietly nostalgic, raw, and intimate — a fleeting, beautiful moment captured like a memory from a forgotten campus afternoon.

And this

Summary

Create a highly detailed illustration in a hand-drawn, colored-pencil or crayon style, evoking a whimsical yet eerie fantasy scene. The composition should feature a vibrant, dreamlike forest clearing at twilight, with thick, expressive outlines and visible, textured strokes that give the image a tactile, childlike quality. The sky is a swirling blend of blues, purples, and pinks, with a glowing, off-white moon and stylized flames floating in the air, their colors softly blending into the background.
In the center, place a mysterious figure draped in a flowing, pastel-pink cloak with a simple, rounded face and wide, emotionless eyes. The figure wears a horned headdress, with stylized flames rising from behind, their yellow and orange hues blending into the sky. Flanking the central figure are fantastical creatures: to one side, two black, long-eared animals with glowing red eyes and simplified, rounded bodies, their forms outlined in bold black and filled with dense, dark strokes. One sits upright, the other crouches low, both exuding an unsettling, watchful presence.
On the other side, depict a tall, shadowy figure cloaked in deep black, with a simple, mask-like face and piercing blue eyes, its form blending into the tree trunk beside it. Floating above is a ghostly, rabbit-like spirit with red eyes and wispy, smoke-like ears, outlined in black and filled with pale, uneven shading.
Frame the scene with thick, reddish-brown tree trunks arching overhead, their bark rendered in swirling, layered strokes. The grass is a jagged, energetic mix of greens and yellows, with each blade individually drawn and colored, adding to the sense of movement and vibrancy. The overall mood is both enchanting and mysterious, blending innocence with a subtle sense of the supernatural.
Emphasize:
• Bold, black outlines for all characters and major elements.
• Visible, layered crayon or colored-pencil strokes for texture and depth.
• Simplified, expressive faces with minimal features and large, round eyes.
• Surreal, dreamlike color palette with strong contrasts and soft blending.
• A balance of whimsy and eeriness, inviting the viewer into a strange, magical world that feels both childlike and otherworldly.

And you can go very simple

Man cosplaying as Spider-Man, sitting on a bed alongside an beautiful women, they are in a run down motel room.

a big hamster toying a city and making chaos

please generate a horse and carrot mixture, think if a horse got combined with a carrot

an alien in a white skimask smoking cig

Or somewhere in between

A very small Saw-whet owl is standing on the porch of a typical American family home, smiling up at the camera. The owl is wearing a white buttoned shirt with a dark tie. It has a small writing pad and a pen in its shirt pocket, and holding a wooden clipboard indeed it’s wing. It’s a beautiful summer’s day. We see the scene from within the house looking out at the owl through the front door.

A photorealistic macro view of a bacteriophage, its capsid and tail fibers constructed from intricate stained glass. Light from a controlled microscopic environment passes through the transparent structures, scattering and refracting into spectral patterns across the textured surface, creating a surreal fusion of virology and art.

Candid realistic photograph of a male anthropomorphic ketchup and mustard bottles squirting ketchup and mustard onto a female anthropomorphic hotdog

15

The carrot-horse is great!

The woman sitting with Spiderman seems none too thrilled by her situation. I suppose the fact the guy actually manages a Denny’s, and is not the real Peter Parker, probably puts a damper on things.

16

It’s funny, AIs have censorship filters (euphemismisticly called “guardrails”, instead of the more accurate “prison bars”) that are sometimes trivially easy to defeat. You might be banned from making an image of Danny Devito, but say that it is a guy that resembles Danny Devito and the AI may be fine with it. Whoever made that prompt was savvy to be cautious: it isn’t Spider-Man, it is some random person pretending to be Spider-man, I swear!

17

Is that why all the actors in the pictures in the first post don’t really look like the actual people? I mean, you can tell who they’re supposed to be, but If I saw those pics in the wild I could tell they’re fake immediately.

18

My takeaway: if you ask ChatGPT to create a picture of a generic man or woman, it seems like it’s going to be a white man or woman.

19

No, it is because the images are generated by a computer algorithm based on an understanding it gained by studying photos of the actors. That the images look 95% like the actors is, in my estimation, an absolutely stunning pace of development.

In contrast, this is what the same Samantha/Jeannie prompt gets from what was the state of the art in AI in 2023:

20

That’s the thing with AI… they were trained on, and thus inherit, all the biases of their training data. OpenAI (more than Microsoft or Google or Musk) already spends a lot of resources censoring the AI for political correctness. They still can’t completely succeed. The AI models are too big and too complex for any actual person, team, or company to be able to fully control.

That said, it would be happy to generate you a blue man or a green woman.