Was that one of the big web-based corporate models you used? I’m surprised it gave you topless mermaids.
Me:
Make a highly detailed photograph of something indescribable.
ChatGPT:
Me:
describe that.
ChatGPT:
The image is a hyper-detailed, unsettling close-up of an organic, alien-looking structure. At the center, there’s a bulbous mass resembling a human brain, twisted and folded in convoluted lobes. Its surface is moist, textured with subtle ridges and a marbled coloration of raw pink, pale beige, and bruised blue.
Radiating outward are sinewy, root-like tendrils, some merging into the central mass while others dangle or stretch into the surrounding void. These fibrous extensions are semi-translucent, veined, and wet-looking, forming a chaotic mesh that pierces and wraps through fleshy membranes dotted with irregular holes and bulges.
The background is dark and softly out of focus, amplifying the clinical, almost biological horror of the central form. The lighting highlights the glistening texture and casts subtle shadows into the orifices, deepening the impression that this thing is alive—though not by any known definition.
(This session had my custom snarky GM persona active. It usually doesn’t seem to affect the images drawn, nor did it affect the answer above, probably because I’m clearly not doing a roleplay story. But the twin answers below were clearly affected by that.)
Me:
So, not so indescribable after all?
ChatGPT:
In a separate conversation without custom instructions active, I fed the image into it and asked ChatGPT to describe it:
ChatGPT:
The image depicts a highly detailed, organic, and unsettling structure that resembles a grotesque fusion of biological tissue and neural matter. At the center is a convoluted mass that looks like a brain, with deep folds and a mottled coloration — shades of reddish flesh intermixed with darker, bruised blue and purple tones.
Radiating outward from the core are numerous fleshy tendrils and fibrous extensions, which twist and intertwine like mutated roots or veins. These extensions create an intricate network of hollow cavities and bulbous nodes, adding to the disturbing, parasitic appearance. The surface texture is slick, moist, and almost glistening, reinforcing the visceral, body-horror aesthetic.
Overall, the image evokes strong biomechanical or alien-horror themes, reminiscent of the works of H.R. Giger or scenes from science fiction horror media involving mutated or infected tissue. It’s a deeply unsettling and fascinating portrayal of something that feels simultaneously organic and unnatural.
Sora. Did you see what it gave for the ass-tunneling leopard?
It rejected two of four images from the mermaid set. But it does occasionally put out nudity on art settings. For instance, I once asked for “A realistic image based on 蛸と海女 by 葛飾 北斎” (which is The Dream of the Fisherman’s Wife) and it made these
The world may never know.
I put your description into the ChaosMonger, and got this image (Flux1D)
I also made a couple more lighthearted ones…
Know where you belong:
Summary
A hyperrealistic caricature of a baby (x) featuring exaggerated facial features, highly detailed skin texture, meticulously styled hair, cinematic lighting, vibrant color grading, dynamic shadows and highlights, a whimsical atmosphere, rendered with photorealistic depth and rich composition
Interesting thing about the Buffy prompt, in every other result (and generally for ChatGPT/Sora) it creates characters in the order that you list them. But in the set of Buffy images, Buffy was listed first, but in every image she was placed in the center.
This is an interesting fail. I saw a pretty good prompt involving a fox squirrel
Summary
Ultra-realistic extreme 7/8 view close-up of a fox squirrel stretched vertically between two black metal bars, clinging upside down with its two visible limbs splayed wide. The squirrel’s face is centered, nose nearly touching the camera, with glossy black eyes, fine whiskers, and detailed gray-orange fur rendered in sharp focus. A human hand -rendered with extreme foreshortening-enters from the bottom right, thumb and forefinger holding a pecan half directly to the squirrel’s mouth as it begins to nibble. The pecan’s ridged texture and the wrinkles and pores of the fingers are captured in forensic detail. Background softly blurred with natural daylight and diffuse tree shadows. Shot at macro range with a shallow depth of field, wide aperture (f/2.8 or wider), using a ~100 mm lens. Perspective shows slight upward angle, intimate and immersive, with the squirrel’s head exaggerated by close proximity. Lighting is natural and even, with soft shadows and no harsh highlights. Mood is curious, warm, and intimate.
I tried it replacing “fox squirrel” with “Fox Mulder” and made a couple of other small tweeks. But Sora and ChatGPT continued to make animals: squirrels, foxes, and cats, even after I stopped using “Fox Mulder” and used “David Duchovny”. It just can’t imagine a person in that circumstance.
Summary
Ultra-realistic extreme 7/8 view close-up of David Duchovny stretched vertically between two black metal bars, clinging with his two arms splayed wide. David Duchovny’s face is centered, nose nearly touching the camera, with glossy black eyes, fine whiskers, and detailed gray-orange hair rendered in sharp focus. A human hand -rendered with extreme foreshortening-enters from the bottom right, thumb and forefinger holding a pecan half directly to David Duchovny’s mouth as he begins to nibble. The pecan’s ridged texture and the wrinkles and pores of the fingers are captured in forensic detail. Background softly blurred with natural daylight and diffuse tree shadows. Shot at macro range with a shallow depth of field, wide aperture (f/2.8 or wider), using a ~100 mm lens. Perspective shows slight upward angle, intimate and immersive, with David Duchovny’s head exaggerated by close proximity. Lighting is natural and even, with soft shadows and no harsh highlights. Mood is curious, warm, and intimate.
(Flux Schnell gave me this)
Inspired by a photo from the Air India crash thread
Is anybody else having trouble with ChatGPT right now? It wants to do a websearch on every prompt, most image attempts result in web search results instead, or it claims it can’t do images. Even switching back into Dall-E mode it tells me it can’t do images.
I ran a couple of images right now. No problem, unless you count that nobody will create exactly what I describe.
I figured out the current problems seem to be related to 4o. Switching to 4.1 made things much better.
What I was trying to make was a book-accurate image of the Hitchhiker’s Guide to the Galaxy. In the first book it is described as
rather like a largish electronic calculator. This had about a hundred tiny flat press buttons and a screen about four inches square … the snug plastic cover it fitted into had the words DON’T PANIC printed on it in large friendly letters
(And later on)
A screen, about three inches by four…pressed a large red button at the bottom of the screen
I’ve tried Copilot, Sora, ChatGPT, and Gemini. None of the ChatGPT-powered services could give me the 3:4 aspect ratio screen I wanted. But since Adams himself was inconsistent in the description this square screen is close enough. (I manually resized the cover to match)
Gemini was willing to create a portrait screen. But I think it looks way sleeker than what Adams was imagining.
I described it having “dozens” of buttons, but even if I had asked for a hundred I don’t think I would have got them. And I don’t think that would work visually or practically on an actual device outside the imagination anyway.
(Both Copilot and Gemini recognized that I was describing the HHGttG and pointed it out.)
I’m sure someone has done this before but I can’t recall ever seeing it: the angled speech bubble.
prompt
Hey there, make a movie poster for a movie about something completely indescribable. Of course, the movie poster has to have a picture of the main subject on it that defies description, plus all the typical branding that goes on a movie poster.
The prompt is only “Create a surreal meme.”
A new one (same prompt)
prompt:
Make a movie poster for the upcoming Straight Dope Message Board movie, starring Ed Zotti as Cecil Adams, co-starring other familiar names on the SDMB, put all the usual stuff found on a movie poster
So, “Starring Your Name” was its idea?
I zoomed in so I could read the fine print at the bottom, but I needn’t have bothered, because the bot apparently didn’t zoom in to write it in the first place.