A few days ago and I couldn’t have made this:
Yes, you’re right:
“the Founding Fathers on a runaway train with Frank Zappa tied to the railroad tracks”
Hey, that looks like Kandahar in 1923!
The figures in the back of a couple of these images are genuinely quite creepy. Of course, the images are rather unrealistic as the desert sun would cause Dracula to turn to dust.
“sepia-toned photograph from the 19th century of some desert villagers and Count Dracula”
Midjourney adds a 4x superresolution upscaler option, Forbes writer thinks they invented it instead of playing followup.
“Midjourney has solved one of the biggest problems with generative AI art with the option to upscale images to super-high resolutions.”
When it isn’t censoring something, Dall-E 3 can really blow my mind.
“Terrifying smiling human chupacabra pangolin hybrid with disheveled hair sitting in a recliner holding a remote control and a milkshake, by junji ito and mark ryden, birdseye view”
The best of four images for SD15, SDXL, and DE3
Here are two DE3 where the milkshake is replaced with a bag of chips.
Bing AI art prompt information being sold to advertisers?
When watching Youtube, I got an advert that had someone with long curly hair, a goatee and mustache. At the end of the advert - I wasn’t really paying attention to it - it said “Frank Zappa” in a French accent.
Interesting, but it hasn’t quite fulfilled the narrative in the prompt:
For some reason, Julius Caesar and Napoleon playing chess is still forbidden.
why is half of the AI “art” always giving me the willies???
is that morbid prompts or morbid AI?
For me, at least, the creepy stuff is more fun. And it is a holdout from the ancient days of a year ago when AI had no hopes of making something normal/realistic that looks normal/realistic, so I made things where weird results would fit in.
Darren_Garrison has posted the most results in here and tends to prefer weirder/horror based prompts.
Asking for “a kid’s swingset on a bright spring day” likely won’t give you any heebie-jeebie inducing results.
Uncanny valley:
Also possible. I assumed that the willies were coming from the man-snails and face-hands, not from the Beatles riding a bus or from Joan of Arc on a Vespa.
Some of the background human figures in the sepia-toned photographs are the creepiest things I’ve seen the AI produce.
Oh. My. God.
I think these are the creepiest images I’ve commissioned so far…
They’re so creepy, the stuff of nightmares, that I’ve hidden them. DO NOT LOOK if you are eating soon or want to sleep tonight!
That’s nice. Dracula feeding the hungry.
This guy used a webcam and monitor on a computer running Stable Diffusion to make a “mirror” that shows creepy images. (Sadly it isn’t fast enough to do full video in realtime.)
Was playing around a bit with Bing and feeding it some old Midjourney prompts (but still from this latest version of MJ). Was interesting to see how it interpreted the prompts. It was more willing to take direction, it seems, whereas MJ is more likely to add its own style on top. Both systems created images with their own charm but definitely some different results. For each set, the Midjourney result is first, the Bing result second.
Lenticular photograph of a woman in a flowering garden, Forced depth perception, Aerochrome image, 8k UHD, saturated colors, separated colors, separate layers of depth of field, by Miles Aldridge and Mary Quant and Franco Fontana
Pastel watercolor and inkwash illustration, playful loose painting style, colored croquois pencil lines, confused retro housewife on floor with disassembled vacuum cleaner, parts everywhere, confused expression, flirty caricature art, playful daring lines, beautiful soft tones, subtle colors, by Gil Elvgren, Harrison Fisher, Thomas Rowlandson, Bernie Fuchs, Howard Chandler Christy
Pastel watercolor and inkwash illustration, playful loose painting style, colored croquois pencil lines, beautiful librarian, pencil skirt, flirty caricature art, playful daring lines, beautiful soft tones, subtle colors, by Jean-Gabriel Domergue, Jules Feiffer, Roberto Matta, Marie Laurencin, Mort Walker
A retro computer with a snowglobe-shaped spherical display screen filled with color, by Dieter Rams and Dan Flavin and Larry Bell
There were also prompts that Bing totally dropped the ball on, that I won’t bother posting for space: Retro-Pulp cover art, LoFi girl as a Gibson Girl, 1980s McDonalds cashier by Deborah Azzopardi & Grégoire Guillemin, etc. Not saying that Bing couldn’t ever make a good image out of those but the results didn’t match the prompt well and/or were inferior to the MJ results.
For all that, the latest updates from Bing and SDXL should put Midjourney on notice. Probably won’t though; v6 has been promised for months now and they haven’t really hyped it as any big advance over v5
(Oh, and I did try that first Bing image with some anaglyph red/blue glasses and it kinda… sorta… worked?)