I don’t know if anyone remembers my complete failure at recreating the Slim Pickens riding the bomb image from Dr. Strangelove before.
Bing does a much better job at this but since ‘Putin’ is a disallowed word, all I can put is ‘Soviet Leader.’ (or Russian Leader)
Someone in another thread didn’t have any luck in generating a Tesla Cybertruck so I tried in MJ and got results suggesting an inkling of what it’s supposed to look like. Better than the other results anyway.
And some assorted render results I’ve liked from the last few days just for the heck of it.
Sometimes “resembles” can work in prompts. Not in this case, though. Even “A man that resembles that bald guy from Russia” got a content warning.
I tried misspelling it too. They’re getting wise to all the tricks.
BTW, a few days back I tried “photo of the xenomorph from Alien working as a sign language interpreter at a presidential press conference” and got a good Trump and a terrible Obama.
I’m abusing the MyFace training I set up. I’ve made out with Taylor Swift, tanned with Elle Fanning, ballroom danced with Natalie Portman, I’ve been a king and a googly eyed monster, but for some reason, I really like this one, even though it isn’t perfect.
I can’t help it. Here are some more. That previous one was The New MVP Quarterback for the New England Patriots.
Dancing with Natalie:
As Indiana Jones and the Magical Weed Garden:
Googly Eyes:
As The Mona Lisa:
BTW, I recently saw a comment that was rather enlightening. Someone mentioned that prompting other systems was like doing a key word web/database search but prompting Bing/DE3 was like writing a narrative. I had already been leaning in that direction on the Bing prompts, but having it pointed out to be conscious about has helped in how I craft prompts. (Folly’s link above is an example of that sort of thing. It is a paragraph, and is probably actually mostly meaningful to the GTP interpreter in Bing/DE3 when it would have been mostly superfluous to Stable Diffusion.)
They look better than the real thing (which isn’t saying much, to be fair.)
Bing basically guided me to that format when I saw how it modified the prompt based on follow-up chat.
Making some billboards. Apparently World of Wombats has a Chinese franchise.
Bonus:
A billboard for the tourist attraction World of Frank Zappa and the Founding Fathers
Here’s a similar tool in Dall E 3 via Chat GPT
Looking over old prompts I stumbled on this again, decided to try it with Bing/DE3
The HBO series The Gilded Age has a cool scene of Thomas Edison flipping the switch on his power station, turning electric lights on for the first time on Pearl Street in lower Manhattan, on September 4, 1882. It was a momentous occasion that transformed the city, then the world. Although the show scene was awesome, I don’t think it was historically accurate (for one thing, it took place at 3PM, but in the show it took place at night with a large crowd of onlookers).
I don’t believe there are any photos of the real event, so I thought I’d have Midjourney take a crack at it. I don’t know if any of these are historically accurate either, but interesting nonetheless:
The same prompt with Bing:
Looking for that million dollar idea (see a other thread) so I can get a set of 8 GH200s…
Never mind where I’ll find the local copy of GTP4 and DE3 to run on it…
There are open source models you can run locally with much less hardware. A MacBook Pro with an M2 chip can run the smaller models at a pretty good speed.
Some of those smaller models are approaching GPT-4 levels of performance.
AI is moving incredibly fast.
The models aren’t interchangeable. I want the half-million dollar rig to run pirate Dall-E 3 because I want the results that come from.Dall-E 3, not the results that come from SDXL.
This is pretty unexpected: Stable Diffusion 1.6.