From "AI can now create any image in seconds, bringing wonder and danger" (WaPo).
DALL-E is now open to all, without a waiting list, so prepare your requests. I'd put my name on the waiting list, but now that I can waltz right in along with anybody and everybody, I can't think of anything I want AI to turn into an image for me. A few weeks ago, I was interested in a TikTokker who was using DALL-E to generate images. I clicked the "follow" button and caused him to come up in my scrolling, but in the last couple weeks, as soon as I see him I immediately swipe him up out of my view. I just don't care what the computer coughed up. It was pointless weirdness.
But if you get the right request, it might be vaguely intriguing:
Posts
40 comments:
I was just watching a Corridor Crew YouTube last night that utilized DALL-E. They uploaded snapshots to a companion program that translated the photos for DALL-E, which then created great likenesses done in whatever style CC asked for. Like them, I found it thrilling to consider gaming and animation possibilities that could star . . . Me. Me as classic Wonder Woman. Me as Commander Shepard. Me as Kratos. (No, haha. That's the one CC made and it was excellent.)
I’ve been having a lot of fun with DALL-E, but it is difficult to control. If you have something specific in mind, the app might not be able to execute it. You have to be open to the spontaneity of the thing. The “mistakes” can lead you in a newer better direction. It’s a very fun tool.
Except when most people hear 'avocado chair' I can pretty much guarantee they aren't thinking of an actual avocado put in blender with a chair. They are thinking of a chair colored in a hideous green shade, and the context is going to have to make clear if that is not the case. There is no 'triggering of a concept that should be exactly the same' in *all cases*. The meaning of each instance of an 'avocado' is highly context specific.
They are slow pedaling this stuff to inure people to the threat.
The window for the Butlerian Jihad is closing.
That avocado chair is pretty good...as good as what I see in museums.
Pablo PicAIsso...
The furry teacup remains a work of art to be venerated for its novel juxtaposition of texture versus function, of course. I bid $150 for the avocado chair. And I will wait for the call from the Met.
Cool app, thanks for the link. My first few attempts were underwhelming, but I'm going to get the hang of it eventually.
I was picturing a wrinkly dark green bean bag, which avocados can impersonate if they are ripened too much. That output looked like something my 3rd-grade self would have done to combine attributes. I don't know if that AI is scary or laughable.
If you understand how these AIs work, you will understand how far from "understanding" this is. 2D images are embarrassingly easy for computers to work on--they are literally grids of integers.
In layman's terms, these programs do nothing more than change a pixel and check the effect on a reward function. If the reward goes up, keep the change, goes down, reject the change.
Beyond that, there are complex and ingenious optimizations such as adversarial networks, to train and speed up the process. But there is absolutely no intelligence here, no understanding, nor anything that could be mistaken for intelligence by a person familiar with the mechanics. These are advanced hill-climbing algorithms.
Midjourney is another AI art program that was used to interpret lines from Mr. Blue Sky.
So AI can now do "art" at the level of (who was the soup can guy from the '60s?).
With the proliferation of art, at what point does it stop being art?
Dall-E is good at recreating artistic styles and mediums, so long as there are plenty of accurately labeled examples available on the internet. It’s eerily good at mimicking the style of R Crumb, for instance. “Mona Lisa in the style of R Crumb.” You might also add other terms to the prompt, to reinforce the effect you want. “1970s, underground comix, Mr. Natural, Fritz the Cat,” and so on.
You can also upload your own photos and, by going into edit mode, erase part of the image and ask the AI to replace what you eliminated with whatever you wish. It often does quite a good job of, for instance, changing people’s clothing or eliminating unwanted elements from a photograph. (That’s called in-painting.) It can also, of course, add fantastical elements. A sea-monster in the lake.
Additionally, you can use what’s called out-painting to ask the AI to add elements to an image outside the existing frame, extending the image in often-convincing ways. OpenAI’s striking example takes Vermeer’s Girl with a Pearl Earring and expands it to panoramic size, revealing her imagined surroundings as they might have been painted by Vermeer.
Personally, I prefer MidJourney. It’s less buttoned-up. And Stable Diffusion can run on a private machine with a decent GPU and doesn’t censor prompts or output. Not that anybody here would create NSFW AI pictures.
I am reminded of all the articles that I never understood that tried to teach me that for topologists or topographers everything was like a donut.
I guess for AI, as for topologists and topographers, avocados and chairs are the same thing.
Looks like a child's potty trainer sitting on its side.
Midjourney is another AI art program that was used to interpret lines from Mr. Blue Sky.
Odd the way it cooks up scary Haloween-y images for "hey hey hey."
Walnut chairs are good for the back.
My first real success, as in I would hang it if it was real, used the search terms "Edward Hopper style painting of a baseball game at Fenway Park". 2 of the 4 results were really great.
I tried to create "Donald Trump selling pizzas", but it responded "It looks like this request may not follow our content policy." Hm.
I tried to create "Donald Trump selling pizzas", but it responded "It looks like this request may not follow our content policy." Hm.
I tried to create "Donald Trump selling pizzas", but it responded "It looks like this request may not follow our content policy." Hm.
I tried to create "Donald Trump selling pizzas", but it responded "It looks like this request may not follow our content policy." Hm.
DALL-E is a tool and like most tools requires practice and skill to operate. Interesting and useful but not the magic that many expect.
Just read two articles today about DALL-E's failures:
See https://spectrum.ieee.org/openai-dall-e-2
Basically, it doesn't understand letters/words, scientific concepts, or faces.
For a more humorous take on AI's failings: https://www.cracked.com/article_35364_5-hyped-up-ai-stories-that-secretly-make-ai-look-really-dumb.html
alien about to attack nightmare scenario
(I think Blogger did that one)
impressionist cobalt blue
angel of death
AI robot armageddon
blogger eating itself
so far ninjas are a problem
beautiful blonde woman ninja
Ninja assassin Monet painting (2nd one is cool)
scary pirate raiding a ship was kind of a disaster
R2D2 on death star somehow ended up on Tatooine
death star, bitch, death star!
worse genie ever
Dr. Strangelove is kind of weird
Can you guess who this is?
They're really censoring the AI robot.
Nothing sexual. Nothing political. Nothing violent.
When I do Madman breaking out of a prison I get a bunch of French mimes
Surreal seems to work better than realistic
Should I be worried that AI can't count?
Van Gogh painting of six penguins
I really like Monet painting of Jack the Ripper
Krypto the superdog painted by Hopper
Krypto the superdog painted by Monet
Krypto the superdog painted by Rembrandt
Krypto the superdog painted by Edvard Munch
(I don't think AI knows Krypto the superdog)
I am running out of free credits!
Degas painting of Albert Einstein in outer space
my favorites so far...
Rembrandt painting of Buster Keaton
Picasso painting of Marilyn Monroe
Munch painting of Krypto the superdog
Monet painting of Jack the Ripper
Mr. Blue Sky at 10:08
thanks Bill!
that was super cool
From a link I sent Althouse last Sunday, which she' evidently ignored in favor of some WaPo puff piece. Oh well!
DALL-E depends on art already made, on photos already taken, aesthetic assumptions statistically derived, and a language—our language—formed over the centuries by acts of communication innumerable about whose nature the great machine knows nothing. It’s largely a cultural-mining operation with a clever assembly line on top. It’s perhaps no coincidence that its name refers to a famous practitioner of surrealism whose work was distinctively suited to reproductions (and look alike variations) on a singularly large scale.”
2 more
flying rats versus ninja assassins in the future
giant eyeball over Tokyo
Post a Comment