Predictions about the impact of AI-generated imagery
- Thread starter ahalpert
- Start date
A dork in a white shirt in a school lobby with 150 other people?
That'd be one weird emoji.
ahalpert
Major Contributor
So, Dall-E 2 is now available for all to use without waiting on a list. You get 50 generative credits per month for free, and then have to pay for additional credits. It seems like you use a credit when you create an image, and then again when you change it. I have no idea what the exact copyright situation is regarding what you produce on their site, nor do I know how your data will be used if you upload an image. So, proceed with caution. https://openai.com/dall-e-2/
I don't know much about the different image generators out there, so I can't compare it to others or even be certain that I'm using it properly. Some of the showcased results are impressive, and I thought I'd give it a whirl. It's a lot of fun, and a shame that one's use is limited.
I thought I'd start with "an historically accurate depiction of jesus christ sitting atop a rocket ship circling the sun". it spits out 4 results. 3 of them showed JC as a white guy. what's really interesting is that they all shared a kind of cartoonish hippy dippy style that I thought was totally befitting the weirdo image prompt. it's gratifying to see an image spit out that I lack the skills to create but can somehow claim is my creation. but there are so very many decisions that would get made along the way in the creation of something like this that I had no part in. in a way, the process takes more away from me than it gives me. also, when I extended the image downward to make it a rectangle rather than a square, it added a second jesus, and I wasn't able to erase him. strange.
so far, so weird. next I tried uploading an image and altering it. I took an image of my backyard and gave it the image prompt "flock of seagulls flying over my backyard" and erased part of the sky where I thought the birds could go.
here's the original shot:
cue seagulls:
doesn't look bad from a distance. but on closer inspection, there are some strange and rather inexcusable artifacts going on around the birds, which shouldn't really happen because it's not even blending new image with old image. the whole section they occupy has been erased and it gets to start anew there.
I would have expected better in that scenario. but let's see how I might use this tool for my own business. what if I wanted to add some art to my website - say, to advertise my wedding services. how about the prompt "color pencil drawing of a handsome young wedding videographer taking a beautiful shot of a gorgeous bride and groom posing in front of a waterfall underneath a rainbow"
maybe something could be useful here. but where's the videographer? ok, let's try that same prompt again but replace "colored pencil drawing" with "photorealistic image." the results are nothing short of bizarre. maybe we're panicking prematurely?
I don't know much about the different image generators out there, so I can't compare it to others or even be certain that I'm using it properly. Some of the showcased results are impressive, and I thought I'd give it a whirl. It's a lot of fun, and a shame that one's use is limited.
I thought I'd start with "an historically accurate depiction of jesus christ sitting atop a rocket ship circling the sun". it spits out 4 results. 3 of them showed JC as a white guy. what's really interesting is that they all shared a kind of cartoonish hippy dippy style that I thought was totally befitting the weirdo image prompt. it's gratifying to see an image spit out that I lack the skills to create but can somehow claim is my creation. but there are so very many decisions that would get made along the way in the creation of something like this that I had no part in. in a way, the process takes more away from me than it gives me. also, when I extended the image downward to make it a rectangle rather than a square, it added a second jesus, and I wasn't able to erase him. strange.
so far, so weird. next I tried uploading an image and altering it. I took an image of my backyard and gave it the image prompt "flock of seagulls flying over my backyard" and erased part of the sky where I thought the birds could go.
here's the original shot:
cue seagulls:
doesn't look bad from a distance. but on closer inspection, there are some strange and rather inexcusable artifacts going on around the birds, which shouldn't really happen because it's not even blending new image with old image. the whole section they occupy has been erased and it gets to start anew there.
I would have expected better in that scenario. but let's see how I might use this tool for my own business. what if I wanted to add some art to my website - say, to advertise my wedding services. how about the prompt "color pencil drawing of a handsome young wedding videographer taking a beautiful shot of a gorgeous bride and groom posing in front of a waterfall underneath a rainbow"
maybe something could be useful here. but where's the videographer? ok, let's try that same prompt again but replace "colored pencil drawing" with "photorealistic image." the results are nothing short of bizarre. maybe we're panicking prematurely?
Attachments
How long does it take the AI to draw an image and it it limited in resolution?
PS. The one with a woman with three shoulders looks as if the computer misread the instructions and decided to embark on an improvisation of its own.
PPS. "color pencil drawing of a handsome young wedding videographer taking a beautiful shot of a gorgeous bride and groom posing in front of a waterfall underneath a rainbow" - I wonder if this sentence contains too many adjectives and confuses the AI.
PS. The one with a woman with three shoulders looks as if the computer misread the instructions and decided to embark on an improvisation of its own.
PPS. "color pencil drawing of a handsome young wedding videographer taking a beautiful shot of a gorgeous bride and groom posing in front of a waterfall underneath a rainbow" - I wonder if this sentence contains too many adjectives and confuses the AI.
OK, this is spreading quickly. This is Pink Floyd's "Comfortably Numb", with the AI generated images based on the lyrics of the song. Given the semi-surreal nature of Roger Waters' lyrics, the images reflect them ... loosely. Which makes them different than the Alan Parker film.
and here's the original film (1982)
and here's the original film (1982)
ahalpert
Major Contributor
Well, the website claims that the more detailed your instructions are, the better the result will be. And judging by some of the output in the photorealistic request which was an identical prompt except for the medium requested, it seems like the computer understood every part of the prompt
Each request you submit takes about 10 seconds to spit out results. I think that that's the amount of processing time each credit will buy you. I think that you're buying a 1024x1024 square of generated pixels each time. My backyard photo was 8000x4000 pixels, and it took multiple generate requests to create new material in the area I erased. I think I spent 4 credits on those lousy birds. The Jesus drawing was 2 squares generated separately. I tried putting a 3rd one beneath the first two, but I didn't like it
Bassman2003
Veteran
Thanks for sharing this Abe. Pretty funny at this early stage. It will probably mature rather quickly once the results improve.
On the AI version of Comfortably Numb, it was probably the desire to be as literal as possible to the quoted lyrics.
And the Qualcomm product manager chimes in on the AI, the smartphones and the "DSLR".
https://www.dpreview.com/news/46827...lp-smartphones-eclipse-dedicated-cameras-soon
https://www.androidauthority.com/smartphone-camera-future-interview-3183730/
More from that interview.
More from that interview.
His presentation is credible ... but ... a smartphone doesn't have to be as good as a "DSLR" to destroy the Japanese photo-video cartel. It just has to be reasonably close (to your back pocket).
NorBro
Major Contributor
A once common insult about a position hurled at those who may have had career trouble is now being done by robots...I didn't see this coming, honestly.
It worries my android brain how & why I didn't see this coming but now this realization has opened up so many thoughts on other jobs being immediately replaced *overnight*.
https://www.yahoo.com/finance/news/w...100917185.html
It worries my android brain how & why I didn't see this coming but now this realization has opened up so many thoughts on other jobs being immediately replaced *overnight*.
https://www.yahoo.com/finance/news/w...100917185.html
ahalpert
Major Contributor
doesn't surprise me. the more repetitive a task and more easily it is described by an algorithm, the easier it is to give it to AI. I don't know what goes into flipping fries, but I assume that the sensory inputs for the cook are limited, the movements predefined. I read a while back about having an AI edit game highlights for sports games (it may have been basketball in specific). All the computer did was search the audio track for the loudest applause sections, then it selected the play that preceded it. the work that requires general intelligence will be the last to go, or may never go
In 1938, the Germans completed construction of the largest single roof building on the planet. The Nazis named the one million square meter plant Kraft durch Freude (Strength through Joy). Its sole original purpose was to manufacture 1,000 light passenger Käfer cars per month. The Beetle never went into full production until after the war but the military version known as Kubelwagen (the bucket car, since it had the first ever bucket seats to keep its drivers and passengers from sliding) and Schwimmwagen (an amphibious car) were approved at the same 1,000 unit per month/12,000 year level.
The building, long renamed Volkswagen, is still in business of automotive manufacturing. In the last pre-Covid/pre-war year (2019), it put out ~ 700,000 cars and light trucks. Besides the obvious fact that the Nazis were dumb asses, that's roughly a 60 times increase in productivity. And you're talking Flippy.
PS. Outside the VW plant is a statue of one Heinrich Nordoff. Pre-war, Nordoff was a plant manager for Opel (in Russelheim). Opel, until recently, was a General Motors trademark (GM sold it to the Italian/Euro Stellantis/Dodge group in 2019) and GM ran the production the "American way' since purchasing the facilities and the brand name from the Adam Opel heirs in 1929. Nordoff, however, was a Parteigenoße and, after the war, GM decided to part with his services. As a fluent English speaker, he then applied at VW, which was run by the British (with Ivan Hirst in charge). The plant output was ~ 400 cars per month. Nordoff made ... a few changes. The eventual production that continued well after Nordoff's death in 1969 was ~ 21.5 million.
The building, long renamed Volkswagen, is still in business of automotive manufacturing. In the last pre-Covid/pre-war year (2019), it put out ~ 700,000 cars and light trucks. Besides the obvious fact that the Nazis were dumb asses, that's roughly a 60 times increase in productivity. And you're talking Flippy.
PS. Outside the VW plant is a statue of one Heinrich Nordoff. Pre-war, Nordoff was a plant manager for Opel (in Russelheim). Opel, until recently, was a General Motors trademark (GM sold it to the Italian/Euro Stellantis/Dodge group in 2019) and GM ran the production the "American way' since purchasing the facilities and the brand name from the Adam Opel heirs in 1929. Nordoff, however, was a Parteigenoße and, after the war, GM decided to part with his services. As a fluent English speaker, he then applied at VW, which was run by the British (with Ivan Hirst in charge). The plant output was ~ 400 cars per month. Nordoff made ... a few changes. The eventual production that continued well after Nordoff's death in 1969 was ~ 21.5 million.
NorBro
Major Contributor
That AI sports highlights editing could work like that, but it would still be missing skill plays that are constantly included in highlights [like ESPN highlights] that don't have the same dB. [For non-sports people, meaning more than just touchdowns and home runs, etc.]
And for the fries...it's not that the task is being done by robots because of course the automated process is perfect for that, but that machines may start to replace people in those types of jobs already so quickly, like now.
I am playing a ridiculous Switch game right now taking place in a very advanced world with machines and aliens and it's actually pretty depressing. lol
Extremely beautiful but what a hell earth has become in it.
And for the fries...it's not that the task is being done by robots because of course the automated process is perfect for that, but that machines may start to replace people in those types of jobs already so quickly, like now.
I am playing a ridiculous Switch game right now taking place in a very advanced world with machines and aliens and it's actually pretty depressing. lol
Extremely beautiful but what a hell earth has become in it.