hckrnws
It seems like we're approaching a sufficiently large set of This X Does Not Exist sites to train a GAN.
thisthisxdoesnotexistdoesnotexist.com
Do these "This Does Not Exist" sites do live image generation (they are so fast), or do they generate the images slowly over days and weeks and show us images from a cache?
I've also wondered this, because I've considered setting one up for fictional creatures. If anyone has any details on the architecture of how these sites are set up.
I assume they generate a new image per time period [e.g. a second] and show the same image to everyone visiting the page rather than doing a generation per page load, simply because 1) they're so fast, and 2) sometimes if you refresh fast enough you can get the same image (at least one some TDNE sites). Would love to read some actual setup details though!
The derivative empty-calories lowest-common non-creative aspect of AI generation is best exemplified by the "this idea does not exist" example where, guess what, the pretend new idea is an "YouTube for Lectures" which is as unoriginal a concept as anything you'd come up with if you really went hard for unoriginal... Basically the lowest bar for art now very much is: could an AI have done it?
To the author: There is also https://thismountaindoesnotexist.com/
(Disclaimer: I created that page)
Thanks! I'll add it to the site this weekend.
Feel free to steal this idea, but I'd love to see a thisgoatsedoesnotexist
Also one for dogs, please
In case you're unaware, "goatse" doesn't have anything to do with goats or animals or anything good.
Don't Google it.
Goatse for dogs? That might get you arrested in quite a few places.
Goatse for dogs is reportedly punishable by caning in Singapore.
And if it ain't, it should.
Unfortunately, thisdogdoesnotexist.com is already gone
Unfortunately and thanks to climate change and sea-level rising, many real beaches could then be featured on "this beach does not exist".
I propose a "this doomsday does not exist" corollary to this project:
- this dark age does not exist (viz. the fall of the Roman empire)
- this ice age does not exist (viz. the host of ice age scares from the late 60's to 80's)
- this apocalypse does not exist (viz. Mayan end of ages, comet scares, ...)
- this world famine does not exist (viz. Paul Ehrlich's entire oeuvre/eschatology)
- this nuclear holocaust does not exist (...yet?)
- this Siberian climate does not exist (viz. the UK going polar around 2020)
- this snow-free UK does not exist (viz. the opposite prediction of the one above, the inspiration for e.g. Marillion's "Seasons End" [1])
- these dead lakes do not exist (viz. the 'acid rain' scare in the 80's-90's)
- this end-of-oil does not exist (viz. "Peak Oil" in 2000, then 2020, then 2010)
- this technological eschaton does not exist (viz. the Y2K scare)
- this rapture does not exist (viz. the countless prophecies of rapture)
- this hockey stick does not exist (viz. most of Al Gore's infamous "An Inconvenient Truth")
In a few years time we'll be able to add 'this climate catastrophe does not exist' to this list while we'll be bombarded with dire predictions on whatever new apocalypse is awaiting us. Which, same as before, will be added to this list while an even newer scare awaits. That is, unless that nuclear holocaust prediction ends up becoming reality of course - time will tell.
Ironically, it seems like https://thisurldoesnotexist.com/ doesn't exist anymore and redirects you to some spam pages
Thanks! I'll remove it from my site.
"This Chemical Does Not Exist" shows valid pair-bonded atoms. Strictly speaking, doesn't every molecule it shows, actually exist?
GANs are really cool. They have recently been used in gravitational wave detectors to produce large numbers of glitches from a few thousand classified so far by Gravity Spy. The great thing about GANs is that they are super intuitive and the "adversarial" aspect is a compelling narrative for humans.
The cats are too good to be generated.
I thought so too, so I tried to find out if they were scraped from the web. I saved about a dozen generated images and used Google Reverse Image Search and TinEye to see if I could find a match on the Internet. But no luck.
Maybe the cat images look so real as compared to generated human faces because our brain is tuned to recognize human faces and can perceive the tiniest flaws, but not so with cat faces.
I don't think my search proves anything however. With hundreds of millions of cat pictures on the web, I wonder if simple transformations on a real photo -- like flipping or slightly rotating the image, squishing it vertically or horizontally by 10%, adjusting the color -- would be enough to hide the original from Google or TinEye?
>Maybe the cat images look so real as compared to generated human faces because our brain is tuned to recognize human faces and can perceive the tiniest flaws, but not so with cat faces.
Lucky for all of you, I spend way too much time staring at cats and while most of the ones generated here are pretty good, there's quite a few where the proportions are off and they look fake or edited to me. If I saw them outside of the context of being generated images, I'm very sure I'd still be staring at them trying to figure out what was wrong.
I think you should be able to find a latent vector that returns a cat that is part of the original training data (or at least very close to it). Most of the outputs will not be real cats at all though. However, it's pretty simple to try and find the latent vector that reproduces a given image, e.g. https://github.com/NVlabs/stylegan2-ada-pytorch/blob/main/pr...
TinEye can certainly find different resolution (and aspect ratio?) versions of images, and somewhat cropped images. I'm not sure how much rotation it can handle, though.
I've seen a couple coat colors that don't seem right. There was one where the cat was dark blue with a very light blue around the muzzle extending up to a strip between the eyes. It was basically your normal white spotting, but it's not genetically possible to have pigmentation in that pattern (barring chimerism, of course).
There were also a couple questionable ones where they had very studdy cheeks, but were tricolor (calico, tortishell, patched tabby). They distinctly looked like intact males, but you need two x chromosomes to have a tricolor cat. Or chimerism, of course, which I've learned over the years makes it impossible to ever say a specific pattern isn't possible.
It did get me wondering though, with the fact that most apparently male cats with tricolor markings genetically have an extra X chromosome, whether they would still get the big stud cheeks. I'd imagine they would, since they've got a Y chromosome and then I assume the X chromosomes turn on and off randomly the same way they do in females.
In less nerdy news, if you go through about a dozen pictures, you'll find that some of them are out of proportion or have their mouths open, but have the coat color in the mouth.
I dunno, I hit refresh a few times and this one seems a bit wonky: https://i.imgur.com/AyBMq5w.png
Having seen lots of ai generated imagery, I believe we are also collectively getting better at recognizing it from the real images. It's feels same as I used to see good 3d games back in 2000s as 'this looks so real' which are now just 'meh'.
"This rental does not exist". With the current housing market, this is to be expected.
I was planning to make a snarky comment here as follows: Ah yes, thisxdoesnotexist.com, let's repeat the fad from a few years ago when StyleGAN 1/2 was all the rage but has been widely outdated since. Still, it can come in useful when giving an AI talk to completely unsuspecting managers.
Though I was pleasantly surprised https://www.derekau.net/this-vessel-does-not-exist at least had been updating their approach, up to DALL-E 2 even. Not yet Stable Diffusion but I do applaud the effort.
(In all seriousness: just goes to show how quick things evolve in this space.)
New Idea: A curated list of "This X does not exist".
Hold my beer
Your beer does not exist.
Now I want a “This ‘x does not exist’ does not exist”
Crafted by Rajat
Source Code