https://thischemicaldoesnotexist.com/ is actually quite interesting. It's generating completely valid chemical structures (some of which may actually exist although this is unlikely due to the large size of chemical space).

And https://thisurldoesnotexist.com/ is quite funny.

Also this one: http://thiscitydoesnotexist.com/

We built this city on TensorFlow!

I'll add it to the site!

I'm a big fan of https://git-man-page-generator.lokaltog.net/ .

So how do I know these things don't exist? How can I know that the algorithm isn't just regurgitating something it was trained on?

You can usually access dataset and run comparisons.

Very impressed with 'This MP does not exist'.

If wanted to do the same to generate 1000s of fake professional headshots, how would I go about it?

Seminal papers in this area AFAIK are StyleGAN1 and StyleGAN2: https://github.com/NVlabs/stylegan2

https://arxiv.org/abs/1912.04958

This is a great collection, but there are some factual errors, I’m pretty certain. The image links use GANs, but text GANs have yet to catch up with “simple” discriminative models. The text links don’t use GANs, and the description of an RNN as not quite a GAN is just wrong in most ways you’d interpret it.

Taken together, 'This Waifu does not exist' and 'This snack does not exist' may well put you off eating snacks while looking at Waifu faces. For life.

"Mollusc Tentacles' Purée Supplemented Pretzles". "Dothraki horse-heart gummy bears". Etc.

If a real-life fatal basilisk image (https://en.wikipedia.org/wiki/BLIT_(short_story) ) is possible, it will be created by something like this. The method is akin to how it is developed in the prequel story: An AI that is adversarially trained and accidentally produces a visual effect that crashes human brains.

There is the McCollough effect, which while not fatal, does have a long effect on the brain.

https://en.wikipedia.org/wiki/McCollough_effect

https://youtu.be/Wm8ZoVQ_OJo

Thanks for pointing that out. Another is the https://en.wikipedia.org/wiki/Tetris_effect .

> An AI that is adversarially trained and accidentally produces a visual effect that crashes human brains.

I very much doubt that would be possible in general. OTOH why using the cumbersome method of training an AI for that? Stroboscopes and flashing lights work perfectly well for a portion of the general population already ;)

We are likely already vaccinating ourselves against them though, as we all experience the development of these weapons at each iteration. It might be a watching a kettle boil situation.

The poor bastards who who arent exposed to the internet everyday are likely going to need some protection though.

To be fair to TWDNE, just generating random faces is among the least interesting use-cases, and one could do better than TWDNE with some more work.

For example, here is a demonstration of how the new model averaging/transfer tricks let you convert a face from human to anime to fox to pony to furry in any direction by any percentage (summing to 100%): https://twitter.com/arfafax/status/1296084902928986113 (And since you can encode any pre-existing photograph into the human face model, you can do things like "what if Barack Obama but only 60% human and 20% fox and 10% anime and 10% My Little Pony?" Impossible-sounding, yet, there you have it.)

The pony/furry models are the same ones used for http://thisfursonadoesnotexist.com/ and https://thisponydoesnotexist.net/ , incidentally. (TFDNE/TPDNE came out much nicer than TWDNE because of better cropping/aligning, longer training runs on entire TPU pods rather than a few GPUs, and transfer learning.)

We're also still working on BigGAN for generating entire anime images. An older model: https://www.gwern.net/Faces#danbooru2019e621-256px-biggan Samples from a current training run: https://cdn.discordapp.com/attachments/693736350561861633/78... https://cdn.discordapp.com/attachments/693736350561861633/78... We had mostly given up, but recently discovered that random cropping was the reason our runs were taking forever and were pretty bad; fixed cropping like top cropping works much better.

I had a lot of fun with "this lyric does not exist" - I don't know if that's more of a statement about modern lyrics lowering my expectations, but some of these could pass the taste test.

I think this is definitely going to be the next smash hit:

Verse 1

I will feel a poop down in the South sea

I will feel a poop down in the South sea

Poop from the past never feared again

Poop from the past shot by a beam

Pre-Chorus

A poop straight from the past

A poop straight from the past

Chorus

Just poop off and come to me

Just poop off and come to me

Just poop off and come to me

A poop from the past

Verse 2

The voice of struggle is clear

And the clear sonic poop of the bomb

Has shattered the icy poop of the night

Felt a warm poop of energy

Pre-Chorus

Poop of delight

Oh, how glad my soul

Chorus

Just poop off and come to me

Just poop off and come to me

Just poop off and come to me

A poop from the past

Bridge

You save my words

We have a poop

Chorus

Just poop off and come to me

Just poop off and come to me

Just poop off and come to me

A poop from the past

Haha very cool. Thank you for sharing! How does https://TheseLyricsDoNotExist.com work? I have only seen StyleGan on images but it seems similar?

Not exactly this site, but I simply fine-tuned GPT2 small: https://m.habr.com/en/post/453232/

Interestingly, these lyrics are now in the public domain. They cannot be copyrighted as computers and algorithms cannot hold copyright.

Practically, yes. Theoretically, it is more interesting.

That particular instance of those lyrics is public domain, as are any copies made from them, directly or indirectly.

If you independently happened to write identical lyrics, you would have copyright on those. That would be a worthless copyright though, because it would be almost impossible to ever do anything with it (hence, why I said "practically, yes").

If you sued someone for infringement, the defendant would argue that you didn't independently write identical lyrics. They would argue that even if you don't remember ever seeing the public domain lyrics or a copy of them or hearing someone sing them, it is far more likely that you did and forgot than it is that you independently happened to come up with the same thing. They'd have no trouble bringing in experts to calculate the probability which would be in the "not expected to happen in the short time the universe as existed" range.

To overcome that, you'd have to be able to very thoroughly document that you really did independently write identical lyrics. Short of you having been away on an interstellar mission and able to prove that you wrote yours while the public domain ones were not in your past light cone, this will be difficult.

Even if the court decides that you really did independently create them, your case is still nearly impossible. The defendant would claim that they had seen or heard the public domain lyrics, and those are the ones they actually copied. Even if you could prove that they saw or heard yours too, even before they saw or heard the public domain ones, it would not be good enough. You'd have to prove that yours are the only ones they have seen.

At least in this case the worst that happens to you is that your copyright is worthless. If the original was not public domain, and you really did independently create something identical, you could end up being the one sued for copyright infringement, and you would probably lose.

Thank you for sharing these. I think they are perfect for my website.

I wonder what happened with "havetheyfaked.me" I wanted to gemerate some fake lookalikes and nos it justo redirects to thispersondoesnotexist

I would like to see something like "This light speed spacecraft design does not exist".

does not exist yet

truly mind blowing to see all those generated faces, many of which look totally legit and may exist given the infinite possibilities of the future...

So waifus are weird, but feet and pony generators are not.

Talk about bias.

I thought this post was going to be "This variable X does not exist" and have a machine-generated formula solving for X.