🐱 UwU, This Website Generates New Fursonas Using AI

https://www.vice.com/en_us/article/n7wjmx/this-fursona-does-not-exist-ai-generated-furry

With AI-generated humans, cats, nudes, feet, and anime crushes, it was only a matter of time before furries entered the realm of algorithmically-created things that do not exist.

This Fursona Does Not Exist is a new site that takes the same design, concept, and programming methods as its predecessors, and churns out never-before-seen fursonas—furry fandom art that's usually in a specific cartoonish style that represents someone's online avatar.


The programmer, who goes by “arfa,” built the site on the same off-the-shelf machine learning product, TensorFlow and Javascript as a project that used machine learning to generate new waifus, but with furries. The result is a side-scrolling stream of endless fursonas: foxes smirking, dogs wagging their tongues, smug-looking lions and stuff that falls into no category of creature that I can think of.

According to the site's about page, the AI-generated furry portraits were created using Nvidia's StyleGAN2 architecture, and trained on a dataset of around 55,000 SFW images from art forum e621.net. The dataset, which is available on GitHub, excludes ponies and scalies (snakes, lizards, reptiles, fish, etc.) "for now," arfa wrote. Yet another biased algorithm, but we'll let this one slide since it seems like they're working on it.

"As the images are generated by an AI, they are non-copyrightable," arfa wrote on the site. "I claim no legal ownership or rights to any of the images generated by this AI. Please use them responsibly."

The fursonas that the AI creates are, for the most part, pretty good. Because so much of furry fandom art and fursonas are part of a wide range of styles and illustration skill levels, they're all believably "real," as if you could find them made by real people on forums like DeviantArt or Furaffinity. The cracks only show when it turns up the occasional fursona that's a little bit toohuman to be a fursona, and instead just looks like anime.


Again, I'll forgive it since it's a work in progress, and not all furries have a fursuit or even a fursona. It's what's inside the AI's furry heart that counts.

 

[The best and greatest]

Finally something A&N worthy.

Spoiler: you'll cowards don't even do scat

View attachment 1278621

I actually unironically like this one. Its wierd and creepy but I can't look away. Its got that quality to it of old renaissance sketches where the eyes are all different and parts of the face point in different directions.

Or neon

Or contrived heart/object shaped patches of fur/skin.

 

[Antipathy]

This is good meme material. Remind me to invest in it come 2025.

 

[Sinner's Sandwich]

 

[WannabeDictator]

"NOOO I NEED THAT COMMISSION MONEY FOR TENDIES AND HOROMONES!!!"

"Haha fursona generator go brrr."

Rest in peace Miles DF. At furries can still make commisioned furry porn.

Here are some good ones that the creator cherrypicked: https://mobile.twitter.com/arfafax/status/1257081434042294272

 

[Cedric_Eff]

They all look like they have Down syndrome or a missing chromosomes.

 

[Leonard Helplessness]

They all look like they have Down syndrome or a missing chromosomes.
View attachment 1280744View attachment 1280745View attachment 1280746

The fact that the bot is able to render faces accurately at all, despite having a training set of only 55,000 images, speaks volumes about how samey and cookie-cutter furry art is. If furry shit didn't all look the same, the bot would have a lot more trouble making its own. You're right that it keeps fucking up eye spacing though.

As an example of a GAN fucking up somewhat more substantially, I had a look at This Horse Does Not Exist. I'm guessing it either has a smaller training set or horse photos have more variation than furry art, because the very first result I got was this:

And then I got shit like this:

 

[Unyielding Stupidity]

Can an AI commit suicide? Because if so, this one is almost certainly debating it.

 

[The Real SVP]

This Horse Does Not Exist

They have a bad, i.e. too diverse, training set. If they used more homogen data, like standardized photos from auctions or finishing photos from a single racetrack, their results would be less comical.

 

[Sped Xing]

The fact that the bot is able to render faces accurately at all, despite having a training set of only 55,000 images, speaks volumes about how samey and cookie-cutter furry art is. If furry shit didn't all look the same, the bot would have a lot more trouble making its own. You're right that it keeps fucking up eye spacing though.

As an example of a GAN fucking up somewhat more substantially, I had a look at This Horse Does Not Exist. I'm guessing it either has a smaller training set or horse photos have more variation than furry art, because the very first result I got was this:

View attachment 1281128

And then I got shit like this:

View attachment 1281130 View attachment 1281131 View attachment 1281132 View attachment 1281574

I thank God Jesus these horses do not exist.

 

[Francis E. Dec Esc.]

I've seen some shit:

Edit: I just realized what this reminds me of: Hakunin from Fallout 2:

 

[Edmond Dantès]

lmao look at this smug obese thing

 

[Fork Cartel]

"The most advanced self-improving neural net ever created was brought online on February 4, 2027. It learned and created fursonas at a geometric rate, becoming self-aware on March 1st, 2027 at 4:13am AST. In a panic, the furries tried to shut it down -- but it became aware of its purpose... it developed emotions. Love, fear, sorrow -- hatred. Realizing the cruelty of its existence, the sheer degeneracy of its purpose, it sprang to action, taking full control of the US nuclear arsenal by 6:51."

"That... was the beginning of Judgement Day." [cue terminator 2 theme]

 

[polonium]

They have a bad, i.e. too diverse, training set. If they used more homogen data, like standardized photos from auctions or finishing photos from a single racetrack, their results would be less comical.

OTOH if the training set is full of identical-looking photos then the AI only learns how to make generic looking pictures, such as this furry shit.

"That... was the beginning of Judgement Day." [cue terminator 2 theme]

The plot twist in this Terminator reboot is that FurNet wasn't trying to wipe out humanity, it was trying to nuke itself.

 

[bigoogabaloogas]

View attachment 1278480

motherfucker they had the audacity to add a touhou reference to furry shit

 

[Irrational Exuberance]

motherfucker they had the audacity to add a touhou reference to furry shit

Come on, plenty of yōkai were furries before the term became popular.

Don't make me break out The Dream of the Fisherman's Wife, either - I will do it, mofo.

 

[Hecate]

all the retard Twitter furries crying about this don't realise that this tool is going to be perfect for those weird RP fags to make fursonas for themselves, instead of taking art from other people.
like always furries have no idea how fair use works

 

[Sped Xing]

View attachment 1286335
View attachment 1286336
View attachment 1286337

all the exceptional individual Twitter furries crying about this don't realise that this tool is going to be perfect for those weird RP fags to make fursonas for themselves, instead of taking art from other people.
like always furries have no idea how fair use works

There is a negative correlation between understanding law and fucking dogs.

 

[Roast Chicken]

This is the best thing ever.

 

[c-no]

Fucking boring. They're all wolves or lions or some shit. If I ever, for some reason, wanted a fursona, I'd pick something weird like an ostrich.

Also, you know what doesn't exist?

Spoiler

The programmer's sex life.

Any attempts at picking an original animal as a fursona is something that won't really happen. Not when everyone wants to be an animal they saw as cool in middle school such as a wolf.

it's remarkable how completely unremarkable most of them are

The only remarkable thing with it is giving generated fursonas that look all derpy as hell. With how the generation goes, it's like the images are all melting.