1. Decide what you want to look like
2. Create a reference sheet (or commission an artist to create one based on a description from step 1)
3. Commission a headshot for an avatar
4. Commission some art of your fursona, possibly with your friends' fursonas
5. GOTO 4
6. At some point, your friends may ask you for your reference sheet so they can gift you an art piece of your fursona and theirs (see optional stage of 4)
Even if you can churn out AI-generated headshots, getting the colors/markings just right for your character is nontrivial for non-generic fursonas (see https://soatok.com/static/soatok-johis-responsive.jpg for example).
And besides, most art commissions are conducted after you have a reference sheet and/or headshot of your character.
If anything, this will give artists something they can point the "steal other people's character art for their roleplay accounts" types of (especially younger) furries towards. "Can't afford to commission an artist? Just use the AI thing and stop the misbehavior!"
StyleGAN works best on single centered objects which are not too diverse in shape. So heads work great if you align them to the middle, and you can get OK results with single centered figures like someone standing, but anything beyond that and it falls apart. We have a theory about that (https://github.com/tensorfork/tensorfork/issues/21) but we've long since moved onto BigGAN, which can model much more diverse datasets like ImageNet successfully.
As far as fursonas go, see my other comment: https://news.ycombinator.com/item?id=23096442 I'd say that GANs are better at letting someone explore fursonas than any human artist is or ever will be. You cannot click a button and have a human artist generate 32 variants in a second for you, toggle on or off attributes, nor can they slide a slider from 'human' to 'neko' to 'dangerously cheesy!'. You might pay for a commission of your final fursona for the highest-quality image with no artifacts, but for exploring...? And as NNs get better over time, they'll creep up the value chain, as it were. (I'm still a little shocked how good the human voices in OpenAI's Jukebox https://openai.com/blog/jukebox/ are, even if the highest level of abstraction is still very shaky and they need to tack on another layer or two to get choruses etc.)
I tend to assume the latter, especially with any of the "Elsagate" style content.
Is there a name for a Reverse Turing Test? Because those videos pass it.
However, this is the one case where more incoherence works better.
One of those interesting side-effects of furry avatars that I noticed is that accounts bearing those avatars were always real people, with the added bonus that you can generally authenticate the human behind the mask if you know how. The reality of being online today is that we have to understand whether we're interacting with real people or just a clever piece of software, and this is far more true for folks who are not technically savvy.
Oddly enough, furries were the last bastion of humanity. (And a welcoming one at that, but I digress.)
This complicates that heuristic somewhat. This brings furry avatars on the same level as human headshots. I now need to, e.g., read and process the full account bio and spend more time authenticating whom I interact with online.
This is great work, but did you really have to?
What nrr is saying is simply:
If someone replies to a comment and they have a human avatar, there was always a chance it was the output of https://www.thispersondoesnotexist.com
Their heuristic for "if it's a furry avatar, it's probably not from a Russian troll farm" is now invalid.
• fallback — e.g. the egg image on twitter
• explicitly algorithmic — e.g. http://identicon.net/
• logotypes — actual logos, icons, arbitrary aesthetically-pleasing typographical art
• headshots of real people
• photographs of real-life things — places, nature, buildings, etc.
• crops from TV/film — headshots of actors playing characters, or whatever you'd call Baby Yoda
• famous works of art — crops of prints of paintings, crops of photos of sculptures
• commercially-marketable art — box art, movie posters, crops from cartoons/anime, professionally-commissioned CG paintings that fit the style of their source material
• unknown, non-marketable art — works that are clearly either self-made, or commissioned as a one-off, where the work has traits that make it specific-enough to someone's tastes that it obviously would never have been produced as spec work without an arranged buyer/planned use
The parent's point is that, for all the categories except the last, there's an obvious way to scrape or generate a million such images, that someone can include in their spambot/voting-ring registering algorithm.
The last category, though—custom competent-but-not-commercial-looking illustrations—were, in some sense, a Proof of Work token for the profile it was attached to: someone had to draw that (and even more, gather requirements to draw that, rather than it just being one keyframe following the same rules of thousands of others.) It cost a few dollars for that person to get that image; and therefore, it's less likely (though not impossible) that ten users with ten different such illustrations in a forum thread were all secretly the same person/bot.
There hasn't even been an AI that can do face-detection on funny-animal cartoons until now, AFAIK, so there was until now no way to even automate+scale scraping of "authentic" profile-pictures from some art-hosting website, let alone a way to automate+scale generating them. But now the cat's out of the bag. Bit of a shame.
There's a few furry-oriented art websites, so this part has actually been really easy for like two decades.
And human labor—even the cheapest human labor—still costs way too much to have real people sitting there cropping pictures to use as profile pictures, if you need a basically-infinite stream of them for your spambot network.
So you'd need, at minimum, a face-detection algorithm that you can rely on to do auto-cropping, so that you can just throw it a whole scrape-dump of the art site, and get back that basically-infinite stream of headshots.
As I said, until this very project, there was no face-detection algorithm that worked on illustrations of, er, "demihuman" faces. (There was one for humans, that would fail horribly on this data; and then specific other ones for cats/dogs/etc in photos, that also would fail here.)
All is fair in war, love, and the name of science (with exceptions).
After spending a lot of time online as well, I've found just as many folks using content they stole that weren't real people, but were generative content.
Also, because the corpus of source material is comparatively pretty small, you can use image hashing and a distance metric to match something you see to its source. It takes more effort to draw something than it does to snap a photo, and the entirety of E621 can fit on a single SSD.
This is quite unlike headshot photographs of people. Authenticating the photographer is almost impossible, and because of the low-effort nature of photos, it's infeasible to have a control data set to match with observed data.
Or you could pay an artist to do that for you.
Neither tactic really scales, especially when your game is coordinated inauthentic behavior i.e. https://www.youtube.com/watch?v=V-1RhQ1uuQ4 and you need thousands of convincing fake accounts.
Furries are a pretty tightly-knit bunch of people, and the folks who occupy the artwork-having set are usually two or three degrees of separation from each other.
As an added bonus, in case the social network effects weren't enough, DMCA claims become easier because the provenance of a piece of art is often quite clear, what with the subculture's emphasis on always attributing the source of the art in question.
1) voteforwest2020
2) weproudto2black
3) aaddictedtoblackk
4) best-usa-today
5) black-to-the-bones
6) blackness-by-your-side
7) blacknproud
8) blacktolive
9) bleepthepolice
10) guns4l1fe
11) sumchckn previously known as: blondeinpolitics, blvckcommunity, classylgbthomie, hwuudoin, politixblondie
12) superblygun
[0]: https://techcrunch.com/2018/03/23/tumblr-confirms-84-accounts-linked-to-kremlin-trolls/
My definition of "infiltrating furry Tumblr" is perhaps a fair bit stricter: Infiltration implies being accepted as a member of the community, which also implies some level of gossip around someone new or cool whom folks are excited to introduce or meet.
The problem is that Tumblr killed these accounts, so there isn't any good record today of what their profiles looked like, what avatars they had, etc. I also can't seem to find anything particularly furry-centric beyond the fact that a malicious actor saw fit to target a subculture for some political trolling.
I'm chalking this up to the interface of a subculture with mainstream society being compromised and not necessarily the subculture itself.
These are characters that are recognizable as being digital representations of those individuals. They aren't the trademarks or personality rights of some faceless commercial entity.
My point is that trying to use artwork depicting a non-commercial character that isn't yours—that doesn't accurately reflect the representation of your digital self but instead someone else's—is as taboo as impersonation. It isn't tolerated, and that kind of behavior is ripe for gossip.
So, yes, while there's a wealth of source material that's ripe for cropping into a profile picture, there's the added hurdle of overcoming the bullshit detectors of furries who will test you to see if you're actually their friend or not.
There are a few facts I feel are worth emphasizing here. Sorry, it's a lot of words, but I want to make sure the required context is here for drawing the logical connectives that are missing.
Notably, furries tend to be acquainted with each other, in the sense that if two people within the community don't personally know each other, the chance is very good that they at least know of each other or have mutual friends who can make introductions. For individuals who exhibit problematic behaviors, interventions happen, and they happen reasonably swiftly.
This goes hand-in-hand with the fact that there are artifacts that act as shibboleths within the community and help to signal "hey, you're one of us." The subculture has a very well-defined market, literally and proverbially, for these kinds of artifacts, and the market is amazingly insular with known players who are also themselves part of the community.
One common example is the artwork I've been talking about, given its relevance here, but I've left a thing or two unstated in my conversation about it so far. There's a reverence of sorts for the artists who do this work, and I'd feel safe in asserting without evidence that artists don't tend to become popular or get work at all unless they've had a good inculcation of what furry as a subculture is all about, up to and including the social connections. Ultimately—and forgive the hand waving—this means that the artwork is readily identifiable as "furry" (as opposed to "just a cartoon") and that there's an understanding as to where it came from. It's able to be authenticated as having originated from within the furry subculture. The implication that there's widespread interest to know the identity of the individual who did a piece of artwork is hopefully clear.
("Ooh, are they open for commissions?" is a common refrain.)
That said, these artifacts extend far beyond just raster images on the Internet. There are T-shirts, stickers, enamel pins, and what seems like myriad other things that are innocuous (or just plain weird, let's admit) to folks outside the community but will start a conversation between folks within. As an anecdote: A homemade sticker on my ThinkPad given to me by an artist friend was seen by a passenger sitting next to me on a flight, and it wound up leading to us talking for the whole flight and getting margaritas while on layover to our separate final destinations. I still talk to that individual today, and knowing them has actually strengthened my connections to other folks.
Those artifacts wind up being a very good way of establishing common ground with someone else within the community very quickly, further cementing the "hey, you're one of us" effect.
… phew.
Let's recap. The furry subculture is particularly tight-knit and egalitarian with strong social connections and community cohesion that are reinforced by cultural artifacts and shared experiences that come with the ability to be verified as genuine.
Or to repeat something I said earlier: "[Copying furry artwork en masse] is generally not advisable; you'll be found out pretty quickly if you do," given a definition of copying probably being closer to co-opting.
I won't assert that the furry subculture is always better than any intelligence agency at this game, but it has a lot of things going for it that certainly make this kind of deceptive, divisive activity harder than it would ordinarily be.
What do you mean by this?
> (for people not familiar, furries are a vibrant online community and one of the last outposts of collaborative, creative Internet culture. They're also heavily LGBTQ and have a deep commitment to inclusiveness and social justice. Pretty great people to have on your side!)
There are a couple of important facts to consider: The furry subculture is very tightly knit, and there's a reasonable chance that you're only two or three degrees of separation from someone else. There's also an emphasis on attributing furry artwork back to its source, not to mention tools for doing just this.
Those two things make authenticating the identity behind an account with a furry avatar comparatively very easy when contrasted to a human headshot photograph.
This one is odd; are those horns or spikes? And what’s that sticking out the back (or is that from the ear?):
https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed30101.jpg
It’s always fun when these kind of systems try to generate text:
https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed25387.jpg
And then we have this one, which is pretty clearly Toriel from Undertale:
https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed14017.jpg
So things like: gender, hair color, face shape, etc?
Basically, an AI-driven character generator that isn't completely random?
GAN models do not need to be specifically architected to enable control, because you can reverse them to get the latents/seed and manipulate that to 'edit' images: https://www.gwern.net/Faces#reversing-stylegan-to-control-modify-images So if someone wanted, they could use Arfa's model to edit images.
- It was already possible to create fursonas based on pictures of human faces.
- It was possible to generate pictures of faces of humans that do not exist.
Pipe these two together and you get this ...
The furry fandom is extremely touchy about "art theft." Considering how many of these images clearly resemble individual artists' styles, I'd tread a little more carefully.
This part seems false. If I take copyrighted photos and run it through something like an "AI" filter to change the colors slightly, the photos aren't suddenly public domain.
If you steal from enough people at once is it then legal?
That is not what is being done here. See also https://en.wikipedia.org/wiki/Transformativeness
> If you steal from enough people at once is it then legal?
How many people does a great human artist steal from to become an artist? If you steal solely from one other artist, you're a plagiarist; if you steal from a thousand other artists, you're an innovator and pioneer of a new style...
Generating a recognizable character is no more 'overfit' than a painter being able to paint George Washington means their brain is 'overfit'. If those characters are part of the data distribution, as they are, then the GAN can and should generate them and countless variants on them.
I don't think the internet, where 4chan or the dancing coffin meme were born, is really a medium that stops at the threat of an outrage.
But, I am wondering if you ran an image similarity search on them vs. the training set would you find matches or are they actually unique?
https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed09985.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed13901.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed31957.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed25075.jpg
Not to say that isn't impressive! Generating convincing fakes, even 1% of the time, even if they're not super unique (There's plenty of people who "look exactly like" in the real world, too!) is a big deal.
I would expect some over prevalence of certain popular characters. For 25075, I would assume https://sonic.fandom.com/wiki/Amy_Rose#Heroes
The fact that nearest-neighbor lookups do not find exact overlaps between the training dataset and a large number of random samples has always been one of the arguments I use against the widespread misconception that GANs 'just memorize' data: https://www.gwern.net/Faces#faq
That so many people are convinced that a given datapoint must be an exact copy of a training datapoint - "it looks exactly like a Zootopia character I recognize!" - is really quite a compliment to the GAN...
I find this claim on the "about page" quite interesting. Some of those images might be so close to the training data that the copyright protection for fictional characters becomes relevant, even if the image is not identical. This is visible in this topic as people recognize characters from popular-culture (video-games or movies), because the training data seems to also contain fanart.
As you say, some of the output images are clearly of specific characters, which turns this from "legally grey" into "definitely not public domain".
This is the whole conundrum of creation and copyright. Every creative work is protected by copyright yet every creative work is the sum of unconscious derivations to varying degrees of something an author has perceived, creative works existing ex nihilo are at best vanishingly rare; personally I'm not even sure they exist, I'm leaning more towards we're just mistaking unusually big jumps of derivations/combinations of those for ex nihilo creative works. We readily recognise "influences" of great artists (whether it is music, literature, painting...).
Doesn't mean creative work should not be protected, but drawing the line of infringing vs not is by definition extremely blurry and subjective.
That talk is as relevant as ever:
https://www.ted.com/talks/lawrence_lessig_laws_that_choke_creativity
Am I now in the public domain?
Not that I'm looking to sue for such things but this "AI did it so it's public" thing ... easily could cross over into real life.
Related business: let me upload photos of my relatives and a family tree, and show me generated faces for other people on the family tree (e.g., common ancestors) based on these inputs and genetics. I wonder what kind of accuracy can be achieved in generating a person's face, based on how many descendants' faces we have photos of, over how many generations, with how much inbreeding, etc.
The gender bias is actually really interesting. I'm curious how much of it is due to:
* Inherent bias in the source dataset. (For example, nekomimi art is overwhelmingly female -- there's a reason they're referred to as "catgirls". The GAN does a great job of distinguishing this from Western anthropomorphic art styles, incidentally.)
* Bias introduced through filtering of the dataset, e.g. by excluding NSFW source material or certain tags, or a bias in what types of faces were recognized.
* Androgynous faces being interpreted as feminine by default.
* GAN-specific characteristics of the output (like smooth features) being interpreted as feminine.
I don't like the word feminine, because it's not really what furry feminine is. It's a human-applied-to-furry concept.
ACTUALLY. As I refresh the page, I get the feeling that it's the result of weird mixing between flat shaded art and mixing anthromorphic and "cargirls" (for lack of a better term- and I don't want to say east vs western, because catgirls dont' fall under furry)- anime inspired furry. It's weird, and makes me ponder. If the underlying datasets are a more similar art style and didn't have anime inspired art mixed in, I think the result would be less confusing and cooler.
Still an awesome project.
Why? I believe there's already so much content out there that this could easily happen today if the only question would be the raw amount. However if you ask people around the water cooler, instead of one guy watching joesmith34's Source Film Maker video on YouTube and another watching janedoe157's flash animation yesterday evening, everyone watched the latest episode of a popular Netflix series.
Nice ref there.
Impressive achievement, I'd love to see the outcome of re-training it with fan-art removed from the training set. There's a lot of Nick Wilde and Judy Hopps in there...
This demonstrates, at the very least, that it isn't simply memorizing the datapoints. You can see that it is able to smoothly transition between images of Zootopia characters and other characters, which indicates that it has learned a lot more about the actual features.
I believe the prevalence of certain characters (Zootopia, Sonic characters, Pokemon) showing up is because a large portion of the input space maps to those regions of the latent space. So I'd expect there to be a roughly equal proportion of images that look like Nick Wilde in the random samples as there are in the training data.
Also, I hate this. I especially hate the loading messages.
Nope! :3
https://faq.dhol.es/@Soatok/furry-fandom/is-furry-a-sex-thing
It's a common misconception. We're more about community than anything.
(That's not saying that sex doesn't happen between adults, but that's not what it's about, for everyone.)
Ok this is just a stereotype that's not true.
...you're messing with me, aren't you. :D
FWIW, I've met more normies in my 10years of software development than absurd people. Absurd people are just memorable, it's a bias.
If you're actually curious, this username was stolen from an absurd Atlanta based rapper named blunt fang and I get more weed culture association than furry association with it.
https://www.youtube.com/watch?v=bDal4GCyEJA
There are not many furries with fursonas that look like Judy, though I'm not sure why that would be gross.
Fursonas tend to mirror the style the content one was watching when they were a kid, so once kids who grow up watching Zootopia become old enough to enter the fandom, then those individuals might bring in a 3D style, but it has yet to happen.
As for the idea that furry are has too many similarities… I don’t think so. The variance in fursona body morphology is huge. For example, one thing I didn’t notice in this set was Sergals [1].
[0] Undertail. And no, not like that.
[1] https://en.wikifur.com/wiki/Sergal
https://www.youtube.com/watch?v=DiSs-sGclcU
https://www.youtube.com/watch?v=3HwdsjXEbOQ
https://www.youtube.com/watch?v=gGOFU62wVco
Zootopia: https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed18100.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed73222.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed46870.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed88525.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed16983.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed38798.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed34557.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed20308.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed65770.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed13269.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed31499.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed78829.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed02032.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed24939.jpg
Pikachu: https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed12878.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed05116.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed49143.jpg
Sonic: https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed22797.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed95651.jpg https://thisfursonadoesnotexist.com/v2/jpgs-2x/seed35704.jpg
[1]: The most popular furry booru. A booru is an image board where everyone can edit the tags of images. Strict tagging rules and advanced search operators make it easy to find specific images. A safe-rated only mirror is available at e926.net.
[2]: https://e926.net/tags?search[category]=4&search[order]=count
* original character https://knowyourmeme.com/memes/original-character-do-not-steal
I guess training on the source that has a massive crush on Zootopia wasn't the best choice for ‘doesn't exist’.
Also, those Sonics are apparently sourced from the gender-swap category.
http://studiohunty.com/punkemon/
http://studiohunty.com/punkemon/ptown/punkachu.html