It has been interesting to see images clearly generated by this service (same artefacts, same resolution) start to flood online dating apps and social media profiles. Previously, fake accounts tended to use photos which could be reverse-image-searched to detect them.
Luckily, it doesn't seem that this approach can be sensibly extended to generate entire scenes, or multiple entire-body photos from different angles. Believability breaks down tremendously in less constrained datasets- see https://www.thiscatdoesnotexist.com for example.
That may very well be because the datasets are not large enough to densely cover the underlying data distribution - which is what these current deep learning models need. Francois Chollet talked about it on Twitter and in some of his papers.
An exciting possibility that is opened by these generative models is synthetic data. Train (well) a generative model on private data, then share the generated syntetic data without privacy violation concerns.
Of course, this opens the way to data laundering - profiting from ilegally obtained data that you're not supposed to have. Possibly by laundering through several jurisdictions, where each local step is legal. What will be the Dutch sandwich of data laundering ? The cyber dystopian novels of the 90s have nothing on this.
The first five or six images that site gave me, I never would have guessed. One or two, completely perfect to my eye. One or two had maybe weird poses, but underneath enough fur that I honestly could never tell from a real cat making weird poses, as cats do. One or two I could see a minor clearly digital artifact that would give it away if I knew to question it, but I wouldn't have picked up on otherwise. Then the site gave me abstract art.
Funnily enough, your "nightmare fuel" screenshots mostly don't look that nightmarish to me... I might mistake some of them for real cats at a passing glance. (Not the first one; that one looks like a root vegetable.) I probably shouldn't get a job as a cat inspector.
EDIT: Ok this is one of the most horrific things I've seen this year. https://ibb.co/Np6KTYq
There's a couple of startups programmatically generating entire new scenes using only words as the input (target use cases are movies/animations/gaming). The output quality isn't yet as high as the OP site, but we're max 18 months out from something like www.thisplacedoesnotexist.com or something like that.
It remains an open question how far it can be pushed. Karras et al have avoided a lot of enhancements they could've used (self-attention being the biggest one) and don't even train that much (8 GPUs is hardly anything these days), and if you look at BigGAN samples on ImageNet, it manages 'entire scenes, or multiple entire-body photos from different angles' far better than StyleGAN 2 on CATS. (It doesn't do faces well, but what do you expect? ImageNet isn't face-focused and faces make up only a tiny fraction of its pixels.) So, since S2 can do such great faces and BigGAN can do great entire scenes, it seems beyond obvious that it is possible to do both simultaneously and it is 'just' a matter of compiling enough data and training on TPU pods long enough.
We hope to run an experiment with https://www.gwern.net/Danbooru2019 and our TFRC TPUs very soon to test the question of how well S2 can scale up...
I have definitely gotten Harry Potter fanfic out of https://talktotransformer.com/ on several occasions. Not, as far as I can tell, long passages or entire sentences copied from Rowling, but clearly identifiable uses of the Potter characters and worldbuilding, which (IIRC) makes it a derivative work under US copyright law; even though GPT-2 doesn't have a copyright in it, Rowling does. Probably the US four-part fair-use test would keep you safe in that case.
I wasn't even feeding it Harry Potter seeds, just generic fantasy prompts. For better or worse, none of them seem to have made it into Dercuano.
· As a slightly rigged example, given the prompt "She pointed her wand and", I get:
She pointed her wand and said, "Greater Nullam (sic) Albus, a man with a beard."
Harry could see a brilliant man in his mid-50s in a maroon silk robe, in a room of softly lit glass and gliding animated candles with hair streaming down his back. It was Muggle science professor, Professor Trelawney.
Professor Trelawney said, "Good evening, Harry Potter. I am pleased to meet you and to see you very well indeed. I have thought of many interesting things to discuss with you today."
"Professor, could you please give me a single book?"
Trelawney nodded curtly. "I would be glad to, but only for a very small
· Given "He struggled with the difficult spell", it came up with a football article. On a second occasion it came up with this, which is sort of magical but definitely not Harry Potter:
He struggled with the difficult spell. "For a few weeks I was in denial, angry, and certainly depressed. I felt like my life was no longer worth living."
So he found a therapist and told her he was a sex addict, something that was not his problem. "Then she said to me: 'Maybe you're right,' " says Gonzales. "And that made me think, OK, this could be my problem. Maybe I'm not broken, and maybe I can learn to be okay without sex."
In August 2012, two years after he had started having sex again, Gonzales woke up in his own bed with his wife next to him. He also had a permanent erection. The next night he had a similar experience. "When you can start having
· Given "Both wizards walked past in their long robes.", it came up with
Both wizards walked past in their long robes. The feather fell from the left sleeve and into the gleaming water of the lake.
"That's enough," Hermione, Harry and Ron said in unison. They turned and walked towards the large tree the stone bridge rested upon.
"Yeah," Ron said, a little amused. "This bridge is a bit much."
"Nothing worth dying for," Harry said, having stopped dead in his tracks as they approached the stone bridge. "C'mon Ron, let's go."
Ron stood there and stared at the bridge.
"Huh," he muttered. "I just realized the rest of us were behind you. I guess we could use your strength. I could do the traveling myself."
· Given the phrase from Tolkien, "Suddenly the Elf gave a cry", it invented part of a book on fairy tales:
Suddenly the Elf gave a cry of frustration, and felt there was nothing he could do to stop the man from chopping the tree, and said, "But my dear young master, you see the young Elk. . . ."
Druid Stewart halted in his work and glared at the Elf with utter fury. "I am more a man of love than of pride," said he, "and I will not have my own lofty feelings crushed by such worthless lumps.
THE ANGELS!
"In the time of my youth, as I recall it, I once stopped to feed the cottage lambs, and I found with them a man and woman.
The man was clearly a bad sort, for he sat listening to the old folk talk and eating his
· Given "The dragon attacked then", on one occasion I got this fragment of Buffy fanfic:
The dragon attacked then ate the village and two of the teenage slayers. The Avatar managed to survive by eating one of the little pets then led the other slayers to kill all of the dragons and protect the villages.
Credits
Written by: David
· On a second occasion it came up with something that seemed a little more original:
The dragon attacked then. Not ever having heard of such an animal before, we spent the entire time staring at it as it roared and thrashed about, causing a mass panic among the passengers who were shocked by what they were witnessing. Most of them could barely stay in their seats with the movements from the beast and barely managed to take photos with their phones. After what seemed like a very long time, it finally came to a rest with a loud thud and the passengers finally gave up, walking to the exit and screaming. The scene was so surreal that I actually forgot for a moment about the fact that we were already in an airplane and about the impending disaster. The passengers boarded a train and continued their journey.
Eventually we reached the Central Railway station, which is more
· Given "I consulted the potions textbook", which is also arguably a bit rigged (how often do potions and textbooks occur together outside of Harry Potter?) it comes up with
I consulted the potions textbook. And here I learned the most astounding fact of all. I had made a potion that restored the mind. It was just that simple. That's right, it worked. It really did. You're getting better. You're really taking things in." Harry lay in his bed and felt a little groggy, but was impressed to see that his eyes were clear, and his mouth was a little dry, but he felt better than he had the previous day. He really had learned a lot. He felt himself getting ready for bed, staring into his window as he thought about the very event that had happened in the past few hours.
It all started with the burning desire to do what he wanted with Ron and Hermione. That made him go to the
· Given "The wizard leaned on his staff", I get
The wizard leaned on his staff and closed his eyes. "But that has not changed. Indeed, the more I have this power, the greater it grows. I never truly felt the full extent of it in the beginning. But today you are stronger, taller, smarter and more powerful. That is a good sign. And because of you, I will see the completion of one of my greatest projects."
The headmaster rose. He walked toward the fire and patted the warlock on the back. "I have waited long enough. You have finally given me an opportunity to become a true student of magic. To study the infinite power I hold, to master it and bestow it onto others. And you have done all this in a single night. It is a privilege I
which starts out not very Potterish but then veers Potterwards.
· Given "The goblin turned the corner", I get
The goblin turned the corner and saw the battle-axe bobbing in the water. His eyes lit up as he stared at the ax and glared at the unicorn. Then the goblin looked back to the pike, quickly raised his wand, and cast the spell: "Search!". "What's the witch's name? Search", yelled the wizard. "Bill Luscombe," the goblin roared. "Vocabulary charms? Go ahead", the wizard yelled back. The goblin bent down, pulled out the memory and said: "Name: Gilderoy Lockhart." The memory was very old. Dumbledore had called it on Dumbledore before and it had been quite a good memory. "Congratulations, dear boy", said the goblin. Then he added "Thank you". Dumbledore said "
· Given "He turned the ring on his finger", I get slightly incoherent Harry Potter gay slash fanfic:
He turned the ring on his finger and it lit up with a red glow, then he started to levitate out of the ring. The ring let out a slight hum when he did so.
A sharp whistling noise could be heard coming from the next room over and the door to that room clicked shut.
Ron smiled and pulled the tall blond man to him, "You know, I've never heard anything like that before, what are you doing?"
Harry smirked and pulled Ron's shirt off, "You're probably wondering why I don't want to go out alone after everything we just did."
"Well," Ron commented, "What were you doing before we met?"
"Well, my people…" Harry said hesitantly, "Some
· Given lines from Tolkien without any proper names in them, it tends to deliver surreal medieval-genre tales:
'Nay! The riders are little more than five leagues distant,' said ilexar. 'I will give you a horse, and go to Ghent to see about our caravan.' They gave him the horse, and ilexar started off. Then a porter of Alcinous' school, who knew the path that led from Knokke to Ghent, said to him: 'Why, of all men I have never seen thee go to Ghent. The full moon is shining, and a queer light is flashing from under the trees on the road that lies by that place. The ponies ride to Ghent in their night-gowns, and set out at the dawn, but they don't come back till night. What chance is there
· Or Legend of Zelda slash fanfic:
Wonder came into the man's eyes, and he laid the weapons hastily before him.
"Y'know, this is my side of things." said Zelda.
"Oh, yeah?" asked Link.
"I'm the doctor. You're the general. You can tell that this one's on edge, but you can't stop her. Plus she's got a gun and two grenades she's using. It's obvious."
"Sorry, I guess I can't. Looks like we've got to make her tell us her side of things or not pass out."
Zelda laughed and pulled Link into a hug.
"Don't worry, it'll be fine. I'll make sure she doesn't do anything too crazy."
So if you prompt it with talk about wizards or magical creatures or magical items using contemporary language, GPT-2 gives you, by default, Harry Potter fanfic, something like half the time. Or Legend of Zelda fanfic. Or Buffy fanfic. Anyway, it's emitting lots of text that clearly carries with it copyrights of human authors, under the current very expansive US copyright law designed to privatize fanfic. Maybe the situation with images is better.
> Anyway, it's emitting lots of text that clearly carries with it copyrights of human authors, under the current very expansive US copyright law designed to privatize fanfic.
That's not copyright: you aren't copying the fixed expression of a Harry Potter work. (Where does any of that quote any of the Harry Potter books?) That's trademark, at best.
There was a pic of an older guy, but what was unsettling was the woman peeking over his shoulder; like a friend or a family member, leaning into the photo.
Her eyes were melting into black and white streaks curling down over her sagging cheeks; her smile was perfectly white, and just a little too long.
I refreshed on instinct, and haven't seen anything like it since.
You have to imagine some strong AI in 2030, video chatting with people using fake human avatars, and people can sometimes tell because there's weird fleshly artifacts moving in the background.
I feel like this is "the 2nd uncanny valley", on the other side of realism. It's more uncanny because the face is so realistic, indistinguishable from a real person. But the weird mistakes around the face reminds us that this is made by something entirely alien, with an alien understanding of what it means to be human. It's like The Thing.
Yep, I've seen some ones like that too. The NN is quite good at knowing what people look like, but it's much worse at knowing what people tend to be next to and sometimes you get some horrors.
I've seen some of those before - the algorithm is a bit strange about people who hang around the edges of the border and sometimes create some fairly horrifying abstractions of humans
To my mind there is a massive need for us in the software world to pause and think about the ethics of all of this. Engineers more broadly have well developed ethical codes, we should too.
So are you saying software developers should have to take the equivalent of a Bar exam in order to practice software development in a given state? That way ethical code violations have teeth because you can be debarred.
As of now anyone and their dog can practice software development and I'm not sure what a nebulous "ethical code" would imply for them.
Probably something more like a PE. From the NSPE website a PE entails:
"To become licensed, engineers must complete a four-year college degree, work under a Professional Engineer for at least four years, pass two intensive competency exams and earn a license from their state's licensure board. Then, to retain their licenses, PEs must continually maintain and improve their skills throughout their careers."
Most PEs that I have worked with are very diligent. You don't get yahoos going that far. Generally, they are top-notch engineers. Think '10X' engineers, but for oil wells and churches. I mean, look at any large structure on Wikipedia, most of the time you'll find the engineer's name in the side placard, not just the firm's name. PEs are serious, dedicated, and smart people. I mean, their stamp is on those documents and they go to jail when the bridge fails.
So, for software, if you have an exam of the PE's caliber, you essentially guarantee '10X' engineers. The workforce is more highly paid, yes, but it is of a much higher quality. Bugs get solved faster, stand-ups don't last 30 minutes, the wiki is updated by everyone, etc. Employers know that their candidates really are good stuff, less whiteboards or take-homes are needed. Also, possibly when something messes up, the engineer is the one responsible, not the company. So offloading the insurance to the employees is not the worst thing for a company.
Honestly, it would not be a bad system and I think would be a much calmer and better one.
i'm seeing some sources that say it was discontinued after 2019 due to lack of interest, and the main "PE" page no longer links to the software PE exam.
I've often wondered if software engineers would benefit from the accountability of licensing like other engineering disciplines, especially as it pertains to safety critical systems.
It's a bit shocking for me to know a mechanical, electrical, or civil design in, say, a power plant may require a Professional Engineer to officially stamp it but any ol' software engineer can write code on that system and fire it up. My understanding is that NCEES lists a controls systems license but I have yet to meet a single person with it or hear of a project that required it.
To a certain extent, I wonder if software development rigor/oversight is in an adolescence similar to what it must have been like for mechanical systems during the early parts of the industrial revolution.
That's because other than the name "software engineering" has nothing to do with "engineering", as in a prescribed process and set of standards and practices that are required to be followed in the design and implementation of projects.
This was a big debate in the 90s. The "industry" headed it off at the gates (partly thanks to communication media such as ... usenet :) It might resurface, but I expect to hear the same arguments used against it, all over again. I'd be suprised if it ever came to pass in the same sense as PE requirements.
1) programming is more like an art than a science, and so should not be regulated like engineering
2) the capital/labor requirements for programming are minimal - basically anyone could do it - so we should not worry about trying to license something that people could (and do) do at home.
3) there's no agreement about what would be required to get a license.
4) free software would die
Of course, some of these really do not hold up to close inspection. It's true that there is an "art" to programming, but not much more than there is in architecture for example, and there's no reason you couldn't require a license for cases where you do work for others rather than yourself. Since the 90s, it's become a _little_ more clear what the components of a s/w dev license might be, but perhaps still not enough to really pin it down. And yes, it would probably kill the free software world.
I wonder if they could parallel other engineering disciplines and give exemptions for specific areas or other avenues to ease the burden.
For example, as long as a certain percentage of principal engineers are licensed, others within the company don’t have to be. Similar with working under a licensed lead or specific article industry exemptions.
It would also give some power to the developers over management in key decisions. I’ve seen managers try to back engineers into a corner and when the engineer ultimately said they wouldn’t sign off on a design, management backed down because they didn’t have an equivalent to “just ship it now” and get it stamped later.
I understand the fear of it, but sometimes it seems like the sw development is run like the Wild West.
Arguably software engineering compared to engineering in a broader sense has a bigger issue with the “we did this because we could, not because we should” attitude.
I feel like a “formalised ethical code” would result in more clever approaches of avoiding it. We don’t need to look far to find the examples (looking at ya ad tech)
I think it would have to start with either regulation or customers requiring it.
I.e., contractually require someone licensed (where the license carries an ethic standard) or legally require it in areas that are 'in the public interest'
The entire point of a code of ethics is to provide a strong minimum standard to keep the morally bankrupt in line. We don't write them for moral people.
That would be a horrible idea. Software developers should be encouraged to think about and discuss ethical issues by other software developers, employers, universities, and people in general.
We need increased awareness of ethics but we don't need more government involvement or any other form of coercion. Governments are already far less ethical then pretty much any other organisation or institution.
This "government involvement is bad" is a crock. Regulation is what makes a society. As an institution, they are no less "ethical" than corporations, in fact, neither ethics nor morals are relevant to corporate or government bodies.
Morals and ethics are about people. Laws and regulation are about how those morals and ethics are reduced to practice.
Voluntary cooperation and peaceful competition is what makes a society. A government is just a group of people who attempt to maintain a monopoly on violence and use it to enforce their will on others. I certainly think they are less ethical then people who do not attempt to enforce their will on others.
Government created laws and regulation are just threats. They tell you nothing about what is right or ethical. They only tell you how a particular group of people want you to behave and what they will try to do to you if you refuse to obey. Anyone with enough capacity for violence and willingness to use it can create a law but no amount of force can change what is right or wrong.
The Free Software Foundation has been promoting a well-developed ethical code for software engineering since the 1980s: https://www.gnu.org/gnu/manifesto.html
It covers most of the widely debated issues in software today, including spam, malware, DRM, invasion of privacy, undue influence by one participant in the ecosystem, copyright infringement, and bait-and-switch terms of service. It even addresses the question of writing software that could be misused.
It isn't widely accepted because a lot of programmers don't like it. They don't like it because they make money by violating it.
Agreed on taking a collective breath and evaluating.
For "real" engineers, is it fair to say their ethical questions are more easily constrained than those in software? I assume it usually reduces to "should I accept a lower-than-average safety margin to save money on a part/structure that might fail and hurt people?" Maybe "should I make this vehicle accommodate extremely tall people?" or "can a child hurt themselves with this device?"
Honestly lawyers feel like the closest analogue to software engineers.
> For "real" engineers, is it fair to say their ethical questions are more easily constrained than those in software?
For capital "E" Engineers (i.e., those with a State license), I would say yes. They are required to abide by specific ethics and some states require regular ethics training. If they certifying a design, they are putting their name/license/livelihood on the line that it meets a minimum standard. This standard is usually defined by specific codes.
The problem I often see in software is that many codes (if they even exist) are looked at as a "nice to have" not a "must have", unless they are contractually obligated. If I show a code-base doesn't adhere to a specific industry standard, the reaction is often "meh". This is in contrast to other engineering disciplines who tend to begrudgingly acknowledge they have to re-design to meet such a standard.
The issue with some of your examples is that much of engineering work happens under an "industrial exemption" meaning they don't have to be a licensed engineer.
Copyright 1999. I was taught ethics in my college work, so clearly everyone else just ignores it like they do for the various other fields where there is an ethics code. People suck.
The issue is that there's no pressure on employers to require it, so there's no advantage on e.g. employability or salary from getting these codes.
I live in the UK and was considering going through the British Computer Society certification process years ago, but dropped it, because while I do care about the ethical side etc. the additional amount of effort it'd take to document skills I already had was not worth it when prospective employers in the fields I've worked in just did not care.
It has a "machine learning" based projection mode where you can project a single photo of a face on the head of a character mesh and it will create somewhat believable set of 3D features from the photograph on the model. It was fun to take a couple of the generated faces from this site and apply them to the models.
I imagine you could do a lot with just those two things if you had the talent of the artist in the video I posted above. You could create fully rigged and clothed characters for games in a fraction of the time it would take to make them entirely from scratch, and for much less expense and cleanup (I would have thought) than full body scanning.
Wow! Imagine allowing people to upload their profile picture in and RPG and suddenly their character actually looks like them. Imagine playing online with your friends and all of their characters look just like them.
So, the specific person may not exist, but likely someone very, very similar in appearance does.
I've come across people in life that I would have sworn were twins. I'm curious how combinatorically low the number of facial features need to be before people resemble one another.
I grew up in the midwest in an area where a lot of Germans settled many decades before I was born.
Later in life I traveled to Germany and just felt ... comfortable. I didn't speak the language or anything, wasn't talking to a lot of people, I simply saw all these faces that I felt I knew...
It was a very sudden change as I had already been to England, France, Italy and etc where things felt foreign and different. Then blamo, Germany and here are all these people I recognize / half recognize.
Similar situation, in that my family line is essentially 100% German, although I grew up elsewhere. When we visited my grandparents' home city, it was kind of a shock how much I looked like everyone else. It didn't make me feel particularly comfortable or uncomfortable either way, but it did surprise me how, well, related we all looked in a way that had never happened to me before.
My favorite part of the trip, though, was getting complimented on my English by other friendly but clueless tourists who asked me for directions and the like. I told them thank you, my parents were very insistent that I practice my English from a young age. Technically the truth! :D
But those people are in fact related. If you could test their genetics, you could prove it. They have different variations on the same face that some successful ancestor of theirs once had long ago, which was conveyed to them in their own generation over hundreds of years by the inheritance of dominant genes.
You will find that people who have such similar faces also tend to think and act similarly also. Not a coincidence.
On a similar tangent, privacy issues aside it would be fun to see if you had an unknown twin on social media. Upload your picture, see if there's other pictures that look like you but have a completely different friend graph.
Such service used to exist: FindFace. It was essentially for finding people's VK accounts having only their photo, but finding people who look like you was one of the popular use cases. It was since closed (kinda) due to privacy outrages.
A similar one called FindClone has recently appeared, though I haven't tried it.
My gut reaction is that the model does an apparently credible job of simulating faces, but cannot take into consideration their underlying biology. I would think a lot of the features these non-existent people have might be biologically incompatible for reasons we don't know. Facial features are incredibly sensitive to exposure to gene:environment interactions. So, I have doubts today they look like anyone real.
When generating a new person (or whatever it is that does not exist) can you know that it isn’t like any of the images that went into the training data set?
How likely is it for it to actually exist after all?
Some of the backgrounds are surprisingly consistent, but some are distorted in ways that can only be done by a deep NN that sort-of knows what it's doing. The fact that the face is (almost always) still completely believable seems like good evidence against overfitting.
The model has 18x512 parameters, and all 9000 parameters are 32-bit floats.
Even assuming only 16 bits of randomness for each parameter (to keep them small enough, so you don't get too wild, because the center of 0 is a pretty homogenized face), 16^9000 is a lot of permuations of faces.
They will, of course, borrow from all of the 70k faces on Flickr that power StyleGAN, in varying degrees.
> Even assuming only 16 bits of randomness for each parameter
Well, this is the question. How do you know that you can make that assumption? Also even though you have ~9000 parameters they could be highly dependent.
It only took me a while to notice that this is an actual human face in the background, just photographed from a peculiar angle and bearing a peculiar expression - not a neural network artifact.
In addition to the background, if you can see earrings, that is usually a giveaway. The generated faces seems to always have different earrings in each ear. Also glasses disappear into the skin, and ears are often different on each side and/or distorted.
Now, the next obvious question is, how do you pick one of these non-existing persons and generate images of him/her doing different things in different places? (Or, of course, real persons. Is there a service where I can upload my photo and it starts to generate images of me doing random - or not so random - things?)
You can reverse-engineer faces into the original random seed/embedding, and then tweak the embedding to 'edit' the face. (There are a bunch of tools and Colab notebooks for that, but the best tool is definitely Artbreeder, which does a lot of other GAN models too: https://artbreeder.com/ )
However, since the model only does faces centered in an image, you're restricted to different headshots: make them smile or frown, or wearing sunglasses, or different hair color, or animate lip flaps, yes, but not 'doing different things in different places', unless you can figure out some way to insert a headshot into a bigger photo.
I remember it from the subproject, < http://thiscatdoesnotexist.com > although that doesn't seem to have improved. It feels more like a "this transporter accident doesn't actually exist dot com". (Just a joke; its fun.)
What's the current state of the art on controllable GANs? Like if you wanted to build a thispersondoesnotexist but with the ability to control e.g. gender, smile, sunglasses, ethnicity?
I believe S1/S2 still are effectively SOTA for static face image editing. (Note that S2 added a projector and has better latent spaces than S1; see the paper.) You can play around with them at https://artbreeder.com/browse among other ways.
An artist friend of mine is using this technology in an interesting way .. she's drawing regular street portraits of people she meets, and then using that as the seed to generate new pictures of people who do not exist.
It has produced some extraordinarily disturbing art, especially for those of us who were her subjects. Some of the images are so wild and out there, it really is an adventure into ones own psyche. I see so much in the generated images that creates an intense emotional response, and its very difficult to differentiate between the lines of the artist and the lines of the code, at least in a way that is easily discernible. Her art as the seed definitely amplifies things - the emotions of eyes, the despair of cheeks and lips - in a way that the generative programming enhances, 1000x ..
Very interesting stuff, and I'm looking forward to her exhibit of this work - especially the live demonstrations.
I'm intrigued about the possibility of using these as stock photos - it dies away with the need for model permissions (or does it?). Presumably the copyright on the image resides with the person who ... programmed the neural net? Trained the neural net? Is running the server?
I would guess the one picking the photos and training parameters would have the largest part of the copyright. Whether courts will decide others have some rights to is TBD.
The persons whose images the net was trained with also might have, certainly if the network was trained with only a few images.
How does copyright work when you have no idea the photo even exists? Are these photos dynamically generated on the fly when a page is loaded? Because if so, then the visitor is the only person to have seen that particular photo, and there'd be no way for any copyright owner to even know they might hold the copyright over that particular photo.
>I'm intrigued about the possibility of using these as stock photos
Another (darker) possibility is to generate copyrighted images. Once you have (say) a trillion photos, it will be hard to generate a new one that doesn't look (in essence) like one you already have.
StyleGAN (v2, which is this, and v1 before it) is spectacularly high quality.
Even more interestingly, you can interpolate between faces very smoothly. (For instance, I was able to give some face off the street a drag makeover, by interpolating that face plus or minus the face of a drag queen minus that performer's face out of drag: https://leebutterman.com/assets/lsb5-plus-trixie-minus-brian... )
But it can be tricky to encode faces if they look out of the ordinary, where "ordinary" is the 70k faces on Flickr that trained StyleGAN.
This has improved so much since the last time I checked it.
It's interesting to see so high quality on image generation using neural networks, while text generation using gpt2, which seems a lot easier at first, is glorified nonsense.
Is part of it deliberate avoidance of women with both ears + earrings showing? That was one of the tells before...one earring or a mismatched pair. Now the women mostly seem to be in poses where you can only see one ear, or no ears (hair covering them).
There is, this time, much less craziness with hair sprouting from the wrong places.
I noticed there are very few black people and none with very dark skin. But there are lots of people of many other ethnicities. Was dark skin avoided because it's inherently harder for the algorithm?
Interesting. I just refreshed the site 100 times to get a large sample size and the darkest skinned image it produced was this girl (https://i.imgur.com/ggJIG94.png). Her skin is much lighter than many of the people in that sample of the training data.
Assuming that that sample is representative of all the training data, it must be that the algorithm is 'choosing' not to produce images of dark skinned people. Perhaps because it isn't very good at them and 'knows' this? I wonder if there are other features it's deliberately avoiding?
> Assuming that that sample is representative of all the training data, it must be that the algorithm is 'choosing' not to produce images of dark skinned people. Perhaps because it isn't very good at them and 'knows' this? I wonder if there are other features it's deliberately avoiding?
An algorithm would not have this kind of insight. (This is just a GAN trained on face data.) To say that an algorithm like this is "deliberately" doing anything is a misunderstanding.
Most generators like this that I've seen, like GPT-2, require you to provide a "seed" of some sort, such as a sentence fragment. They then build off of this seed. I don't know if this is implemented that way, but if so, perhaps the developer provided a set of seed data that leads to this result. There may also be a sort of "averaging" involved (see another commenter's note about the women having similar features), and depending on both the training data and the seed, this may result in a preference towards certain features.
The original FFHQ was designed to cover a lot of ethnicities, but maybe it didn't cover very dark skin enough? Alternately, it might be related to the 'truncation trick': images at the 'edge' of the latent space get dropped to assure high quality at the cost of covering the entire distribution. If very black skin got mapped to a compact cluster at the edge, say, the truncation trick could hypothetically lead to few such samples despite representation. To test this you would generate images with very high psi to see if they come out more often.
What is the likelihood that the generated person does exist?
I got the feeling this could turn out the same as people using fake email addresses to later discover that the domain in fact exists (that's why you should always use example.com).
Good question. I would guess it's very low if we're talking exact match. It would have to match N features. There's effectively an infinite number of combinations, many more than the 8 billion people curently living.
In reality, there are already real-world doppelgangers, where a typical observer would confuse two (unrelated) individuals. I'm guessing that could happen with a synthetic image too, but it's probably not any worse than a real-world matching error. (The bigger concern is the related one of deepfakes, which don't need any kind of synthesis, they just act directly on the real person's image.)
What's really amazing to me is the rendering quality, if that's the right word for it, the resolution. This seems different to me than the "structural realism" in the sense of how plausible the face is in it being representative. Not sure if there are terms for this but it's amazing to me how realistic things like skin texture are independent of "higher level" features.
I wonder could the generated faces now be used to actually do further training? Even from scratch for other, similar GANs? Doing so might remove any privacy concerns about real world data being used to train these things (going forward)
I looked at about 50 faces trying to find one with a strong jawline or hard cheekbones. Although the features and hair are very realistic, the algorithm does seem to stray away from distinctive features. The women especially have uniformly round and soft faces.
Makes me so happy I took "Computers, Ethics and Social Responsibility" in college.
These algorithms are tools and at their worst they are powerful weapons - much more dangerous than some nuclear bomb that wipes out a few hundred square miles.
The background, clothing and especially other people in the image are still a dead giveaway that it's not a real person. It's kind of weird how the algorithm borks something relatively simple as the texture of clothing.
I don't know but I imagine somebody might want to generate an algorithm that puts random clothes on people. There's so much money going on in the fashion industry and beautifying. If you can see how a new shirt fits on an existing person you can also fit a shirt on a fake person right?
I just realized the potential this has for ruining dating apps by filling up the majority of people on the program with fake generated people that just waste enough of the participants time that they never get anywhere.
Some people generated there seem oddly familiar. Too bad, I am not good with remembering names or I would frequently go: "oh that is <xyz> from <abc.inc>. How did his/her picture end up here?"
Om my book “x doesn’t exist” demonstrations are absolutely worthless if they don’t as a minimum show at least one “closest” member of the training set.
You could set up a script that just randomly picked one out of 50 images of my closest family and no one would know the difference from going to it.
The point is that you aren’t demonstrating the implied value in the generated content. Deformations and artifacts are easy to add to images. The concept here is that this is supposedly an image of someone who doesn’t exist, if what you see is just a test-set image that’s deformed or has an artifact added to make it look like it was automatically created then it’s not actually doing anything... but the point is that you have no frame of reference to judge that because you are only seeing the supposed non existing face.
So it’s only impressive is you take it for granted that it’s doing what is claimed.
It’s impressive if you assume it’s impressive.
What I’m saying is simply that if they want to claim it’s impressive you should be able to judge that based on the displayed result itself. Just include the nearest match and it’s simple as hell for everyone to judge if it actually did something cool or not.
Luckily, it doesn't seem that this approach can be sensibly extended to generate entire scenes, or multiple entire-body photos from different angles. Believability breaks down tremendously in less constrained datasets- see https://www.thiscatdoesnotexist.com for example.