Catch up on stories from the past week (and beyond) at the Slashdot story archive

 


Forgot your password?
Close
typodupeerror
×
AI Technology

'This Person Does Not Exist' Website Uses AI To Create Realistic Yet Horrifying Faces (inverse.com) 145

Posted by msmash from the stranger-things dept.
A website that uses AI -- Generative Adversarial Networks (GANs) -- to generate photos of people who do not exist is circulating on social media and forums this week. A news writeup adds: Every time the site is refreshed, a shockingly realistic -- but totally fake --picture of a person's face appears. Uber software engineer Phillip Wang created the page to demonstrate what GANs are capable of, and then posted it to the public Facebook group "Artificial Intelligence & Deep Learning" on Tuesday. The underlying code that made this possible, titled StyleGAN, was written by Nvidia and featured in a paper that has yet to be peer-reviewed. This exact type of neural network has the potential to revolutionize video game and 3D-modeling technology, but, as with almost any kind of technology, it could also be used for more sinister purposes.
This discussion has been archived. No new comments can be posted.

'This Person Does Not Exist' Website Uses AI To Create Realistic Yet Horrifying Faces More

'This Person Does Not Exist' Website Uses AI To Create Realistic Yet Horrifying Faces

Comments Filter:

  • Should I be concerned? (Score:5, Funny)

    by Nidi62 ( 1525137 ) on Thursday February 14, 2019 @02:25PM (#58122446)
    I pulled up the website and the picture looked a lot like me.....

    • Re: (Score:3)

      by sinij ( 911942 )
      Activate protocol 2! This is not a drill, activate protocol 2! The human consciousness simulation escaped the containment.

      • Re:Should I be concerned? (Score:5, Funny)

        by Red_Forman ( 5546482 ) on Thursday February 14, 2019 @03:08PM (#58122740)

        Dude, calm the fuck down a bit. We know it's not a drill. It's a screwdriver.

      • I'm curious....are these images 'copyrighted' as that they are generated?

        Just thinking it would be a useful tool if you needed to just put a face on an image or pamphlet, etc......

        • are these images 'copyrighted' as that they are generated

          That's an interesting legal-theoretical question and it might make a good topic for a law school essay.

          As a matter of mere practice, however, given that no two runs of the program should produce identical faces, it seems unlikely that anyone would even think to look for a watermark (assuming there is one) or other identifying feature connecting the image to the the putative copyright holder. TLDR: You're unlikely to be caught.

          • All the pictures I got had a "blotch" at the top. Probably not an official "watermark", but seems reasonably unique enough for them to argue the picture is theirs.

            Ahh, scratch that. Got some without the blotch.

            • The quality isn't that great.

              Many of the pictures I saw had lopsided faces, and many others had "blotchy" looks, like someone had a skin graft from a donor whose skin wasn't quite the same color.

              All in all, I don't think it's ready for prime time.

        • There's a generated sample set of images on the same site as the paper. I'm assuming the guy who set up the site is serving up those sample images.

          Here's what the NVIDIA github repo has to say about the datasets:

          "All material, excluding the Flickr-Faces-HQ dataset, is made available under Creative Commons BY-NC 4.0 license by NVIDIA Corporation. You can use, redistribute, and adapt the material for non-commercial purposes, as long as you give appropriate credit by citing our paper and indicating any change

          • Damn, nevermind, apparently the website is using the canned generative model for generating faces and is dynamically creating new ones on the fly. I have no idea how you would license the output...

            • Yeah I refreshed over & over, and got a unique face each time; imo, only one face looked slightly creepy-ish to me.

    • I just tried it, and got a picture of a normal looking young woman. Not "horrifying" at all.

      The site is really slow, and sucks up a lot of local CPU running JavaScript.

      • The picture itself isn't "horrifying", it's the fact that the picture (which as you say looks perfectly normal) isn't real. That "normal looking young woman" is a complete fiction of a neural network, despite looking (to the human eye anyways) perfectly real. People typically assume images of people are real, so being able to create completely realistic looking humans out of nothing allows an entirely new level of fake news.

        • The picture itself isn't "horrifying", it's the fact that the picture (which as you say looks perfectly normal) isn't real.

          I don't see anything "horrifying" about it. GANs are interesting and the results are sometimes impressive. But "horror"? No.

          Also, the only thing "new" about this website is that the images are supposedly generated on-the-fly. The faces are not much different than published results for other face GANs. GANs have been around since 2014, and high quality face generators have been around since 2017.

        • Most of the stuff you see in magazines is just as unreal, the work of Photoshop, Gimp, whatever, makeup artists, hair stylists, and lots of doctoring..... not to mention the plastic surgery.

          Ok, if there's an ear in the wrong place, it's horrifying. Otherwise, it's just new CGI for the next LOTR episodes.

  • Bet you've not heard that term in a while!

    The site for me is loading the image slower than an 80's fax-modem set to highest resolution.

  • The Technology Is Already Being Used Negatively (Score:5, Interesting)

    by dryriver ( 1010635 ) on Thursday February 14, 2019 @02:30PM (#58122476)
    A friend who lives in a repressive country where over 90% of the media is pro-government propaganda told me that the when he does a reverse image-search of the little "author image" next to the opinion pieces lauding the government's actions, neither Google nor Tineye can find the person online. It seems that the opinion pieces are authored by supposed "journalists" who's face can only be found next to the opinion piece - and nowhere else online. The authors names and their face images are fictional, even though their faces appear to belong to a real person. It seems like this tech was around before Nvidia supposedly "pioneered" it.

    • Re:The Technology Is Already Being Used Negatively (Score:4, Interesting)

      by 93 Escort Wagon ( 326346 ) on Thursday February 14, 2019 @02:50PM (#58122638)

      This actually sounds a bit like the bulk news outlet services which a lot of small newspapers use nowadays - I believe it was covered in an NPR story a few years back. The news items are collected and aggregated overseas, sometimes rewritten slightly to "localize", and then released using a made up "generic white American" name (and sometimes a stock photo) for the byline.

      • The bulk news outlets are Reuters, AP (Associated Press), and less frequently UPI (United Press International). They hire reporters to travel the world writing up stories, then sell the stories to smaller news services which cannot afford to send one of their own reporters overseas for every single international story. When properly attributed, their stories start with an (AP) or (Reuters) or (UPI) tag at the beginning of the text. When unattributed, they're easy to pick out since a google search of a sn

    • Re:The Technology Is Already Being Used Negatively (Score:4, Insightful)

      by ScentCone ( 795499 ) on Thursday February 14, 2019 @02:57PM (#58122690)
      Or ... someone could have spent 30 seconds in Photoshop with half a dozen normal face shots of different people, and produced exactly what you're talking about while setting up some fake journo profiles. That's been done for years. And is not the same as an AI-ish thing synthesizing faces.

    • Re: (Score:2)

      by Vanyle ( 5553318 )

      Why wouldn't they just leave the images out? I go on several news sites where it is difficult to find the article's author's image

  • I can remember dabbling with fractals as a basis for creating artificial landscapes back in the 80's. (And no, it wasn't my lawn which you can kindly depart from.)

    It's interesting to see the level of detail, and the types of asymmetries and other 'imperfections' . It'd also be interesting to determine statistical probabilities of a reasonably close match to an actual person.

  • Slashdotted again!

  • Not 100% (Score:5, Interesting)

    by Nidi62 ( 1525137 ) on Thursday February 14, 2019 @02:45PM (#58122602)
    It's not 100% realistic yet. The faces look good, but with the hair and body there are a lot of issues. I pulled up one picture and the face is a significantly lighter tone than the neck/chest area which leads to issue number 2, the woman does not appear to have a neck. The ears are just close enough to realistic to notice they aren't right, and the hair coming down over the body isn't correct either. But the facial features are really good.
    • Got one that seemed to have an earring embedded in her cheek...

    • I've seen a few examples where there's a weird stretching of the mouth or, in one, fingers coming out of the person's chin. My guess is that it's sampled a huge number of images and is stitching these together in some complicated fashion to create a "does not exist" person. In the event that the image contains an odd object (microphone near mouth, fingers by chin), the AI chokes and the image gets weird.

      Still, it's very impressive. Now enhance this so it's a video and have the AI create a realistic (but not

    • Some of those faces can be instantly identified as fakes. Some have eyes that are beyond uncanny valley and down right hideous.

      Others... well I saw at least 2 images where I couldn't find a fault despite looking for many minutes.

    • Re: (Score:2)

      by sad_ ( 7868 )

      depending on the usage it's damn near perfect.
      as the example given in the summary, if this tech would be used in games for example, you would barely notice these defects nor would they matter that much.

  • Do something useful with it (Score:5, Interesting)

    by Comboman ( 895500 ) on Thursday February 14, 2019 @02:58PM (#58122692)
    Someone should write a bot to post each new photo to Facebook (along with a randomly generated name) to salt their facial recognition algorithm.

    • Re: (Score:2)

      by PKFC ( 580410 )

      If I had one of these photos, I wouldn't have been banned from Facebook. That and had a name that wasn't "Jus D'Orange"

  • Horrifying maybe for the wrong reason (Score:4, Interesting)

    by FilmedInNoir ( 1392323 ) on Thursday February 14, 2019 @03:14PM (#58122774)
    Some of the pics have noticeable artifacts that give the appearance of a severe scar... or possibly gills in some cases.
  • As it is, the created faces just look like random people to me. There is nothing "horrifying" about it.

  • Generate the rest of their bodies as well and we'll finally have a replacement for Tumblr!

  • I didn't know the industry needed up'ending, but this could forever change "Big Hero-shot".

  • Obvious stock image input (Score:5, Informative)

    by spaceyhackerlady ( 462530 ) on Thursday February 14, 2019 @04:19PM (#58123132)

    It's obvious the neural network was trained on stock images, including lots of celebrities. I saw Angelina Jolie's eyes, Brad Pitt's jaw, Caitlin Jenner's hair.

    ...laura

    • It's obvious the neural network was trained on stock images, including lots of celebrities.

      Probably more the latter than the former, since there are quite a lot of celebrity datasets, like Celeb-A, VGG's Celebrity in places, VGG's Celebrity Together dataset, VoxCeleb and so on.

    • I noticed that too. (btw hello, we have not talked in a while)

      Then it occurred to me, if I were not aware that I was looking at manufactured images, I would not have noticed on casual inspection.

      I went through a few dozen images. Amazing stuff. Still needs human caretaking to ensure a perfect final product.

      But yeah, "I saw Angelina Jolie's eyes, Brad Pitt's jaw, Caitlin Jenner's hair.", too. It was weird. Like I was seeing some aspect of what makes up a real human. I saw one photo that looked absolutely rea

  • They all have nearly perfect teeth. And the shape of the two front teeth is really similar between all people that show their teeth. I don't think these are real people...

  • What if you see one of the images and fall in love with the person in it? They don't exist, but you can't get the idea of them out of your mind and you are ruined. Forever.

    That would be a nightmare.

  • 1991 called and they want credit back for this new face "morphing" thing.
    http://www.criticalcommons.org... [criticalcommons.org]

    Pause any morph in the state between two faces and you have the very realistic looking face of a person who does not exist. Couple that with our modern software algorithms that identify various landmark points on a person's face (used by SnapChat and other filters to put sunglasses on a person's face in the right location, etc), and voila... you have infinite face creation with no manual input by blend

    • Re: (Score:1)

      by Anonymous Coward

      I appreciate the “it’s all been done before” sentiment, and I too remember when that Jackson morphing video came out.

      However, the technique used in the new article is really different. If you want to simplify it to simple concepts, the Jackson jethod is interpolation between two endpoints (in the space of all possible real faces), whereas the current method is to progressively generate images and refine them based on whether they are face-like.

    • Re: (Score:2)

      by ffkom ( 3519199 )
      Ah, I remember the good old days of "Morph+" on the Amiga... when this was new & cool. But the effect dropped astonishingly quick out of fashion.

    • Re: (Score:2)

      by clovis ( 4684 )

      You reminded me of one of my favorites youtubes
      https://www.youtube.com/watch?... [youtube.com]

  • They can make all the fake profiles they want. reference [scientolog...roject.com]

  • That book with four separate sets of pages, so you could choose the hair, eyes, nose and mouth to make hilarious combinations.

    Good times.

  • I looked at some of the pics. Either the pictures are real, or they're composites of real people. There's too much background detail. The reflections in eyeglasses show content. Almost every pic had some easily identifiable artifact like a smudge, a paperclip growing out of the top of someone's head, or the temple from a pair of eyeglasses just hanging out without the rest of the glasses.

    For me to buy into this, just get the semi-fake person to start sending tweets with the AI-driven fake tweetbot [slashdot.org].
  • Comment removed based on user account deletion
  • The real magic isnt that it is able to create new faces, but in fact, how it is able to more accurately blend the individual real-world body parts (ie, the eyes, hair, mouth, wrinkles all come from real-world examples) in the scene by understanding the correct lighting and textures. We have been able to randomly generate non-existent faces for decades, but doing so, so the parts blend well has always been the hard part. It's a Face-Time filter applying more passes. I still fail to understand why the buzz

  • It seems to create weird artifacts around the edges of hair and ears.

  • I got a better name for the service. I guess you can't trust anybody on the Internet (anymore).

Slashdot Top Deals

"Plastic gun. Ingenious. More coffee, please." -- The Phantom comics

Close