I'll simply observe that it is easy to tell a fake face when presented an either/or choice and when specifically asked to. Most of the time we aren't looking as closely, so while I see some commenters being very happy about their accomplishments, I don't personally see a reason to rejoice.<p>Regardless, the AP news article[1] linked under the "methods" page provides some useful reading on how to detect these faces, for anyone interested.<p>[1] <a href="https://apnews.com/article/ap-top-news-artificial-intelligence-social-platforms-think-tanks-politics-bc2f19097a4c4fffaa00de6770b8a60d" rel="nofollow">https://apnews.com/article/ap-top-news-artificial-intelligen...</a>
After 5 minutes, I got tired and started seeing some of the same pictures again. 100% right all the time. For me, the trick is to assess the background, ear shape, synthetic textile (if any), and skin conditions.
On a streak so far.<p>I'm playing "can you tell which picture has a non blurry background and has no artifacts?"<p>edit: My first mistake is when I thought a piece of fabric on a human was unusually warped.
It told me that the following is the real face:<p><a href="https://www.whichfaceisreal.com/realimages/02794.jpeg" rel="nofollow">https://www.whichfaceisreal.com/realimages/02794.jpeg</a>
For all the people boasting about how easily they can detect it: yes, you have to deeply look at possible artifacts (especially in teeth/ears) but sometimes it's not that easy and I'm pretty sure it would fool most of the population, especially if not giving a reference, real image on the side. Photoshopped images can also be spotted easily by keen eyes, but they still do their job, which is deceiving the majority.<p>Edit: typos
I find the result impressive.<p>I'm sure this fools a majority of people, contrary to the comments here. Obviously, with detailed analysis, you can probably spot the difference, but in day-to-day activity, and without knowing that one picture is fake, you will fool even more.
Eyes and backgrounds. Easy peasy, 10/10.<p>Backgrounds should be generated by a different model and face should be pasted in, now that would be a real challenge! Models that fix eyes already exist.
The backgrounds are a dead giveaway for me most of the time. Granted, I’m a professional photographer and spend a lot of time looking at photos taken with various lenses, so have become pretty familiar with depth of field and all that jazz.<p>That, or the backgrounds have the weird discombobulated shapes and structures that only vaguely resemble real things, which I’ve also noticed in other AI generation tools.<p>Either way, it still fools me sometimes and it’s pretty remarkable how quickly this has all been happening.
After doing 30 I was able to differentiate very quickly, it's surprising how easy it is to detect these. You can tell by abnormalities in ears and AI probably wont show you hands because it struggles a lot. The backgrounds often look correct but dont make architectural sense. I also noticed if I dont look at the person in the eyes it sometimes is a tell, I'm not sure why though.
It's the background. The faces look half decent but all the AI backgrounds are fucked in some way. After a few misses getting my bearings I started getting nearly 100% success rate, and within a second and a half in most cases.
I found that in direct comparison, the background often was enough to tell the difference - but that was mostly because one of the images had a detailed background with text or architecture, which I know the AI would struggle with.<p>I think a similar test that is not asking for a direct comparison but just "is this image real?" would be much harder, since there is no better "safe" choice to fall back on.
My detection ratio was 100% successful and I didn't pay attention to anything in particular, it just clicked. I don't know what gave it away. I suspect that is because I looked through so many pics on thispersondoesnotexist.com, my brain's own neural network learned how to detect them (which is still a blackbox to the consciousness).
I got the first 3 wrong, then I started looking at the necks and the background and got all of them right (although not always 100% certain I was going to get it).<p>A few of them to have some artifacts on the face that give it away, but this is very impressive.
The pointer shouldn't turn into a magnifying glass when I hover over a picture. That signals that I can zoom in and look at details more closely.<p>Use a normal pointer.
In contrast with everyone else, I struggled a lot with this when just looking at the faces. I made twenty attempts and got ten successes and ten failures. After reading other comments, when attempting again by looking at the backgrounds, I tried ten more times and went nine and one.<p>But I believe I am somewhat face-blind. I have never understood how people were able to describe faces to the cops to make those mockups of criminal suspects. I also struggle to recognize faces sometimes, including celebrities and new dating partners. At a past job, I remember thinking two of my coworkers were the same coworker until I saw them at the same lunch outing and it suddenly clicked. I recently got confused by two characters in an action movie with less than a dozen characters total, and realized shortly after that they had different ethnicities.
It asks to click on the person who is "real", which gets a bit strange when "real" seems to be someone in green contacts and a wig for cosplay: <a href="https://www.whichfaceisreal.com/realimages/12481.jpeg" rel="nofollow">https://www.whichfaceisreal.com/realimages/12481.jpeg</a><p>Biggest issue seems to be a number of images of people consuming their deformed selves:
<a href="https://www.whichfaceisreal.com/fakeimages/image-2019-02-18_165449.jpeg" rel="nofollow">https://www.whichfaceisreal.com/fakeimages/image-2019-02-18_...</a>
<a href="https://www.whichfaceisreal.com/fakeimages/image-2019-02-17_222220.jpeg" rel="nofollow">https://www.whichfaceisreal.com/fakeimages/image-2019-02-17_...</a>
The teeth were a big giveaway for me. The gap between the central incisors should roughly line up with the nose but the fakes are almost always noticably offset.
I'd love to see what images from the training set look most similar to a given generated face.<p>It's hard to decide whether these are impressive without knowing whether each face is just a real face with some minor adjustments.
I got 18 in a row before I missed. There's something around the corner of the eyes that's weird, but I'll be damned if I could figure out how exactly to articulate it.
On a slightly related note, whenever I see a generated face with other faces in the background, and those faces are warped in strange ways, I get a very unpleasant sensation, like a chill going up my spine. Does anyone else get this?<p>Example: <a href="https://imgur.com/a/eK0jMZx" rel="nofollow">https://imgur.com/a/eK0jMZx</a>. I can look at it after getting used to it, but at first glance I have to look away.
100% right for 10 minutes on an iPhone (zooming in as needed).<p>Other giveaways I haven’t seen mentioned in the discussion: vague earrings (fake). Coherent details in glasses reflections (real). If second person in picture has good details, probably real. Second person has bad details, too easy, fake. Gratuitous wisps of disconnected hair, fake. Actual clearly coherent finely detailed design on glasses frames or clothing, real.
This game seems quite easy. When I <i>know</i> one of them is computer generated and one not, it's easy to pick the real one. Like a multiple choice question is much easier than otherwise.<p>I didnt get a single one wrong, and am now playing with the rule that I have to decide within a few seconds, still all right.<p>Still, they're pretty good, If one of the CG images came up by itself in the course of other business I wouldn't bat an eyelid.
I always find these "which is real" comparisons interesting because there is always some type of distrotion around the borders of the face, like the AI has a good idea what a face looks like but things get fuzzy when it tries to create the stray hairs a person always has sticking out.
It's usually the glitchy artifacts which give the clues.<p>Not the face itself, bur what's around it/background/other objects etc.<p>Check for weird-looking "something's not right" objects/backgrounds and you'll get most of it fine.
I agree with another commenter that "Which face is real?" is somewhat easy to determine. In this scenario, it's A or B. You already know one face is fake, and one is real. It would be substantially more challenging if the question was rather "Can you spot all the AI generated faces?" and it turns out 40% of the time there is no AI generated face at all.<p>AI vs. Real can become somewhat easy to identify over multiple repetitions - AI vs. Real, Real vs. Real, AI vs. AI. are all scenarios that should be included to increase the difficulty imo.
Same as the other commenters, I got the first couple wrong, but then quickly realised what I was looking for. You can see artefacts in the skin of many of the faces, and often the ears were the giveaway.
Where is the proof? How can we trust that this website is honest about which are using real and fake faces? This might be some grad student's psychology experiment, or some artist's comment on our understanding of reality. If you can fake a face, you can fake a website. If i wanted to sell a database of "real faces" i moght just generate them myself using AI and sell them to researchers as real, forever polluting such tests. That would certainly clear up any copyright issues.
I’m viewing this on a mobile device so I can’t zoom in too efficiently to do all the subtle detail stuff. What I caught onto was the real photos were subtly messy with imperfections. The AI ones have this idealized look to them that has a slight airbrushed effect in aggregate. If people’s faces had blemishes or imperfect skin it’s more likely real. Somehow those imperfections get chopped in AI, probably because they’re so idiosyncratic they don’t survive the AI transformations which look at features en masse.
Software is good at making faces now, but concentrating on the the periphery (background, ears, earrings) is still easy to spot a fake. Also computers don’t know how hands look, at all.
Perfect score, but it’s such an impressive technology. Traditional graphics with triangles and ray tracing are fake at a glance, but here you need attention to detail and a bit of wit.
I got the first 5 wrong because I thought I was meant to pick the computer, not the real person. After that I did ~20, got them all correct.<p>There's alot of tells. Glasses make the edge of the eyes look strange. Around the ears, hats, sometimes the backgrounds etc, the blurs are wrong or corrupt.<p>However, if I didnt know 1 of the 2 were generated, I wouldn't look for anything and would probably just assume it's real, unless there was real obvious corruption on the face.
The faces look pretty good for the most part, but there a few things that usually give away the generated photos:<p>- artifacts in backgrounds;<p>- weird patterns in clothes;<p>- clothes that are very ill-fitting.
Since this uses StyleGAN, it's relatively easy to tell when an image is fake or real, since the networks seems to have trouble with backgrounds and faces that are directly adjacent to the main face.<p>However, since diffusion models are all the rage now, I think we would perform significantly worse with landscapes or images of fruits and animals, especially if the task is "distinguish between the real and fake art".
This site was put up in 2019 presumably with images from 2015-2019 era algorithms. This was early in the viability of these image generation techniques, so the authors work is kind of precinct.<p>However the state of the art of image generation has moved - I suspect a 2022 version of this would be substantially harder.
At first I was tricked a handful of times, but I trained myself in what to look for. At first uneven blemishes proved a useful heuristic, but then when I looked deeper I found the edges and backgrounds were even more effective. The fakes somehow feel like they are in this... Oily world of illusions.
Any picture with more than one person is real. If there is a second partial face, or a shoulder or hair, or any other sign of another human, then it's the real one. They need to clean up their data, unless they're testing for people figuring that out.
Is this a serious topic worthy of serious responses from high ranking HN readers?<p>Depressing. They're both photos. A photo (of 'reality') is at its very best, at very best, already just a representation of the subject. Both are (technically) fake, aren't they?
Some of these are just straight up insane fever dreams if you evaluate the entire photo instead of just the face. After easily getting 80%+ correct, I had to stop. It wasn't from boredom, but getting creeped out by how grotesque some of these fakes were.
Apparently this face is real: <<a href="https://www.whichfaceisreal.com/realimages/59264.jpeg" rel="nofollow">https://www.whichfaceisreal.com/realimages/59264.jpeg</a>>.<p>If you say so, I guess.
Eerie. It seems like most people here could tell, but with my morning vision and on my phone screen I did terribly. Not sure if it’s because I can’t see as clearly as usual or I have some deficiency in identifying faces.
Easiest thing for me right now is the background.<p>If it has blurry and screwed up bokeh or random patterns, it's probably the fake one. If there's something incredibly detailed, but blurred, it's probably real.
Eyes, teeth, and Ears, that’s my endgame… plenty of anomalies there, took me about one per second to get 20 right in a rows then I stopped playing. I figured I was trining the the system… your welcome!
I played about 10 of them and got them all right. It’s very impressive, but I just happened to know that ears and teeth are particularly problematic for these generative models (for now).
This was surprisingly easy when looking out for artifacts. StyleGAN2 significantly reduces artifacts, i'd be very interested to see StyleGAN2 on this website as well!
The biggest tell for me after ears, hair and backgrounds was hard-lighting. None of these generation models 'learn' shadows with point-like lighting.
I wonder if it is possible that a generated face matches a real face. A person with real face is detected by algorithm as unreal and cannot access facilities.
Dead simple if you just focus on the ears.<p>Every single fake image has poorly rendered ears, which makes perfect sense as the contouring would be hard to get right for an AI
Oddly enough I was rocking 100% accuracy by ignoring the face and looking at the background.<p>Maybe a face-centric ML isn't so good at backgrounds. Shocker!
just look at outlines, especially EARS area or earings at women, AI can't make real ears yet, face is already OK (though AI can't really make skin imperfections), but ears are giveaway<p>tried 3, got all correct, no point to waste my time anymore, same issues as always with these photos
We rarely see real faces any more in photographs or videos. On YouTube Biden looks like 60 and without wrinkles, everyone uses tons of filters.<p>Perhaps that is the reason people are doing badly in these tests.