StyleGANv2 Explained!

This video explores changes to the StyleGAN architecture to remove certain artifacts, increase training speed, and achieve a much smoother latent space interpolation! This paper also presents an interesting Deepfake detection algorithm enabled by their improvements to latent space interpolation.
Robert Luxemburg's StyleGAN2 Interpolation Loop: • StyleGAN2 Interpolatio...
Paper Link: arxiv.org/pdf/1912.04958.pdf
Tero Karras StyleGAN Video: • StyleGAN2
NVIDIA StyleGAN Repository: github.com/NVlabs/stylegan2
Thanks for watching! Please subscribe!

[@stef902]

imagine if it generated a person that doesn't exist and it looked exactly like you.

[@ashfaqahmed1408]

that happened to someone on reddit

[@delamin8r]

The obvious conclusion would be that you do not exist.

... and you start to devenish

[@burenyuu8815]

theres creepypastas with this. Apparently, a woman found herself and a few days laters, she was found dead. she doesnt exist anymore

[@rebstrike4507]

@@ashfaqahmed1408 Did it really? Where's the post?

[@jessicaw725]

"Don't panic" I'M PANICKING

[@kopser11]

Donttttt

[@porchcatproductions9931]

It makes me feel anxious and I don’t know why

[@rhodwolff144]

You look like a generated person

[@sapcebou5648]

omg ikr THIS IS SO CREEPYYY

[@DuCKie.03.514]

Because im gonna lose my job

[@dvmpgmhl1191]

These people may not exist but I like to think that they do in a parallel universe and we're the ones being shown on their version of StyleGAN.

[@Donevello123]

nakakatakot hahah

[@cupofkoa]

They kind of do exist. They are interpolated from existing faces. It's likely that a lot of these faces look very similar to the training data. You could probably produce an image from the generator then find the top 1-10 images that the generator interpolated from. So if you happen to know someone that's in the training set you'd say "hey, that looks like Darren", probably because it's >50% is actually him.

[@callmedaddy6969]

🤯

[@zoeydeu2261]

Mind blown

[@paulacrespo5299]

lmao imagine

[@ziconghuang7139]

This gave me existential crisis. Looking at these faces, I can imagine their ordinary life. I can see how they live through a normal boring day. They are one of us, we share the same human experience of love and sorrows. But they are not real. They've never been born.

[@pamelahills8985]

What freaks me out even more is that the background looks like a real photo , there are often people in corners of the pictures, like a cropped picture of some friends. It’s sooooooooo creepy

[@gavins9846]

The sperms that never made it

[@cheesobogo3199]

How would we know maybe those people do exist somewhere out there..thats scary

[@terrandroid]

So extremely weird, I have the exact same thing.

[@ingenide]

I saw one with two people but the second person’s face was freakishly distorted

[@londons_legacy]

Imagine someone on catfish using these photos

[@shawnio]

this shits been around for more than 20 years

[@ch4vel]

Thx for the idea heheh

[@tatict]

😂😂

[@patooeyspit]

imagine?

[@zion30164]

i’m deadass watching catfish rn💀

[@thejbo777]

I have absolutely no idea what he is talking about.

[@BR-hi6yt]

He's a Klingon - from planet Latent Vector

[@jefgoossens2285]

I was looking for this... Absolutely no clue

[@milanstevic8424]

ah, so we're assuming this guy was born with this knowledge? why don't you start with something simpler if this is too much for you. or if you don't you can always stfu. when was the last time you went into the pilot's cockpit and went like "I have absolutely no idea what these buttons are for"? and you don't; so why is this place any special? if you really want to learn this stuff then do it. if you don't, don't try to get validation to remain stupid. because this is exactly what you're doing. people don't normally go "oh I'm smart, I know this" and similarly "I have absolutely no idea what he is talking about" without an explicit reason. your reason is to get socially validated, because you're > absolutely full of ideas < how to make a stampede of certain like-minded people *like* your ignorant comment, by throwing an appeal to intellectual insecurity. if this is not for you, that's ok, this is exactly why our prankster god made Nicki Minaj and Kim Kardashian.

[@thejbo777]

@@milanstevic8424 Dude why did you write a whole essay lol

[@milanstevic8424]

@@thejbo777 because there are as many letters in my essay, as there are comments like yours. maybe if you all team up and each one of you reads just one letter, maybe you can share with each other this magnificent information and gather a zoomed out picture of what's going on with the actual message. I'm sorry for being unable to split the letters into something even more digestible. in a regular world, where people regularly use more than one neuron, this is also called a meaningful comment, when there is more than one basic sentence involved. this is usually necessary to convey more than just an expression of a primal instinct, a social validation utterance, or a general purpose internet troll.

[@johngrigsby5351]

This description is exactly what I hear the professor saying, in a recurring nightmare of mine, as he hands out final exams for a class I never attended.

[@anthonymerola1489]

I’ve had a similar reoccurring dream, where I had three classes with time conflicts, which were needed to graduate. I had to take finals for classes I couldn’t attend for that reason.

[@yudhst]

So it's like how you make a characters on The Sims, but with greater graphic quality

[@dagnytheartist]

Conspiracy theory; real life is a very advanced version of The Sims. The reason real time moves slower than other Sims games is because the game has to process all the high-quality graphics. But time gets faster because the game is getting better at loading our graphics faster over time.

[@nehiryalcndag4031]

@@dagnytheartist unique

[@aniyaturner6]

omg i love the sims

[@DuCKie.03.514]

Infact, we are all just randomized sims lol

[@cleetusboon]

Is there an English version of this video?

[@rodrigoidalino9228]

u kidding? Well, there is always the playback speed which you can change. You`re welcome.

[@cleetusboon]

@@rodrigoidalino9228 I was just looking for a simple explanation for how this tech works. I didn't understand this at all.

[@gymdump6]

Ryan Ashcroft same dude

[@face0nmars]

😄

[@bizzaremonkey756]

@@cleetusboon just be smarter

[@aqualisa]

I like looking at the details in the pictures, weird stuff. The earrings, the reflection in their sunglasses, and the sometimes rather creepy backgrounds. Fascinating.

[@dacoolestdude7376]

sometimes there will be a hand of the og hoto and the ai will absolutely frick it up and mold it to the shoulder and stuff and its nasty

[@valedictorianism]

Glad to realize I wasn't the only one who noticed that.

[@X152535]

It sounds like English but I don't understand a single word. Is the speech also AI generated?

[@jacobladder5556]

This language does not exist.

[@coronaguy1829]

I can understand it very well

[@Keziah2447178]

@@coronaguy1829 cool

[@animallover5626]

@@coronaguy1829 Same, he does talk fast though

[@animallover5626]

If you set playback speed to .75 it sounds closer to a normal speed

[@Aomicplane]

I don't understand anything in this video

[@elsiee3502]

Arnodude53 bytheway, i love ur profile picture 🥺

[@FreedInPieces]

Unless you have some knowledge about this topic already, you'll be lost. I was! They don't really explain what the variables are. The diagram was mindbending.

[@5starkarma]

@@FreedInPieces I believe the variables are rather constant in terms of naming conventions and they are vectors and matrices according to linear algebra (at least I believ)

[@MichelRivera]

LOL

[@ava-dv4zp]

i love your gorillaz pfp!!

[@eitanshirman9072]

I can't express what I feel looking at these photos: I just think how cool it'd be to meet this person

[@hector1404]

It is amazing technology, fascinates how the first pics of the trained model are just patches of colors and then slowly starts taking form, the progression is outstanding

[@YaYa80808]

Sometimes there would be other faces in the picture of the generated person and the other face would be completely terrifying

[@hector1404]

Yeah, some details on some pictures will show horrifying details, stuff of nightmares, could work for some creepy pasta : P

[@iSlandeRon]

Yeah and the back ground are all creepy af

[@Brainmelt427]

Yeah it's like they took a selfie with their sleep paralysis demon and tried to crop it out of frame.

[@valedictorianism]

Exactly! Goya said it best: "El sueño de la razón produce monstruos" (The sleep of reason breeds monsters).

[@lunchbag_larry]

people are getting too smart i think we should start over back in the renaissance sometime

[@ok-kk3ic]

More like before religion started to take shape.

[@datatwo7405]

If only our sense of compassion and ethics would progress as swiftly as our ability to create all our toys.

[@philidor9657]

I would be a genius 100 years ago instead of barely average. Lets go back

[@BigboiiTone]

Idiot

[@e.erin.]

I think you’re right. We’re about to max out.

[@youtubesuckss7058]

This transition part made me realise how magnificent a creation human beings are, amazing! We're amazing dude

[@mezzops1716]

its like hitting randomize character in a video game.

[@QuranicRemembrance]

I was lowkey scared when I first saw the first image.

[@condew6103]

Actually, they are well past the "uncanny valley" of a synthetic image that is unsettling because it isn't quite human. But I would be scared of a shell corporation with a president and dozens of "board members" none of which are real, but on a cursory internet search would come up looking totally legit.

[@Jay-Jones]

Me too

[@alvinxyz7419]

i fell in love with one of girl generated by AI

[@lyudachka11]

Loner gang

[@dumb.minnesotan.moron.2005]

lonely simp moment

[@solidkingcobra]

You're just rippin' of the plot of Ex Machina.

[@zoidedits5859]

​@Willian Shut up, Mr. Woke 14 year old.

[@savanisis7802]

like the movie- HER with scarlet Johannsson and Joaquin Phoenix!

[@sandwichchristensen3007]

Whenever I see one of these I think about how whether there's a possibility that someone in the world looks like that randomly generated picture or not

[@craigbikes8831]

This is literal sorcery

[@Ryaaaannnnnnn]

These videos are gold, please keep them coming!

[@connorshorten6311]

Thank you so much!!

[@animeshkarnewar3]

Very nice explanatory video! I loved the "replacing progressive growing with multi-scaled gradients" part :) :) :D ;D

[@connorshorten6311]

Thank you so much!!

[@ezerad1]

They said don't panic click on this to find out... i feel like I'm panicking more lol

[@ivanyaadams9442]

I wonder if it is possible to develop a detecting system.Dude seriously.... these seem freekingly authentic

[@DialecticRed]

I think there tends to be a general pattern with a LOT of phenomena where there is a huge offensive advantage, with defenses being unable to keep up. I think that's become the case for hacking (with the exception of SHA-256 encryption and similar), and it's also the case for wars (nukes have become such a powerful offensive maneuver they've effectively become more defensive than anything else), lock picking, infiltration (easier to be a spy than to ensure that your organization or whatever has no spies), and you could probably pull even more examples. Idk if there's like a word for this pattern or something, it's just something I have noticed. That being said, I've tried generating a lot of images, and a lot of times something might just feel very off (I think it's most apparent in it the children it generates, which I can only assume probably has to child photography restrictions and data collection laws for individuals under 18). Also, the differences become the most apparent in the backgrounds of the photos, which often give it away. There's also some photos that just get a little messed up, I see a lot of them where eyes might be oriented in an awkward and unnatural way, or anytime someone is wearing a headband or something that interacts with their hair without covering it, it tends to get messed up. The same is what happens whenever hands are in the frame, but even worse. They just become . . . blobs. At the end of the day, if it ever becomes a real issue then there could be authentication systems in place such as requiring users to upload photos of themselves with their finger on their nose, their tongue out, or something else to be able to authenticate that a user is using their actual face as a pfp.

[@TheRevitQueen]

there is a lot of good work happening on that front. kzbin.info/www/bejne/iKCqeYl_YtxsmKc

[@loganh2735]

I used to check the hairline, but it's improved so much that it's not a giveaway anymore. There is a mostly reliable way to detect it IF it's not cropped: The eyes and mouth are generally in the same spot. This is by design, but unfortunately it varies so much that it can generate faces at 45° which fail that test. So really, we're the detecting system. The AI can get 90% of visual cues right, but it will always miss something.

[@jeffmiller3499]

Balenciaga spring 2020 collection

[@Dead-Not-Sleeping]

Great explanation, thanks. On a somewhat related topic, I always speed up informational videos as it would seem that they are usually recorded with the lowest common denominator in mind, and then halved. The point is, I love the fact that you weren't 'dicking around.' :)

[@drabberfrog]

Who's here from daily dose of Internet?

[@laelbm3360]

Here!

[@journeygalore]

Me!

[@quinn470]

me lol

[@merisko4162]

im here from unus annus even though its sadly gone.

[@gypster2004]

My brain translating while listening to this: Она нервно всматривалась через край. Она понимала в своем уме, что вид должен быть красивым, но все, что она чувствовала, это страх. В росте всегда было что-то, что беспокоило ее, и теперь она могла чувствовать всю силу этого беспокойства. Она неохотно подкралась немного ближе с поддержкой своих друзей, поскольку страх продолжал нарастить. Она не могла не чувствовать, что вот-вот произойдет что-то ужасное.

[@mykebryan8299]

I hit refresh for hours. This algorithm is not capable of producing a dark skinned person. So, we’re safe LOL

[@Bouncymotive]

myke bryan yessir facts 😂😹

[@KartonRealista2]

Right now it does! Now no one is safe...

[@ira6605]

LMAO really? I got two dark skinned people within first 10 hits. I think its a chance event you did not get any. :D

[@mmmmmmmmmmmmmmmmmmmmmmmmmm3153]

LMAAAOOO

[@ThePC007]

Lots of dark-skinned people when I tried it, lol.

[@preetcreation5667]

The way of making images matters whether through any application or other way. I think it is amazing and we have to be familiar with this image theory, normalization artefacts and everything which is explained in this video lesson

[@Sebastian-fc5dc]

Wow! This was really helpful! Thank you so much.

[@connorshorten6311]

Thank you!

[@amortalbeing]

This was great. keep up the good work

[@connorshorten6311]

Thank you so much!!

[@markcounseling]

Thanks for clarifying that. I was concerned about the tractability of the inception score's pathway vector given that the semantic tuning of the instance normalization caused the FID score of the Stylegan to obsessively grow and produce droplet-like effects that would not scale properly into the latent space. But now that you've clarified that artifacts at the 64x64 scale are not automatically restructured by the latent vector W into the channel-wise instance statistic, I can see that any 3x3 convolution will be parameterized according to Gaussian noise demodulation layers, and the s-sub-I unit invariance will indeed convolute all semantic substance properties. Again, many thanks.

[@LordOfSilense]

VX Junkies are wilding

[@sbalogh53]

I am trying to decide if you are trolling or serious. I can't tell. :)

[@markcounseling]

@@sbalogh53 The deft magician never reveals his secrets.

[@Zoe-qn8uo]

By looking at the picture you can visualise that they have a regular life... but the thing is... they don’t. They never existed to start with :0

[@tpstrat14]

If you’re freaked out, the only way out is through.... keep refreshing and notice the similarities in the lighting, the expressions, and the weird glitches in the background. This freaked me out so much but I kept going and now I can tell they’re not real. Existential crisis avoided. But when they figure out how to make videos of people that don’t exist, I’m done

[@riggs20]

Deepfake videos are real and even more terrifying.

[@murrayshekelberg9754]

@@riggs20 They are getting good with deepfake voices now, as well. Its come to a point that no video of a person can be taken as real. We are in the dystopian future literature warned us about but nobody has noticed yet.

[@lightchasr]

Hollywood already does this. No?

[@TheHardikupadhyay87]

i understood everyting in 17mins that was very clear thanks

[@stentor9640]

This whole thing reminds of the movie "It Follows"

[@thedude8046]

It blew my mind when i first heard about this. It feels weird when i look at the pictures.

[@innamordo]

Would love to see an interactive version of TPNDE for example a slider for the Z vector, to morph the face!

[@kranthi9010267088]

Very nice explanation..Thank you

[@alyuadifebryansyah9884]

whew so we can predict how our child look by combining the pics of the parents

[@zerobyte802]

My question - I'm assuming the algorithm starts with real photos of real people and then manipulates them with these various parameters and vectors and so forth. Is that correct? This almost looks like "morph" where instead of having a target picture, you're just telling the network: Make it different! GO! Then it hits stop just before the image turns into gooey playdough people. Is that pretty much a lay-explanation of what's going on? I've seen some of the images where a beanie hat has somehow merged with the person's hair and then their head. It's also funny if there's a piece of person visible in the background that isn't the subject of the photo. They're generally quite distorted and deformed-looking because I assume the algorithm isn't actually paying attention to them.

[@babyshoemcclanahan]

thanks for the explanation i got it now.

[@nooriiem91]

Hi there, I would like to include one of your images in my uni project as an example, firstly am I allowed too? and secondly if permitted, how should I reference this?

[@pfarabee]

That last point.. yes, exactly. That technique for detecting fake images would only be useful if you had access to the exact trained network they used, which is completely reliant on the training set used. If they used a different training set, or even an existing set with significant additions/omissions, you would probably not get such a perfect match.

[@jonathynblythe8096]

Is there any way StyleGAN can be downloaded as a software application for one's own desktop? I've seen videos of github and all that along with the python language and downloading other applications. But for the layman, none of this makes any sense. We're used to downloading software with .exe in it. The tutorials I've looked at for github stuff lose me at about half way through. Thanks.

[@garyriehle239]

Do these start from images of real people and then transform them, or are they completely generative?

[@MikeHowles]

He had me until 10:03 where he said "it was a little bit confusing for me".

[@cjpirtle]

Tik Tok brought me here.

[@connorshorten6311]

Lol, awesome

[@madisonfox2470]

Cooper Pirtle same

[@connormclees579]

Same

[@connormclees579]

Cooper Pirtle same

[@skater_boi_4876]

Same

[@simonstrandgaard5503]

Great explanations.

[@connorshorten6311]

Thank you!!

[@ThomSonnyYeah]

The art work generator is blowing my mind

[@misterSex358]

Great video! Waiting for new AI weekly update videos!

[@connorshorten6311]

Thank you so much for your interest in the AI Weekly Update videos!! I am taking a short break from it, but will get back to it soon!

[@Salad-Cream-Binge]

I understood all of this completely

[@inkognitongue765]

Does anyone know a video or book that explains this better (ideally in German)? Because I don't understand anything but I'm interested in the topic.

[@samuraicupcake289]

The interpolation animation is mind-blowing but also a little scary.

[@TheThirdLieberkind]

from my testing it seems stylegan v2 is much slower than v1: stylegan v1 I compute initial 8x8 pixel kimg every ~1.2 seconds, stylegan v2 config A a kimg every ~17 seconds, stylegan v2 config D a kimg every ~52 seconds. all with vanilla setup - I wonder what that is about

[@10Teabags]

Fell in love with this person who does not exist.

[@RealSkyDiver2]

Is there and English version of this video?

[@JesseStoddard]

What we should all be asking every single one of the people involved in any and all of this technology: "Okay, that's all fine and good. I get the 'who, what, where, how'... but far more important is WHY? AND What does this mean to our future? HOW can anyone possibly not see this as a HUGE potential problem to humanity on infinite levels?"

[@chupameoscolhoes]

Chills "14 Videos That Are Freaking People Out" brought me here x')

[@marinin_]

Wait, chills got a video about this? Gotta watch it omg

[@chupameoscolhoes]

@@marinin_ yes , and it gave me chills just by the way he spoke about that site. I actualy got myself some creepy generated images from that website x')

[@lightttt5146]

anyway to implement image2stylegan? id like to use a custom image to modify or what kind of parametres should i use in github.com/NVlabs/stylegan2 to combine 2 images together?

[@larrysmith9204]

Thank you for video

[@aniketshinde4416]

Any framework implementation for the ssame tutorial will be highly appreciated,

[@DanielJesus0]

are you also working detection systems for its use?

[@seoexpertsandyrowley6598]

Mind is blown.... 🤯

[@austinpage9463]

Surely it's possible to have more than 8 gpu's right? Would the training time continue to get lower as you add more gpu's?

[@xxoo3712]

Once it's generated you made it exist, so every generated picture could be and definitely will be someone in the past present or future

[@chreeshawn]

thanks ruby!

[@judylwbidoov4195]

Is it weird that my friend and I were refreshing it and we found the same image at the same time? We both got the exact same picture of a girl with a blue and red headband. Is it supposed to have duplicates?

[@VoltisArt]

Many "random" computer-generated things are based on an algorithm that uses a current timestamp as the input key or seed. If you both managed to refresh at the same exact millisecond, it's possible that's what happened. It's like that one-in-a-thousand (1024 actually) chance of tossing the same side of a coin ten times in a row.

[@sscamt]

Can we use this innivation in stock market by technical analsysis module.

[@toaka5568]

This reminds me of Babel library if you go to photo search you get this random patterns of colored pixels this is somehow similar but in a rather organised rational and concivable version of these photo to form a face which means it's a random process but somehow controllable and adjustable .

[@zarsiw]

the library of babel? like borges?

[@sophiapicchi2677]

ok this calmed me down

[@alisonlivingston]

I went to the site to generate my own "cats that don't exist" and had nightmares for days. It's like road kill without the gore.

[@Lydix]

The cats have gotten MUCH better. The horses though... ouchies.

[@machko1]

Imagine falling in love with one of these faces...

[@i.ehrenfest349]

Not having time for 17 minutes of high speed explanation: are these composite photos, or are they generated some other way?

[@VoltisArt]

I do have an honest question after about 100 refreshes of the site. All the faces seem to be young children or adults. Where are those between 8 and 21?

[@abstract_duck]

could this be done In Blender's node editor?

[@Clikaco]

Bro, don’t you think it would be a good idea to establish some policy around this potentially reality shattering technology before we “dive into this latent vector code”? We have enough issues as it is right now.

[@-GodIsMyJudge-]

I agree, the potential negative applications of technology like this and deep fakes are lost on many people.

[@rusticagenerica]

Any recommendation for a good video about GANs?

[@TasmanianTigerGrrr]

I had a feeling it worked by normalizing W-space semantics in a cross-referenced coefficient J-slide.

[@sdoron79]

Can i use these photos on my academic research? people are not willing to do so..

[@stefanSS1480]

I will use this for profile pictures someday if I need it

[@GB3770]

Can this AI or any AI generate a face and show it from multiple angles? These are all front shots - how about a side shot or just at a slight angle/different background? Can it do this? If so how? Asking for an AI paper I'm writing...

[@Bestgameplayer10]

You ever figure this out?

[@jvoidhuey3088]

My question is this, could you feed example images into this as a library that the fakes will solely use to create from? Could this also create full 3d models to import into virtual worlds? I wouldn't be surprised if we have the tech already to do something like this. VR is quickly gaining steam. There will be big interest in social media VR worlds. VR Chat and I hate to say it, Rec Room, is already showing this; with VR Chat being the closest thing to READY PLAYER ONE's 'Oasis" concept. I think that people will also create their own micro social media worlds as VR progresses with AI. People will be able to create their own virtual worlds with their own virtual characters that they will set custom AI personalities onto. Friends will then share their creations, or connect to a Universe of micro worlds, or just simply connect theirs to others much like a friends list so it stays selective. Being able to create life life 3d characters will be a must for something like this. I would love to recreate famous thinkers like Tesla, Einstein, and others, who can both teach, and keep me up to date on scientific breakthroughs. I would really like to see this tech shown branch out. It has a lot of potential.

[@DanielBrownsan]

Is there a ELI5 version of this? It *almost* sounds like he’s making things up.

[@spiedz2130]

interesting vidoe!

[@put_istine]

It reminds me when we were kids there were some fun app where you can put 2 or more images of faces and some morph algorithm will create a new faces based on previous images, sometimes not so good as here. And it was looong time ago lol

[@clray123]

The Animorphs, you can find them on LGR channel.

[@AashishBalkhande]

All i can say is BIG BRAIN. I am still wrapping my head on first 15 seconds XD

[@selimasworld451]

5:20 i- im scared

[@Verklunkenzwiebel]

whatif you let this loose on photo's of starsystems, with colorized Huble photo's. Currently, creating a colorized photo from a stack taken om M31 or similar is a pain. Also I'd imagine that the Gan could clean it up, en de-noise it like mad!

[@nadasonic6]

what's the worst that could happen?

[@ryanc4s926]

me who doesn't understand shit he said: HMMMM veryy cool amazing!!

[@LeonePC]

They should add this to the The Sims

[@fubarexress6359]

if you don't have a doctorate degree in computer science/ advanced coding training you will have ZERO CLUE about what this man is saying

[@sliceofbutter5454]

Why did i find my deceased friend on there..?

[@shironookami414]

OMFG it's creepypasta time.

[@lucidraisin]

Oh you silly kids 🤦🤦🤦

[@riri2286]

@@lucidraisin ify😂

[@jl3606]

Doppelgänger

[@the.powerinside]

Itsyaboy Su ahhh! Ein Deutscher! Geil

[@masterwerneck]

R the generated images totally CGI kinda or use real faces parts like eyes, ears and so on of real people?

[@clray123]

They first use real images to train the network, then tweak some parameters to generate back new images from those.

[@JaraFPV]

Did they finally fixed the upper front teeth issues?

[@jeanicadevany4546]

Lol i also noticed that teeth looked fake

[@crispyjake4221]

I like to look at hundreds of these pictures a day to see if I can find one that looks like someone I know

[@timemasterhms]

Who owns the images that get created?