Blind Test: AI vs Real Photo, Can YOU Tell?

[@cvonp]

As a photographer and stock photo contributor, it's easy to be concerned about AI scraping stock image collections as part of their learning models. On the other hand, human art and photography students similarly gain inspiration from works created by others, often mimicking styles-either consciously or unconsciously-as they work to develop their own artistic identities. Regarding Midjourney, it's a challenge to determine precisely what MJ is 'borrowing' from the images it has learned from. In Unmesh's example, he explains that he used a seed photo at 00:57; however, later on at 10:52, we observe the remnants of a watermark from a stock library. Yet, the image generated by Midjourney is evidently derived [almost exclusively] from the seed image Unmesh provided. This raises the question: what exactly did Midjourney borrow that it left the artifacts of a watermark?

[@dakara4877]

As an artist, it can often be an honor to have others mimic your work. It shows great appreciation and you know real work went into learning the style or technique. However, most don't copy verbatim, they take inspiration with their own nuance. Also, because it requires real skill and effort, nobody can simply mass produce millions of your work. All of this is in stark contrast to how AI operates.

[@enchantedwebmaster9163]

You clearly don't even know how AI "operates" so how would you know?@@dakara4877

[@elliotfriedmanmusic]

8/10 correct, but I've generated 10,000+ images with MidJourney so after a while you begin to notice patterns with the model. That said, v6 is very close to the point where it's impossible to distinguish between AI and reality, particularly if you use MJ as a base and enhance it further in Photoshop or other AI upscalers. By v8/v9 I expect the detail to match conventional photography, even fully zoomed in. Crazy times!

[@iamakkkshay]

Same here 8/10 just lost partially in the first Product photo and landscape one😅

[@MsNellyuan]

I won't say its impossible , it may look impossible in this video because some of these picture aren't just strictly AI generated , Unmesh also added in a little of his touch-up . So by definition those aren't just entirely done by AI , it has been polished by a Professional . The first thing i always look for is the Iris . The AI always got it wrong .

[@stevenlitvintchouk3131]

I'm waiting for the big reveal at the end of this video where we find out that Unmesh was generated by AI.

[@Pandybelly]

I hate everything about how it is passed off as real photography. Extending a studio background is one thing, but creating images from AI and passing them off as photography is appalling. It's worse now that nobody questions anything they see.

[@emmygreen8108]

This video made me mad because this game was rigged. I'm out here looking for ai generated flaws, and this man is touching them up.

[@alfie_petersen]

I only got half correct. Its amazing how good AI images are becoming. All these AI models must use source material created by others to train, whether its text or images. Theres bound to be a few copyright claims looming

[@AlkuntarGaming]

That's insane! AI is getting too smart... 👀

[@jmh3d]

These tests were a lot more fun when I was trying to tell if something was 3D (ie. manmade) or not. I was good at those. This changes the meaning of CGI forever. That term is now literal and about to cause all sorts of problems, unfortunately.

[@kasloth994]

I've got 6. The only one I think was unfair was the test with the burgers, mainly because stock and commercially used photos of them were never realistic (aka not looking like actual edible food) to begin with. They always looked fake

[@spidermonkeynr1]

If I'm honest . Ai lack Imagination no fantasy. It's easy to pick out a briljant photographer by the trick they use so create a picutre/make a picture

[@Tochkannn]

Its not a fun game when you just say ‘both ai’ every time😕

[@rzl1234]

Intellectual property is a heavy ANVIL on technological innovations... if it's not subtracting, it's not stealing!

[@akashrna1]

2 of my answer is correct, Mockup bottles, and chocolate. AI is a great asset for photography and graphic designing (ooooohhhh big help for graphic designing). at least we need the main subject to place on the ai images, right?

[@adjustedfilm]

Why did the camera go to therapy? It had too many issues with its "focus," and it couldn't picture a positive frame of mind!

[@jimtincher7357]

I think photography needs to be protected... much like Hollywood just did last fall.

[@AmedeeBoulette]

Wow! I pay for KZbin Premium, yet I get a 11:22 minutes ad in the shape of a game! What a time to be alive….

[@vince7528]

It's gonna create a huge intellectual property issue

[@FusionDeveloper]

This should be called "I edited AI generated images to try to make them look real, can you still tell it was AI Generated versus real photos with no editing?" He didn't take an AI generated output and compare it to a real photo. He took an AI Generated photo and did tons of manual editing in photoshop and inpainting. That isn't a fair comparison, because I am looking for AI generated things and he is undoing them. You could do the same and take real photos and add in obvious AI Generated features to make it look AI Generated. I don't think this is a fair video and by the reveal of the 2nd photo, I feel deceived by the video title and verbal challenge. I got the first image guessed correctly, but knew something wasn't fair/right. I was 50/50 on the second image and as soon as I started seeing what was done, I was angry about the whole video. At this point, I stopped watching the video, even though now I know what the actual rules of this game are.

[@RyeClarke]

I haven't finished watching yet but just with the first one... Ohhh, scary.

[@erwinzer0]

This thing needs to be regulated

[@publicminx]

the patterns in your video are: the worse a photo looks the more likely its a real one - while the better quality/more detailed one is likely AI created. unless it looks very AI like then its likely real. plus the pattern you use yourself by trying to trick one and variate ;)

[@mgreeneyes666]

Ok… Let’s pause for a moment. Isn’t AI just making compilations of images that have already been made, taken, or drawn and composing it into one image. It’s not like AI is drawing this out of thin air.🤷🏽 Some images created are in copyright violations.🤷🏽

[@marlaura5079]

My exact thoughts. And its worse when it comes to people. They can generate someone that is very similar to someone in real life. Imagine being a lawyer for example and having an image almost same as you being used for an instagram lingerie model. Or a male stripper. At least the lawyers gonna earn a lot with this. I can almost see a lot of court cases in future, maybe a law restricting ai. Like they did specifics laws for the internet.

[@zerorusher]

It's well established that using other artists work in a transformative way is no copyright infringement. Thats the very nature of the way people learn art, first you study other people work, than you (maybe) come up with something unique. Yes, AI doesn't create nothing out other thin air, but unless we are to believe that art is magic (which its not) neither do we. We create based on experiences and lots and lots of references of things other people created. If someone makes some very unique art and AI "copies" it, sure, sue the company because it will be obvious that the AI model is infringing copyright. But if someone's work or style is common enough to be statistically significant, on what ground could someone claim having copyrights over some work with is not even really original in the first place? Sure, IA brings plenty of potential for misuse, as any advanced and powerful technology...

[@dantheman2907]

​@@zerorusherAI doesn't learn in any way that's similar to the way a human learns. The fact that it can spit out near identical copies of movie scenes, or tries to add watermarks and autographs proves that it is not only trained on, but also retains some copyright infringing material. It's not even using it transformatively in the case of movie scenes, so it certainly would not fall under fair use as many AI proponents will tell you, as the companies creating these models are doing so for a profit.

[@skeptyka]

Nope, it's not a collage. It's no different than you seeing photos and being inspired by them to make something similar.

[@zerorusher]

@@dantheman2907 a trained human will also spill near identical movie scenes it has seen if asked to. In all honesty, do you know how diffusion models are trained and work? There's no way of discussing generative AI without understanding the underlying technology, these models create imagens by progressively improving on random noise, directed by a neural network trained on millions of images to learn the Underlying semantic relationship between the prompt and the image. Yes, the model can create images very similar to copyrighted material and if your work is original you'll be able to prove your work was used without authorization. Otherwise, by definition the image created isn't similar enough to be identified with the copyrighted work, so the use is, by definition, transformative. This technology won't go away as photography didn't go away because of a bunch of painters got pissed when a câmera became capable enough to achieve realism. Even if some specific model is shutdown (which won't happen) because of issues with the training data, others will emerge based on fully open and free datasets. The underlying technology powering these models is the real disruptive technology.

[@_Frank_23]

Plot twist: Unmesh is also created by Ai

[@visibleinvisibility8557]

I don't think it'll do much to photography. People are always going to want actual photos of them and not an AI generation that vaguely looks like them. For retouching, though, as AI gets better at that, retouchers may find it harder and harder to find work if AI can do the same thing in 5 minutes for half the price. Also may not go well for models. AI is easier to work with than people and AI doesn't flake at the last minute.

[@AG_before]

8. I've been using MJ for a while and was going to stop until v6 ALPHA recently came out. This is, hands-down, the most photorealistic version ever, and I'm not even into photorealism. When you add Magnific and slight editing in Photoshop, it's just not possible. As for stock photo sites, they will have to do what they are, which is just to be transparent about the images provided so that clients can make their own choices. Transparency is key. As for the watermarks, even for many MJ users, it's a sore point, and many are unhappy with it. It's blatant proof that data is being scraped from stock photography. But how much and in what manner? They've been sued and questioned enough times, yet gotten away with it every time. Most AI-image generators have. Lawmakers MUST make laws applicable very clearly so that everyone is properly compensated and people using them can feel that laws are not being broken. There's no way to stop technology. It always wins, but the least we can do is ensure no one is abused while allowing others creativity with a new medium. But I believe the damage done already, is unfortunate enough. Let's be positive about the future. Great video. Thank you. 👍

[@MsNellyuan]

stable difussion can do all of the above plus offering more settings ( all for free too ) . The downside being , it requires a lot of times , efforts and understanding of their models to set up.

[@kathyb249]

Holy crap, that was rough!! I am bad at guessing!!

[@barrycainphotography]

Holy cow I just noticed you started using a Mac.

[@ojaswiverma5157]

Hats Off to your new video editor. The wiggle to the camera is adding an oomph to the overall look. Adding 3D elements is a new approach. Loved the overall funky look. And what happened to the outro. Who turned the TV off. What happened to "like the video and subscribe, and not just subscribe, ring the bell icon so that you my friend don't miss any other tips, tricks and tutorial. I would like to take this time and thank all these wonderful people from patreon....."

[@DaleSheltonsPage]

I know it’s been missing for some time now, but I also miss the “I hope you’re having a good day and making it a great one.” I’ve often wondered if that’s a common greeting in Hindi.

[@Serifinity]

Great video, I would say that any photo that has had post processing is no longer real. Surely unless the photo is an exact capture of what the photographer is seeing, then any post processing makes it a synthetic image. AI is just a continuation of filters and Photoshop post processing.

[@hotlineoperator]

The real picture as a reference is cheating in the test. AI's big problem is the proportions, without a reference it has challenges.

[@sociallyineptsnapper]

Trick question based ones are inherently unfair because even if you feel there is something wrong you are forced to choose one, which is why they’re trick questions but…

[@amy_soucy]

Ok, so I got them all correct. BUT in the same way you get lottery numbers all correct. 😂

[@AnkitLuthra-s4o]

Creativity is inspiration. Steal like an Artist!

[@MrChuckbackus]

Thank you for this. Score 6 out of 10 correct. The cats were my most difficult to determine.

[@JOMprodBeta]

the cat one was hard but when i saw the lights in the eyes i knew which was real because when im doing a photoshoot or lighting a video i understand how the lighting works when reflecting in the eyes. OTHERWISE i would have chosen the cat that was laying on the side of its face. WOW so hard

[@greaper42]

Had to go look it up... Heard it but never knew the name. You'd have to REALLY know that song to catch it.

[@noamcohen9443]

It won't take too long until aI will be able to make things look completely real. That said, I dont believe ai will replace artist or photographers at any point but rather help them achieve better results with less effort. As long as we 'keep creating' as umesh said- ai will accompany us, not replace us.

[@sashr7842]

Thank you!

[@ezeqpoeck]

1, only 1 point... holy...

[@MsNellyuan]

The problem with new tech is , the laws and copyright stuff always take awhile to catch up . In this case , it has not yet catching up with AI . If anyone is using the stock photos to train these AI models , those artist , photographer and models deserved to be compensated . Without those stocks , these AI models simply wouldn't exist . AI aren't thinking like human (yet) , they don't really understand the anatomy of human , the lightings , the physics of the real world , contrary to popular belief. They are only hanging onto the specific patterns trained using the stocks , which can sometimes produce accurate result but most of the time , not really because they do not understand the fundamental of it like human artist. The same reason why they always get the delicate fingers wrong because they don't know how fingers can't twist a certain direction.

[@KingZero69]

CLOWN TAKE LIK BRO 🤡

[@andrewsidamo2469]

Aren't you allowed to use stock images with watermarks if you're OK with them?

[@alfons6350]

The ai people gave me still shivers.

[@urbaguilera]

if it looks too perfect, flawless, it's probably AI 😀

[@hehheheheheehehehehehhehehehhe]

0:24 that graphic makes this video look like it was made for 4 year olds

[@Flopsaurus]

If you're going in and editing all the tell-tale details out, it's not really AI generated anymore

[@lun321]

I heard you can do this on your own. I'd love to learn how.

[@Markcain268]

Pretty soon, Ai will be able to comment on social media to save us the bother so we can get on with more important things.

[@Maxmillion_Gregory]

I fail miserably. Out of 4 pictures I choose the the fake one as real & the real one as fake. It isn't easy to distinguish between AI gen & real images these days.

[@ankushjoy4322]

The cat one got me I thought the middle one is ai but left one I didn't expect

[@jackbauer2698]

Who isn't fake these days tbh.

[@infographie]

Excellent.

[@CriaVielen]

For the first one, the woman on the left has misshapen pupils and a gap splitting one of her teeth in half

[@tiffanyw3794]

Can you do a video on the updated camera raw filter. It looks different from the one in your mock-up video

[@leedodge4227]

I hit less than 50% with four correct, so I did worse than a coin toss. I was watching on a tablet, on my monitor, maybe a little better.

[@HashebAnzar]

I scored 0. 😅

[@SebastianSkadisson]

Gonna make this post a bit spaced out so it doesn't spoil. Here's my score and what happened. I got 6 points, which is including the one bonus point you gave out for one of the gotcha moments. Apples: i picked the left one as the real one but none of them were real -> 0 points. Ladies under Cherry Blossom: i picked the right one because the left one had that typical "balenciaga AI" flair in it's facial structure -> +1 point. Bodywash items: i chose AI because of the way the relfections just cut off, especially between the two dark tubs on the left side, and vanished on some items and your robotic movement hint -> +1 point. Wine and coffee (what a combo): Chose the right side, the coffee, as real because of the ambient lighting, the wine misses a bit of ambient lighting, but both are AI-generated -> 0 points. Oh, bonus-point because of Coffee and wine gotcha -> +1 point. :D Thanks. Old waiting room vs. baroque style lounge: I picked the right side as real because the left side (the waiting room) defied physics with it's chairs missing both legs in parts, without sacking down or slanting that much. But none of them were real. -> 0 points. Hamburbur: I picked the bottom-right because i looked specifically at what i call "crowding", when AI gets a prompt to put ingredients or layers into something it often generates too much of some of the layers, meaning in this example, i thought, you'd get equal distribution of lettuce for example where lettuce in reality wouldn't always be so perfectly distributed. But i was wrong, the top-left was the only real photo. -> 0 points. Stylish Handsanitizer: Chose real because of the ambient light again, but the bottom two leaves of each branch being placed so similarly made me second guess a little for a short time. -> +1 point. Cates: Picked the middle one as AI-generated because of the way the fur in the back of the cat behaved, but two were fake, not only one, so i think i only get half a point. -> +0.5 points. Cute smiling lady in leisure wear: That was super easy - i love dimples and know how they anatomically behave. The dimples on the left didn't behave anatomically correct, so i almost instantly picked the right one as real. -> +1 points. Landscapes: Picked the left one as real because the other two looked like Bob Ross paintings made hyper-realistic to me, but the middle one was real too, so only half a point. -> +0.5 points. Chocolate: Chose AI-generated because the chocolate didn't interact with what i thought was hot chocolate sauce, but it was very hard because the person in the reflection looks a bit like Gollum thanks to the perspective. But it was real, just clever usage of real world materials. -> 0 points. I'm working with AI myself and my trick in determining if something is AI-generated or not is the same way teachers read the essays of their pupils: "Did the creator understand the practical theory behind what they have created?". Like with physics of a chair or the anatomical behavior of how dimples move the skin on a body.

[@Rangeela-Rasool]

I thought Ai image means whole image is generated by Computer itself, but no Ai tool need source to generate image

[@TheDmitrii13]

Ну, после того как я обработаю фото в редакторе, оно тоже будет уже не фото в полной мере, а созданной на его основе картинкой. То же делает и генератор изображений.

[@tonyjames5444]

I remember back when PS first started to take off and we heard it'll spell the end for many areas of commercial photography, it didn't happen, then along came digital and we heard the same thing and that also wasn't the case. Not really heard the same said about AI which is ironic because this will without doubt spell the end for many photography businesses.

[@publicminx]

the more images and videos are AI generated the more the AI learns from 'its own' created stuff in the net ;)

[@natalieminks7412]

I scored 7 of 10. Since AI has to use existing images wouldn't that be in the rehelm of copyright infringement? Many humans have been sued for images looking very similar to others and since AI has to use other's work, since it's can't create something on it's own. Who is responsible for the infringement, the company which wrote the code or the person typing in the idea to be generated?

[@Lulustime]

Can you respond to make my day I would really appreciate that also your Chanel always cheers me up when I watch it and it takes my mind of the stress I have

[@someguy7582]

Midjourney has a "house style" that you can spot easily once you've seen enough of its images. I can only describe it as like a heavily Photoshopped studio photo lit from the side/top to emphasize a 3D shape, deep shadows that almost lean chiaroscuro, and a subtle soft glow. The images that seem too "flat" in comparison are real. Raw mode reduces the effect, but the multiple rounds of RLHF and aesthetic scoring pushing it towards that style are baked into the base model.

[@KingZero69]

raw mode: exists. 🤡

[@WhatLurksBeneath]

Also --s 0

[@Sotoskiller2009]

I didn't get any right but one. It's spooky tbh

[@wiriadinata]

Dengggg....mind blowing

[@dcedce2907]

How to find out if it's AI of not. Pull the plug, yank the batteries. Now you know.

[@Foodiefoto]

That is crazy...I got 6 of 10 correct, but Im not gonna lie there were two that were simply a guess...

[@nightmarejungle5871]

if u didnt tell me the glass of wine picture was ai generated id probbly never knew thats insane i tought its real cause it looks very realistic

[@Aymen19827]

He said wich is real wich implies one is real And because there is not a real one so i chose the most realistic

[@EEEEEEEEEEEEEEEE1]

6:04 look at cheese

[@SIMPLIFIEDFILMING]

I was close to about 80% correct. Using my advantage as a filmmaker and how light reflects in different lenses I was able to establish real from fake. At least 80% correct

[@sadepennbrook]

Watermarks? Uh oh...😬

[@ernie28ernie]

Really hard to tell what's photographed and what's generated... I'm really glad, that photography is 'only' my hobby and I don't have to earn money with it! I guess AI will cost a lot of photo-related jobs... Sadly, I fear, we will get more and more pseudo-perfect, but boring content. I see copyright laws from the consumer as well as from the creator side, and they feel off. But training a commercial AI with non-free content cannot be right. Pandoras Box?

[@Lulustime]

If you respond that would make my day

[@Fleischkopf]

those AI cats fall right in my uncanny valley. The eyes look super weird. i could tell in seconds which ones were fake. but with the locations and products not at all 😅 damn, that middle cat gives me the creeps please do never ever show it again 🤣

[@marlaura5079]

I can always tell if some picture is AI or not. The ai is somewhat perfectly blurried. Like it was made for perfect and it is. I dont like ai at all. I see a lot of instagram accounts of ai generated models, these are fake women, ai generated people. Thats scarying. They create a whole fake persona, is not an actress dressed up as a character. So its not only changing photography, but arts, cinema, recorded performances, model work, fashion, marketing, advertisinh, design, and human hability to create things. I think ai came to destroy originality and human creativity and thats sad, Im avoiding it, I wont help kill human creative work.

[@KingZero69]

nice COPE lil bro.... 🤡😂

[@miqlas52]

... only at the first (apples) images I was wrong ...

[@Michael-rh5uo]

Crazy 4.5 out of 10

[@quicktastic]

Photos have been deviously manipulated in many ways for years now. The term 'photoshopped' has come to mean that an image is fake. The Russians have famously altered images to remove certain people going back to the Stalin era. AI is just making it easier and faster. It will probably lead to a sharp drop in the value of photography as it will become nearly impossible to present any image as original. Already established photographers with good reputations may be able to continue on for a while, but younger people will become less and less interested in actual photography as an art.

[@historyoflahore4408]

5

[@Lulustime]

I love your channel

[@mutayyab01]

Amazing

[@zenblue-bk]

How about 3d clone , photos are easy to make

[@micbab-vg2mu]

Thank you for great video:)

[@EEEEEEEEEEEEEEEE1]

3:40 ther was unrealistic reflection

[@jasmarr]

This is insane!! .. I only got one correct 🤯. Ai is better than the real ones here.

[@rasolzamini8678]

Kat was hardest !

[@mikekobal]

I failed miserably on all 😭

[@cvonp]

Another brilliant video, Unmesh👍 Midjourney 6 is an amazing advancement, unfortunately it still has quite a lot of difficulty with human hands and feet - often misaligning/forgetting fingers, or my personal favorite: attaching hands and feet (or even heads!) facing the wrong way - is that supposed to be hello? Goodbye? 😂

[@oualid_elrey]

1 is my score 😂😂

[@MrPeace-mm7ws]

I can clearly see in Unmesh eyes AI is going to dangerous for photographers and image retoucher.

[@AeditZ]

8. hardest part is landscapes

[@toddysurcharge771]

Same for me on both counts. Portraits are easier because AI pop too much like they were shot with lenses that don’t exist (like a 70-200 f/1.8)

[@josiahgeoffrey]

My score was 3 or 4. Not bad, but not good either haha

[@editit1540]

I got 7.5/10 pretty good I guess

[@namakudamono]

Great video! AI image creation is becoming so convincing! I would love to see a video (or a series of videos) related to integrating AI image generation with real photography. For example, I currently shoot home interior products at actual model room locations in order to show the product in a typical environment. It would be incredible to able to use s studio-shot image of the product, isolate it from the solid colour background, and then use AI to generate a realistic background environment with the correct angle and lighting etc. Are there any tutorials for achieving this kind of thing online? Keep the great work!

[@mandolina6280]

We're living in a century of lies.

[@valdemarcaballero5298]

with so many cell phones (with camera) we are F$·***K, now with AI we are DEATH 😭😭😭😭😭 ✝

[@bakunawa-wu2zh]

it's definitely stealing stock photos 😅

[@markbillson4919]

To be honest non of the photos here are or look "Real". A more suitable title for the video would be "Has this image been heavily Photoshopped, or is it AI generated?"

[@annabyhoff1288]

So what’s the point of making a fake portrait of a person who doesn’t exist?