I hadn't been exposed to the new GPT-Image-2.5 model before this, so didn't have any "tells" to go off of, and found it almost impossible to begin with, but it seems the brain is pretty good at adapting for visual stuff! After 2 rounds, I was able to build up a 16-image combo, answering most within a second or two.
I think the "tell" I ended up with is that almost all of the AI images tend to center around a very obvious main "subject" (a flower, a bike, a wrench, etc) – presumably an inherent artefact of generating from a prompt – with everything else around the subject having a very strange depth to it. The depth, focus, and bokeh around the main subject just never look quite right. If you look at an image and find it has a very strong central subject with a little too much depth separation than expected, there's a good chance it's AI.
I recognise that this method of finding a consistent "tell" will probably not hold for future models, unfortunately. Everything else looks almost perfect at this point.
It's hard to put into words but the "texture" of GPT-Image-2.5 is just a bit off, like the difference between blue noise and white noise. It's like someone turned up the local contrast just a little too high.
There are occasionally more obvious tells as well, if you look closely. Some examples that I saw:
- A boat cast a shadow, but the mooring rope didn't
- A bench leg was out of perspective, and didn't touch the ground
- In a pile of logs,the ends were sharpened in an unnatural way[0]
- In several there seems to be an inconsistent contrast between something that looks more dirty, worn, weathered, etc. while the surroundings are more clean and smooth.
Some also feel off in a way that I can't really articulate.
[0]: I think the relationship to the prompt was interesting here, the prompt actually asked for a stack of wooden fence posts, but they were really thick for fence posts, and looked more like firewood with a sharpened point. Interestingly, the prompt included the phrase "without excessive sharpening" which I think meant increasing the sharpness of the image, but I wonder if the AI interpreted as referring to the sharpness of the fence posts.
The fonts it uses for signs and posters. They often don't match the scene. The posters on a corkboard next to school gymnasium don't look like they were made by students or teachers, most likely amateurs at graphic design, they look professionally made. As for signs, a rustic looking sign isn't going to use helvetica-like typeface, it looks entirely out of place. Again, it looks like a professional made it in 2026, not an amateur merchant 50 years ago.
I realized this is only somewhat difficult because of the 10s timer. Most people probably don’t spend more than 10s looking at a picture on social media, excellent project.
The scene should feel incidentally noticed during daily life, not arranged for an advertisement. Natural perspective and subtle everyday wear. No cinematic grading, stock-photo staging, deliberate blur, artificial grain, fake low resolution, extreme background blur, exaggerated HDR or oversharpening. No collage, border, caption or added watermark.
At some point during the process of building up this collection of turd-polishing techniques, the thought should’ve occurred that this is actually a lot of effort to get generations confused for reality. But I suspect this process was being carried out by something that couldn’t have an original thought.
A troll farm will happily go through that effort then (if it's a high effort part of the campaign rather than just 'flooding the zone') pick the images that are most likely to pass.
Another pro-AI psyop to show that AI generated content is perfectly fine, created by cunning machines and spread through mindless meat proxies glad to do their bidding.
We all been forced to help train Google's CAPTCHA for years for free, with no benefit. At least here you decide to opt-in or not, and it can potentially help you be more... discerning.
Highest I got was 26 in a row. AI seems to love generating "close up" images of a subject. Its like every image has to have an obvious subject (that was in the prompt), and it struggles to make scenes without a clear subject and more chaotic vibe. I notice this a lot with these "AI movies", every single camera shot has an obvious subject that makes them unbearable to watch, they can't just chill out and have a messy scene.
The time limit did make it quite challenging. I got 1,000 points with a max streak of 8 in a row. On second playthrough I got 3,550 / 12. It seems the AI images all come from a particular area of latent space that's quite easy to recognize once you get dialled in.
So it's more of a "can you detect the GPT Image 2.5 house style", which is an easier problem.
I agree, there's patterns that aren't hard to recognize if you have some experience.
I got 2400 with 10 in a row on first attempt.
Here's the entire decision matrix:
- Is the color, mood, composition, and level of detail exactly like most Nano Banana Pro images? If yes, AI.
- Is the composition/framing utterly perfect? If yes, AI.
- Is the subject ephemeral or random, like a candid photo or something otherwise unlikely to be photographed? If yes, not AI.
- Final decision, if there's any doubt, go entirely off of how centered the subject is.
I think some of this has more to do with the data set than the methodology. You could easily come up with a bunch of examples to fool me specifically, but alas, they didn't.
Neat app but it makes me sad. Also I'm very curious as to where the real photos come from and how we can verify that they are indeed real. The timer thing doesn't really add any value, just stress.
My most effective strategy is to ask: is there any reason for a human to take this picture?
One AI-generated image was a bedside night stand, with no clutter other than a set of keys on top of a receipt. It looked perfectly real, no glitches or weirdness... other than the idea of someone taking a picture of it.
But then, some people specialize in taking photos that others would not because they have in mind stock photos, art or something else. They see the world differently.
it's harder to tell at a glance rather than pulling out a magnifying glass and counting the fingers of everyone in each group; the time limit pressures the player into making underqualified guesses..
Highest score is 16,800.
Highest streak is 60 in a row.
I just came to HN to see if I won!
Also, I do research on how people perceive images and video. I’ve published a couple papers and did my dissertation on how folks understand moving images.
If you’re the developers of this game I’d love to help you parse this data. B L T P H D at V T dot E D U.
Something we’ve been discussing in our lab is if people can detect the “taste” of AI. Seems if you play this game enough times, you can at least recognize the differences pretty consistently. I’d love to know how rounds of play improves people’s accuracy.
Next time you eat at McDonald’s try to notice the similarities in how things taste. McDonald’s has a flavor profile across all their foods. Everything is a little sweeter and a little saltier but exactly “McDonald’s-like.” My favorite example is the sauces for chicken nuggets. Regardless of the flavor you get, all the sauces have the same consistency and sort of taste like sweet and sour sauce. The honey mustard, the bbq, the buffalo sauce, and even the ranch all taste McDonaldish.
My working theory is that AI generated content also has a tatse. Once you can taste it, you are able to easily discern genuine content (which has random tastes) from generated content (which taste like the model it comes from).
Not bad. I’d appreciate a much longer time limit. Got really rushed at the end. But this was crazy hard on my phone, I was straight guessing most of the time. I think I could be a little better with more time on a desktop screen.
Another idea would be games that were all pictures of a single category- like 10 AI and 10 real pictures of basketballs, or food on plates, etc.
But I think that your experience is actually the useful one that proves a point. Most people consume content on their phone (e.g instagram) and would usually just scroll by photos. They wouldn’t deeply analyze them.
It would be cool if they published stats on how well people do on average, and how that changes over time and with different models. I feel like in a few years we'll either all be much better at this or completely desensitized, and sadly I'm guessing we're trending towards the latter.
I'm sure it's not the intended way to play but guessing randomly seems to have a net positive score, so getting a high score is as simple as spamming the buttons as fast as you can. I got 10k on mobile through this strategy.
The general lack of humans helps conceal the difference though. Bypasses the uncanny valley problem. I'd contend that most of us would do a bit better on truly random photos
Surprised how good the nature/scenery shots are though. When I last played with a (local) image gen I tried making forest scenes and it was terrible at it
12/17. Due to the speed, I felt like I was just randomly guessing, but 70% correct seems better than random guessing. I did suspect some things.
How mundane the AI images were felt like a means to trick people. It doesn’t seem like most people making AI images and posting them are generating still life images of everyday life. What would be the point?
They're not challenging that at all. The time limit makes it a game rather than an experiment. "You can't tell it's AI if you don't have enough time to examine the details" is not exactly a compelling claim...
Another poster in this threads commented that their strategy was to question why anyone would take the photo that was presented. I think this touches on one of the most important issues with this whole schoolyard level argument of "I can tell" and "no you can't." That is: what is even the point of all of this?
I had funding for a project I had proposed at work yanked out from under me because I had refused to vibe code the whole thing. I had proposed the project to create a tool that would be beneficial to my division, which represents a good 25% of the company's revenue (we're a consulting company). It had the enthusiastic support of my entire direct leadership line, very heavy hitters in the company. But the executives in charge of the internal R&D budget have their own agenda of pushing AI out on everything to gin up this idea that our company is "AI forward." Nevermind that our IT infrastructure is a joke, we need to play a game to convince non-existent investors that we are suddenly, out of the blue, after 50 years, a "technology company" when what we really do well is manage client relationships.
The executive I was closest to at one point told me I had failed because I hadn't used AI to tell me what the requirements for the project should be, that I was just doing "whatever I wanted."
If I were really doing whatever I wanted, I would not be building some kind of data management analysis workflow whatever. I designed the project around exactly the sort of thing that would benefit our direct line analysts. I don't need to be told what the requirements are, I know then intimately from working with these people for years. But because I was not using AI to tell me what to build and using AI to build it for me, I am being "behind the times." Another executive told me it wouldn't matter if I succeeded at the project if I didn't use AI to build all of it. Another time he told me it wouldn't matter if the project was any good, so long as I used AI to build it.
What is even the point? Why would I abdicate everything to AI? I know what I'm doing and I know how to do it. But that is unimaginable to the AI-pilled executive class. They are so used to not knowing how to do anything that they can't imagine anyone else actually knows how to do their job.
And again, how is "AI or human" relevant to discussing? Given the potential of an indistinguishable discussion (which will eventually be the case). Although also eventually it's AI that will consistently provide the superior discussion experience, then... what? Those against AI actively seek out inferior experiences? Should be interesting.
Well, now instead of paying the photographers for the stock photos to use as headline pictures, people instead pay the AI firms for the stock pictures to use as headline pictures.
The same could be said of commissioned paintings when photography became a thing. And yet here we are, and I doubt anyone - except maybe artists - would say the world is worse off.
Precisely. There is no need to wait for "the position" to be formulated since a) it won't be; b) it's mostly "AI is bad because, full stop" anyhow, not even a strawman.
> it's mostly "AI is bad because, full stop" anyhow
Ooh, the rare* double straw man. Since you look like you could use the exercise in intellectual honesty, I’ll pose the question directly: can you think of any reason that the ability for anyone to generate fake photographs indistinguishable from reality with no skill, time, or money required might make the world a worse place? Cause harm?
I hadn't been exposed to the new GPT-Image-2.5 model before this, so didn't have any "tells" to go off of, and found it almost impossible to begin with, but it seems the brain is pretty good at adapting for visual stuff! After 2 rounds, I was able to build up a 16-image combo, answering most within a second or two.
I think the "tell" I ended up with is that almost all of the AI images tend to center around a very obvious main "subject" (a flower, a bike, a wrench, etc) – presumably an inherent artefact of generating from a prompt – with everything else around the subject having a very strange depth to it. The depth, focus, and bokeh around the main subject just never look quite right. If you look at an image and find it has a very strong central subject with a little too much depth separation than expected, there's a good chance it's AI.
I recognise that this method of finding a consistent "tell" will probably not hold for future models, unfortunately. Everything else looks almost perfect at this point.
It's hard to put into words but the "texture" of GPT-Image-2.5 is just a bit off, like the difference between blue noise and white noise. It's like someone turned up the local contrast just a little too high.
Exactly! The generated images taste different. Also…if you see a perfectly folded towel, it’s probably AI. At least in this set of images.
There are occasionally more obvious tells as well, if you look closely. Some examples that I saw:
- A boat cast a shadow, but the mooring rope didn't
- A bench leg was out of perspective, and didn't touch the ground
- In a pile of logs,the ends were sharpened in an unnatural way[0]
- In several there seems to be an inconsistent contrast between something that looks more dirty, worn, weathered, etc. while the surroundings are more clean and smooth.
Some also feel off in a way that I can't really articulate.
[0]: I think the relationship to the prompt was interesting here, the prompt actually asked for a stack of wooden fence posts, but they were really thick for fence posts, and looked more like firewood with a sharpened point. Interestingly, the prompt included the phrase "without excessive sharpening" which I think meant increasing the sharpness of the image, but I wonder if the AI interpreted as referring to the sharpness of the fence posts.
The fonts it uses for signs and posters. They often don't match the scene. The posters on a corkboard next to school gymnasium don't look like they were made by students or teachers, most likely amateurs at graphic design, they look professionally made. As for signs, a rustic looking sign isn't going to use helvetica-like typeface, it looks entirely out of place. Again, it looks like a professional made it in 2026, not an amateur merchant 50 years ago.
Exactly, if the words are too easy to read, and clearly not part of a photo used to advertise - then it’s probably AI.
I realized this is only somewhat difficult because of the 10s timer. Most people probably don’t spend more than 10s looking at a picture on social media, excellent project.
I thought that at first, but a comment below enlightened me. These types of games are a great way to get free training data.
Great point, in the privacy disclosure they confirm they're collecting the training data.
> Reality Check records visits, round starts and finishes, image views, answers, response times....records are stored in Cloudflare D1
Well I guess it's good that I picked AI image for every image...
Every generation prompt has this suffixed:
At some point during the process of building up this collection of turd-polishing techniques, the thought should’ve occurred that this is actually a lot of effort to get generations confused for reality. But I suspect this process was being carried out by something that couldn’t have an original thought.A troll farm will happily go through that effort then (if it's a high effort part of the campaign rather than just 'flooding the zone') pick the images that are most likely to pass.
freethink is inefficient, schedule this biomass to the remediation chamber
> 60 seconds to find out.
> 90s round, 10s per image
> Play 60 seconds
This is confusing. What does "90s round" mean if it's a 60 second game? Is it about the 1990s?
"Made with ♥ by Codex"
Another pro-AI psyop to show that AI generated content is perfectly fine, created by cunning machines and spread through mindless meat proxies glad to do their bidding.
/s but also not really.
Slop all the way down, I guess at least if we’re being charitable. If we’re being less charitable this is trying to gamify getting free training data.
We all been forced to help train Google's CAPTCHA for years for free, with no benefit. At least here you decide to opt-in or not, and it can potentially help you be more... discerning.
Now I'm worried Google adds this check to its captcha questions to improve it's own model: tell the ai image
Wait, did I miss something? Wasn't that the point the entire time?
Highest I got was 26 in a row. AI seems to love generating "close up" images of a subject. Its like every image has to have an obvious subject (that was in the prompt), and it struggles to make scenes without a clear subject and more chaotic vibe. I notice this a lot with these "AI movies", every single camera shot has an obvious subject that makes them unbearable to watch, they can't just chill out and have a messy scene.
The time limit did make it quite challenging. I got 1,000 points with a max streak of 8 in a row. On second playthrough I got 3,550 / 12. It seems the AI images all come from a particular area of latent space that's quite easy to recognize once you get dialled in.
So it's more of a "can you detect the GPT Image 2.5 house style", which is an easier problem.
I'd appreciate an untimed mode.
I agree, there's patterns that aren't hard to recognize if you have some experience.
I got 2400 with 10 in a row on first attempt.
Here's the entire decision matrix:
- Is the color, mood, composition, and level of detail exactly like most Nano Banana Pro images? If yes, AI.
- Is the composition/framing utterly perfect? If yes, AI.
- Is the subject ephemeral or random, like a candid photo or something otherwise unlikely to be photographed? If yes, not AI.
- Final decision, if there's any doubt, go entirely off of how centered the subject is.
I think some of this has more to do with the data set than the methodology. You could easily come up with a bunch of examples to fool me specifically, but alas, they didn't.
Neat app but it makes me sad. Also I'm very curious as to where the real photos come from and how we can verify that they are indeed real. The timer thing doesn't really add any value, just stress.
The result page has links to the source of each real photo.
My most effective strategy is to ask: is there any reason for a human to take this picture?
One AI-generated image was a bedside night stand, with no clutter other than a set of keys on top of a receipt. It looked perfectly real, no glitches or weirdness... other than the idea of someone taking a picture of it.
Not a perfect strategy, though; I thought this one was AI for just that reason: https://commons.wikimedia.org/wiki/File:Potted_plant_1_2018-...
But then, some people specialize in taking photos that others would not because they have in mind stock photos, art or something else. They see the world differently.
There are tons of real stock photos with that type of mundane content.
Stock photos and AI images might as well be in the same category tbh.
[dead]
Impossible to do on a phone. Zooming in is misinterpreted as a choice. Fix the game.
Also the time limit is stupid and pointless.
>Also the time limit is stupid and pointless.
it's harder to tell at a glance rather than pulling out a magnifying glass and counting the fingers of everyone in each group; the time limit pressures the player into making underqualified guesses..
it's not pointless.
Someone got a low score
I didn't get the zoom bug but zooming in to look at it and zooming out to get back to the buttons to practically the entire 10s
The time limit is obviously an important part of this
Highest score is 16,800. Highest streak is 60 in a row.
I just came to HN to see if I won!
Also, I do research on how people perceive images and video. I’ve published a couple papers and did my dissertation on how folks understand moving images.
If you’re the developers of this game I’d love to help you parse this data. B L T P H D at V T dot E D U.
Something we’ve been discussing in our lab is if people can detect the “taste” of AI. Seems if you play this game enough times, you can at least recognize the differences pretty consistently. I’d love to know how rounds of play improves people’s accuracy.
Next time you eat at McDonald’s try to notice the similarities in how things taste. McDonald’s has a flavor profile across all their foods. Everything is a little sweeter and a little saltier but exactly “McDonald’s-like.” My favorite example is the sauces for chicken nuggets. Regardless of the flavor you get, all the sauces have the same consistency and sort of taste like sweet and sour sauce. The honey mustard, the bbq, the buffalo sauce, and even the ranch all taste McDonaldish.
My working theory is that AI generated content also has a tatse. Once you can taste it, you are able to easily discern genuine content (which has random tastes) from generated content (which taste like the model it comes from).
I guess the UI is ai generated, and it impossibly to see the images well enough to judge on a phone.
Hmm… training data siphon?
Not bad. I’d appreciate a much longer time limit. Got really rushed at the end. But this was crazy hard on my phone, I was straight guessing most of the time. I think I could be a little better with more time on a desktop screen.
Another idea would be games that were all pictures of a single category- like 10 AI and 10 real pictures of basketballs, or food on plates, etc.
But I think that your experience is actually the useful one that proves a point. Most people consume content on their phone (e.g instagram) and would usually just scroll by photos. They wouldn’t deeply analyze them.
I think the point is to be able to tell at a glance. People don't examine every image they look at.
I mean that is a conclusion that I agree with, but I’d also like to see if I can do better if given more time on a bigger screen.
It would be cool if they published stats on how well people do on average, and how that changes over time and with different models. I feel like in a few years we'll either all be much better at this or completely desensitized, and sadly I'm guessing we're trending towards the latter.
Good reality check for people who proudly boasted about "always" being able to spot AI images.
My biggest clue is the odd repeating details. Especially on grass, tree leaves, pores, etc
Not all image models (or sampler settings) do this, but with gpt it's very obvious.
With this in mind, I misjudged 3 images on the default game.
I failed horribly and might as well used a random coin toss
try a couple times and you’ll start getting much better than coin toss.
The trouble with these tests are that they are themselves training AI. I don't want to let designers know the "tells".
I don't get the scoring system: I got 0 points, but I got a streak of 2 in a row. Bizarre.
Please give us a fraction or percent of how many we got correct.
Is this a crowd sourced training effort to make their image generator better?
not doable in short time with high certainty which means that from now on an image is only worth a thousand words
Got 4k by just spamming 2.
I'm sure it's not the intended way to play but guessing randomly seems to have a net positive score, so getting a high score is as simple as spamming the buttons as fast as you can. I got 10k on mobile through this strategy.
Play the lottery! I got -1350 spamming 2, -400 spamming 1. Feeling unlucky.
I got 14/16 correct, and the two I got wrong had no people and I was unsure about both. Should have went with my gut.
Wow - I did much worse than I'd have thought.
The general lack of humans helps conceal the difference though. Bypasses the uncanny valley problem. I'd contend that most of us would do a bit better on truly random photos
Surprised how good the nature/scenery shots are though. When I last played with a (local) image gen I tried making forest scenes and it was terrible at it
TIME'S UP
2,050 points Best streak: 9 in a row
Hard af.
12/17. Due to the speed, I felt like I was just randomly guessing, but 70% correct seems better than random guessing. I did suspect some things.
How mundane the AI images were felt like a means to trick people. It doesn’t seem like most people making AI images and posting them are generating still life images of everyday life. What would be the point?
this is fair only if i have enough time to decipher.
Now add Geoguesser-like battle royale mode with accuracy points for guessing which model generated the image
best strategy: just click as fast as possible.
My method is simple:
Did I create the image myself?
Yes: Not AI
No: AI
AI vs Aye, it was I
Can I tell what’s AI generated?
Well if I can I’m not going to help train more slop image generation.
Bouncing on AI generated button as fast as possible without looking at the images got me 6,000 points.
We get it. AI slop is getting good enough to replace photography. This isn't good for humans, and makes the world a worse place.
The fact that the suck machine works isn't something to be proud of.
That’s not what they’re saying. They’re challenging the “I can always tell” people who are both proud and regularly wrong.
As fake news becomes more and more real and with photo evidence, this matters more and more.
They're not challenging that at all. The time limit makes it a game rather than an experiment. "You can't tell it's AI if you don't have enough time to examine the details" is not exactly a compelling claim...
Another poster in this threads commented that their strategy was to question why anyone would take the photo that was presented. I think this touches on one of the most important issues with this whole schoolyard level argument of "I can tell" and "no you can't." That is: what is even the point of all of this?
I had funding for a project I had proposed at work yanked out from under me because I had refused to vibe code the whole thing. I had proposed the project to create a tool that would be beneficial to my division, which represents a good 25% of the company's revenue (we're a consulting company). It had the enthusiastic support of my entire direct leadership line, very heavy hitters in the company. But the executives in charge of the internal R&D budget have their own agenda of pushing AI out on everything to gin up this idea that our company is "AI forward." Nevermind that our IT infrastructure is a joke, we need to play a game to convince non-existent investors that we are suddenly, out of the blue, after 50 years, a "technology company" when what we really do well is manage client relationships.
The executive I was closest to at one point told me I had failed because I hadn't used AI to tell me what the requirements for the project should be, that I was just doing "whatever I wanted."
If I were really doing whatever I wanted, I would not be building some kind of data management analysis workflow whatever. I designed the project around exactly the sort of thing that would benefit our direct line analysts. I don't need to be told what the requirements are, I know then intimately from working with these people for years. But because I was not using AI to tell me what to build and using AI to build it for me, I am being "behind the times." Another executive told me it wouldn't matter if I succeeded at the project if I didn't use AI to build all of it. Another time he told me it wouldn't matter if the project was any good, so long as I used AI to build it.
What is even the point? Why would I abdicate everything to AI? I know what I'm doing and I know how to do it. But that is unimaginable to the AI-pilled executive class. They are so used to not knowing how to do anything that they can't imagine anyone else actually knows how to do their job.
That sounds like a really horrible situation and some terrible leadership and I'm sorry to hear that.
... was I the person you meant to respond to, though?
Yes, it was meant to add to your point about the stupid "can/can't tell" argument.
Yeah, exactly, they're right, we're successfully making shit worse. Yay us.
> makes the world a worse place
In what way?
First, prove you're not an AI, and then we can discuss whether being able to generate AI output indistinguishable from human output can cause harm.
And again, how is "AI or human" relevant to discussing? Given the potential of an indistinguishable discussion (which will eventually be the case). Although also eventually it's AI that will consistently provide the superior discussion experience, then... what? Those against AI actively seek out inferior experiences? Should be interesting.
Well, now instead of paying the photographers for the stock photos to use as headline pictures, people instead pay the AI firms for the stock pictures to use as headline pictures.
Which is "worse" by some metric, I'm sure.
The same could be said of commissioned paintings when photography became a thing. And yet here we are, and I doubt anyone - except maybe artists - would say the world is worse off.
Stop, stop, he’s already dead! You guys are getting straw everywhere.
Couldn’t wait more than 3 minutes for someone’s actual position before setting up that straw man to take down?
The actual position: "First, prove you're not an AI, and then we can discuss..." Yeah, no, I'll take myself being strangely prescient this morning.
The partial comment you’re quoting is
- not even a position
- not something you were remotely “prescient” about
Precisely. There is no need to wait for "the position" to be formulated since a) it won't be; b) it's mostly "AI is bad because, full stop" anyhow, not even a strawman.
> it's mostly "AI is bad because, full stop" anyhow
Ooh, the rare* double straw man. Since you look like you could use the exercise in intellectual honesty, I’ll pose the question directly: can you think of any reason that the ability for anyone to generate fake photographs indistinguishable from reality with no skill, time, or money required might make the world a worse place? Cause harm?
* I wish
No, but we still can tell that this web site is AI-generated slop.
uh yeah...we need hardware attestation and a protocol to make it work cross platform like, yesterday...
holy shit