This Image Does Not Exist | Hacker News Reader

This Image Does Not Exist

thisimagedoesnotexist.com

170 pointsalbertzeyer127 comments

I scored 27/30, with around 12 being human-created.

I've been keeping a close eye on DALL-E 2, especially images posted on Twitter and Reddit[0], and the biggest sign that it's a DALL-E 2 image to me are the edges of objects not being as crisp as you'd expect in a human-created image.

The composition that DALL-E 2 creates is typically fantastic, but it isn't perfect when it comes to the edges, details, and textures (fabric, for example).

I still think it's incredible for inspiration — I don't think a typical artist/designer will use DALL-E 2's output as the final image, but rather as a great starting/mid point for their final image. I can see myself using DALL-E 2 to generate something akin to a wireframe that I'll then go in and add the details to manually.

[0] https://www.reddit.com/r/dalle2/

Biggest signs for me were the edges you mentioned, compositions that werent technically bad but were visually uninteresting (and had minor logical flaws like missing details in a reflection), but also the subject matter is the sort of thing that could be described in a succint text prompt designed to challenge a machine (I actually guessed a couple of the prompts, and "tessellated Hello Kitty" and "Origami car" are less likely to be human art concepts than challenges for machines). I'm browsing on a phone so some of the fine detail is lost.

The fact that I'm using the subject matter as much as the image itself to discern AI art is a sign of how good the tech is though. Especially since the subject matter in question is typically "stuff that would be really challenging for an AI to conceptualise" and the image itself (not necessarily the AI's only attempt of course) is a sound attempt at rendering it.

I see most of the utility as the opposite: the artist still needs to come up with the original concept, the overall compositions are unlikely to be better than an artist would conceive, but even in the hands of a non-artist a bit of text can produce "good enough" images (that can be cropped or touched up to remove rough edges) instantaneously which would take a lot of time for a skilled artist to produce.

> are less likely to be human art concepts than challenges for machines

Counter-example: there was one image I got that was a Gameboy as a concrete brick. That seems like a prompt you’d give DALL-E 2, but it was actually of human creation

Yeah, but I guessed that one correctly because that sort of art (taking symbolism like a game-boy and putting it in a "cool" medium) is pretty common on the internet already.

It makes sense that someone would think to make that because a gameboy is basically already a grey brick. So it's really just a change of texture- it's a comparison of two things that look similar that a person would probably make but an AI might not. it has some artistic intent.

The other DALL-E prompts are just "a <random subject> doing <random thing> in <random medium>", like "a dog building a fence in watercolor".

Would it make sense for a person to paint a dog building a fence in watercolor? Probably not unless they're writing a children's book.

Biggest/best tell for me is not the image but the subject matter. The images with some interesting twist always come from the humans. For instance one image shown to me had human hands at a laptop, but they were coming through the screen to type. That is playing with human perception, twisting reality to make an artistic point (the self-typing laptop where the user comes from inside the machine). I've never seen an AI-generated image that projects any artistic depth. They are beautiful and sometimes funny, but never interesting.

Of course most human-created images are equally shallow, but if you do see depth then you know it came from a human. Put this test to an art historian/appraiser or architect and they would probably score very well.

This was my "tell" as well. I don't know a ton about DALL-E, but it seems like most of the pictures are in the style of "show X doing Y in the style of Z": "Show a bear carrying a pizza in a realistic style". The composition of the DALL-E ones is almost "random association", e.g. acid-trip type associations. For each image I'd ask myself "What purpose would a human have for creating this?" and it usually helped get me the right answer.

But what if the human is tasked with creating an image that looks like it was created by a machine? If you gave human artists the same taskings (X and Y in style of Z) then they would also create a bunch of purposeless images. That is the dataset I would create if I wanted to promote my AI project.

I think that a human's portrayal would sometimes still be noticeably different.

For example, if you prompted a human and DALL-E to draw "a fork in an electrical outlet", the human would probably portray the fork as shocking and exploding, whereas DALL-E would probably just portray a fork being put into an electrical outlet and nothing happening. Maybe that's a bad example because there may be enough depictions of forks in electrical outlets for DALL-E to learn to depict them as exploding, but my point is that DALL-E can't reason generally about the composition of the image.

The human-created images had something interesting to say, whereas the computer-generated images did not. I do find this to be an interesting variation of the Turing Test.

I made it "honestly" until photo 18 (where I got a score of 17/18), then I realized that all photos were auto-generated and spammed the "robot" button. Apparently there was a single human-generated picture in my set.

> The composition that DALL-E 2 creates is typically fantastic, but it isn't perfect when it comes to the edges, details, and textures (fabric, for example).

For me the trick for discerning them is asking "what's the purpose of this detail?", namely: if a figure is writing, what are they writing? If a figure is wearing a shirt, does it look like a shirt I've ever seen? Why are they standing the way they are standing? And what material is this statue made of?

There's me trying to evaluate intent, context, talent -- when instead I should look for blurry edges. I imagine a future when AI is so sophisticated and human-like they're granted all the privileges of living, sentient beings but they're still clumsy with a knife and fork and they fart too often in public.

Yeah, I used edges as a tell.

For instance, the train at dusk on a bridge was really well composed overall, but some of the bridge cables were slightly transparent.

I'm slightly surprised to report a 100% score. However, I think I'm actually recognising DALL-E-2's _style_, and not "this picture is CG". It feels more like I'm identifying the work of an artist I'm very familiar with, rather than identifying fundamentally different categories of creator.

I think all it would take to "defeat" me here would be greater variety of high-quality image generation AIs.

I expect in a few years that will be the case, unless the "style" of DALL-E-2 is something fundamental, instead of an artifact of its specific implementation.

If you open up the image in a new tab to see the full resolution and look at fine details it's very easy to spot AI ones - AI seems to especially struggle on reflective details, it obviously doesn't know it's meant to be a reflection, so it gets details subtly wrong or weird in the mirror image.

Also anything mechanical is just way off on AI generated things - a car with the fuel door in the middle of the passenger door, a missing tire but you see no axle and nothing makes mechanical sense.

Also, fucked up hands/fingers/letters and such details often give it away

Interestingly these are the kind of things -fingers, writings, reflections- that also betray the “generative” nature of dreams.

https://lucid.fandom.com/wiki/Reality_check

But still, its pretty miraculous.

Also Dalle makes square images so if it’s not square it’s human made

On my phone this was indiscernible

I can see DALL-E-2 stuff spreading around WhatsApp groups like wildfire, masquerading as something else

Even now just selecting Dalle propmpts that output things that look like basic ray tracer tests will probably make it undecidable for anyone not bothering to recrating the scene and chech that it's physically coherent.

well, Theres still plenty telltale signs of AI handwork here. Quasi-faces on places no human would place them or would articulate a detail much better. edge coherence is also very important. Seeing smudges of paint where it doesn't belong at all, lines no human would be able to make and errors only humans make but where AI struggles with.

My point is that Dalle can do things like the Cornell box pretty much perfectly, so if you want to make a test where it's impossible to decide who made it you should sample those type of prompts.

If enough humans on the pool this webpage gets its images from start copying DALL-E-2 effectively that would also "defeat" you.

I am failing spectacularly, love it.

Also - please speed up the delay switching to a new image! I bet this would feel way better if the delay was in the 0-0.5s range instead of multiple seconds. (I bet it’s some runaway JS, investigating…)

Edit: `changeImage` is blocked by `await saveDB`, and the POST is 500ing. (e.g. uBlock makes that AWS request fail fast – why desktop people are less affected)

Maybe use `Promise.allSettled` to make them parallel, and prepare the data being sent beforehand to avoid state race conditions :) thanks for leaving sourcemaps on!

> Also - please speed up the delay switching to a new image!

I came here to suggest the opposite, as I didn't have time to read the full prompt which generated the image, when it was a generated picture.

So I guess the conclusion is instead: allow users to click a second time on any of the buttons/picture to proceed at their own leisure, so the ones who want to read can read, and the ones who want to speed through it, can do that too.

Nice, that’s even better :) add some keyboard controls (maybe left for artist, right for robot?, and then either key to advance) and folks can zoom through!

Only tangentially related, I assembled a children's book made by GPT-3 and Midjourney, which is a generative art collaboration that is along the lines of DALL-E.

(Edit: see JacobiX reply, looks like it robbed the concept, possibly because I didn't give it any direction.)

I just prompted GPT-3 (davinci-002 engine on openai) to write a children's book and include illustration suggestions. I didn't even suggest a topic, just said 'children's book'. I fed the illustration suggestions into Midjourney with the qualifier of 'children's book illustration of ...' and pulled the first image of each (it always generates four).

I then asked it for a cover page, last page and a background image for the text.

I assembled the story text written by GPT-3 and images rendered by Midjourney in this google slides preso. The only real creative input i had was the font. I did iterate on the text background image prompt a couple of times to find something that would work:

(Note this link works in incognito if you prefer not to have your google login associated)

https://docs.google.com/presentation/d/e/2PACX-1vT4XWNx2SdEg...

I've been having a lot of fun with this prompt. Part in asterisks is where you set the topic. The last 'Title of the book:' is where you submit, it really helps keeping the generative part on track.

----

    Write a four page children's book with the following:
    
    Title of the book:
    Description of the cover page with enough detail for an illustrator to draw:
    Description of the background imagery used for the pages with text on them:
    First page text in quotes:
    Description of the opposite page illustration:
    Second page text in quotes:
    Description of the opposite page illustration:
    Third page text in quotes:
    Description of the third page illustration:
    Fourth page text in quotes:
    Description of the fourth page illustration:
    Description of the final page:
    
    The subject of the children's book is *how to prevent buffer overflows in c code*.

    Title of the book:

It should be noted that there is a real children's book with the same name and a very similar content: The Very Lonely Firefly by Eric Carle.

For this particular case, it seems like GPT-3 is paraphrasing the story ...

Interesting. I thought it sounded familiar. Guiding it with a subject is probably a better way to avoid that kind of thing.

Here's what resulted from my prompt above for a children's book about buffer overflows - https://docs.google.com/presentation/d/e/2PACX-1vSp7wc1xCgiv...

Doesn't always work hahaha.

Feeding it a suggested topic to avoid derivative works.

Input:

The subject of the children's book is how to autorotate a helicopter after its engines have failed.

Title of the book:

Output:

"Autorotating a Helicopter"

Description of the cover page with enough detail for an illustrator to draw: The cover will feature a large illustration of a helicopter in the air, with the text "Autorotating a Helicopter" in big, bold letters.

Description of the background imagery used for the pages with text on them: The pages will have a white background with black text.

First page text in quotes: "If you're ever in a helicopter and the engines fail, don't panic! Just autorotate."

Description of the opposite page illustration: The illustration on this page will show a helicopter with its engines on fire, falling towards the ground.

Second page text in quotes: "To autorotate, pull back on the cyclic and maintain a level attitude."

Description of the opposite page illustration: The illustration on this page will show a helicopter with its engines on fire, but the blades are still spinning.

Third page text in quotes: "To land, flare the helicopter just before touchdown."

Description of the third page illustration: The illustration on this page will show a helicopter landing safely on the ground, with the blades still spinning.

Fourth page text in quotes: "Remember, practice makes perfect! So if you ever find yourself in a helicopter with failed engines, you'll be prepared."

Description of the fourth page illustration: The illustration on this page will show a helicopter in the air, with the engines on fire.

Description of the final page: The final page will feature a large illustration of a helicopter landing safely on the ground, with the text "The End" in big, bold letters.

a few possible illustrations - https://ibb.co/c2ZYQps

Now I'm tempted to make a dynamic book with a different story each time you open the cover.

This is something that I've been thinking of as well. GPT-3 can be a bit unruly in its output but if you execute ten prompts you'll get something parsable 6 times.

24/30 You can definitely recognize DALL-E's style of mashing two elements together. Most of the time it will look smudgy at the points where two images are combined. Also, if there are any eyes in the image it will mess them up in some way, almost every time.

There is also some kind of artifact on DALL-E's images which makes gradient areas look very low color depth.

Hardest ones to get right are low quality images and images of abstract artwork or the ones that are made in a "smudgy" style already.

It is pretty good overall but I would say that it's almost impossible to fool anyone if the images were in high resolution and you could zoom them in.

It's probably most effective in small images like thumbnails for example.

It's kind of spooky to think how good it will be in a few years and what it will be used for.

Is it me or is this very depressing and scary? I always assumed the creative industries would be relatively resistant to automation, with more algorithmic jobs more easily replaceable. But with this you never need to buy stock pictures ever again, right? I saw several stock-picture-esque images of indistinguishable quality, and yet with incredible specificity to the user-supplied description.

I thought this way originally but I have realised these tools are being used right now to massively amplify the creative output of individuals. AI text to voice tech is being used for a lot of hilarious content right now. I saw an image upscailer ai used in a 1 week project to rebuild a ps2 game to make it look modern.

There is a long runway of human guided ai generation. Imagine if a single human was able to produce an entire animated TV show and all it took was good story telling skills without requiring 100 animators.

This is the right way to think about this IMO. Wouldn’t it be amazing if tools like this massively increased the scope of what is possible for a single artist?

Something I think about a lot is video game creation - what if AI tools let you quickly create 3D objects and a world without a big, dedicated team of artists?

(Check this out by Nvidia! https://m.youtube.com/watch?v=5j8I7V6blqM)

Humorously, the narration on that video also sounds like an AI. The rising inflection on nearly every clause almost sounds like they’re recording every sentence separately

It’s similar to how photography grants anyone the technical ability to render a photo-realistic image, but the majority of photos we see are easily discarded from memory. It did put painters out of the portrait business, but also freed them from a realist ideal.

I find it amazing and exciting - but then I don't earn my living as an illustrator.

I suspect there was similar wailings and gnashings of teeth when photography was invented. Humans adapt and it's us that decides what to value. We are quite fond of other people so tastes will alter to preserve that balance.

There was a clear career path for anyone out of a job due to photographs though. Become a photographer. Which was a perfectly fine career for a very long time. It still is, really, despite the high quality cameras available to many people, because it requires artistic and technical skill.

The career path for an artist trying to find something else while AI is developing at this pace is...less clear, and definitely scarier.

Photography didn't replace visual art though (which is just as well for painters and sculptors really, the skill needed to operate an early camera was pretty orthogonal). A subset doing the grunt work of simply recording details for prosperity had a competitive alternative, but even in the subset of image production where photorealism is considered desirable, people still commission painters for "lifelike" portraits and Canaletto renderings of Venice are still worth a fortune despite a near infinitude of similar images on Instagram.

Same goes for this really, but perhaps even more so. It's impressive at producing eye-catching photos quality images of simple concepts you might use in advertising campaigns, and can do a good job of reproducing well defined cartoon styles. But the artists working in those sort of roles are already working digitally and using image libraries, and this is a massive productivity boost for them - dozens of "$brand on beach" concept images for no effort at all, and they'll still need all the retouching and layering and colour balancing and combining with other images. And as soon as the brief gets more precise about details of layout and colour scheme or what they want the model to look like, they're either back to their original creative flow or designing text prompts and curating images for the brief is an advanced skill in itself. Non artists can produce "good enough" images too (looks great for memes), but much like non-artists wielding cameras they're mostly satisfying their own whims rather than competing with the business of art

Conceptual art remains a possibility.

The quality is not indistinguishable to any deliberate observation. Plus, I think people will begin to casually notice how off it is once they become ‘trained’ as to the quirks of these images. That is, unless the quality of the generated images continues to steadily improve.

Every time I think I have come to understand the limits of ML generated content, I see something a few months later that completely surpasses what I believed possible.

Creative industries are still safe from automation. This doesn't do anything except generate nice stock art or album covers. Unless you're in those industries, there's nothing to worry about. It can't even do text.

>It can't even do text.

An unpaid intern adding text to AI generated images would be trivial.

The only human-made images I got (2 out of 30) were also the only ones that were not square. Given that most GANs only do square images, human made ones should be square also if we want this to be fair. And as some others have said, DALL-E 2 really seems to have a "style" that's quite easy to spot once you've seen a few images already, so even without the square hint I think would have gotten close to 30/30. But I would like to see the results on a better crafted dataset with "naive" people, like a kind of visual Turing test.

Nice test but it would be great if it didn't wait forever to pass to the next image after answering.

Yes, the site is so slow it's unusable. Maybe the result of being on the front page of HN, but still; it could all work client side.

Indeed; all the responses are already in the JS file: https://thisimagedoesnotexist.com/static/js/main.dddea102.ch...

yub, I waited 6 forevers and then came here to write this.

Having already seen a few DALL-E generated images, there seemed to features I could pick up on that would allow me to identify them such as proportions, composition or unnecessary random detail. For example, in one image, there was a trail from what looked like a spaceship in the sky but the spaceship itself was obscured by a face in the foreground. It was hard to believe any human artist would choose that composition as it's quite jarring to look at.

I got 26/30, but only because I knew of DALL-E 2's "style", i.e. the robot ones were mostly drawings. if they gave a more "photorealistic" images, I could have been thrown off easily I think

Does it ever give human-created/real images? I got 30 artificial images.

me too. I assumed that was the "point", but turns out we can invent image-generating AIs but not a halfway decent random number generator.

The first 20 were all human for me. I was starting to think it was some sort of joke or commentary.

26/30. Since 90%+ of images seem to be have been generated by a machine, a better challenge would be to pick which image was generated by a human from a gallery of images.

That’s interesting, in my batch it seemed like 90% were from humans.

If you knew ahead of time that it was 50/50, then you could adjust your odds as you went on; i.e., on image 28, if there had been 14 CG ones and 13 human ones, you'd know a priori it had a 2/3 chance of being human.

It would be nice if they told you in more detail how the algorithm worked. How well curated are the AI and human images? It seems like selecting the absolute best AI ones and the worst human ones would make it easy to make things look better than they really are.

A genuinely "fair" way to do it I think would be this:

1. Randomly collect a few thousand images from DeviantArt

2. Ask people to write a short description of each image, similar to the prompts given to DALL-E

3. Generate DALL-E images from all the prompts

Randomly show a human or AI image with 50% probability.

That should give you a good measure of the actual state of the art.

I've tried the first time, then just restarted and voted machine every time. 30/30.

Are there any actual human-created images? I didn't stumble onto any. Maybe that's the point?

Yep I got 1/29 human created. Easier at that point to just guess robot every time which ruins the game

I got 30 AI-generated images as well, also wondered if that was the whole point.

Yes there are some. There were 3 in the set of 30 I was shown.

Do you mean that all 30 were AI generated or are you trying to make some point about what it means to be human-created?

I meant that the website kept telling me all the images I saw were computer-generated.

Yes I think it would be way better if both classes have a 50% chance.

If you flip a coin 30 times you won't get 15 heads and 15 tails. But getting 30 heads in a row would be one in a billion (literally); HN is popular but I doubt there's a billion people going to the site...

Getting any combination of flips is one in a billion.

HHHHHHHHHHHHHHHHHHHHHHHHHHHHHH

and

HHTHTHTTTTHTHTHTHTHTHHHTHTHTHTHTH

are equally as likely outcomes.

Right, but there is exactly one outcome with all heads, and there are at least millions of combinations with 50% heads and 50% tails.

To make it more obvious, let's take just 4 flips. There is exactly one combination with 4 heads:

HHHH

Whereas all the following will give you 2 heads and 2 tails:

HHTT HTHT HTTH THHT THTH TTHH

So you're 6x more likely to get 50% heads/tails than 100% heads. Scale that up to 30; the chances of getting 15 heads and 15 tails is not the most common thing but not unusual. The chances of getting all heads is quite rare -- if every human on earth did this test, we'd expect around 7 such instances total.

I think I had around 12 human-created images in my batch.

I got 11 human-created.

I found one.

Got 30/30 twice in a row (second time had maybe 4 repeats from the first game). Not sure what the fuss is, only tricky one was what looked like a child's painting but the tell was the handwriting being too good and it being a strange blend of crayon and paint that I don't think a child would have much patience for.

Overall most AI images are clearly identifiable by lack of fine detail in things like fingers or leaves and a telltail method of blending between colours in low detail areas.

If you spam the buttons you can get several points from one image and skip images.

Got me my highest score of 21!

To be utterly pedantic:

The image definitely does exist. I know they're trying to riff off of "this human does not exist" -- but that site works because the word "human" can mean a person in real life. I do not know of a usage of "image" to mean an artwork that must be painted by a human.

One thing I noticed is that the robot is not very good at composition. Awkward crops can be a bit of a giveaway. I suspect this could an artefact of the training process, where the images were cropped and shifted randomly (speculation).

Perhaps giving the "real" art a random crop would level the playing field slightly.

27/30. You can tell the AI ones because they don't look like they were made with the same constraints as a human; they merge, distort, twist together lines, shapes, and colors, rather than have images split into the compartmentalized pieces that humans like to do. I find the best place to check for AI art is to look at the overlaps; this is how I figured out the kindergartener one, for example: there were no obvious "AI-like" patterns except for the one section where the drawn lady's legs intersected with the ground line, and you could see that DALL-E 2 didn't know what to do, putting one leg over the line and one underneath, which can't be done unless one leg was drawn, then the ground, then the other, which is not what a human does.

I guessed 29 out of 30 correct. Just chose machine for all.

Perhaps have more human generated images, although that might be a limited set and people will just soon remember which images are human made. Impressive though.

Added a few dozens of images earlier today and add more tomorrow. Let me know how it goes.

I scored 29/30, only missing one because there were so many machine generated ones I started to just click the machine button repeatedly. Then I checked more closely and got all the rest right.

The machine generated ones are usually easy to spot if you look closely, because parts of the image are just nonsense (e.g. weird patterns, things out of place, odd focus). Machine generated images clearly have no actual understanding of the real world.

I got a ‘human made’ image wrong because areas of detail became a smeary pattern. The problem was very likely caused by computer-made compression artifacts, and I wouldn’t be surprised if similar generational AI becomes part of immediate post processing of cell phone snapshots.

It makes me wonder, are models being unintentionally trained to produce jpg artifacts? How much better would a generator be if it had been trained on, say, pngs?

25/30, with most of the errors being in the art styles.

Some key things that let me to immediate spotting:

- DALL-E 2 is really awful at symmetry, in art pieces it's less obvious because even humans are terrible at replicating features equally.

- A lot of human images are pictures from Unsplash, so they have some pretty recognizable features, in particular framing, shallow depth of field or more commonly, very noticeable ISO noise

Malformed anatomy is still a pretty good artifact of AI. Problem is some human artists aren't that great either, so easy to confuse with AI.

It’s usually that the AI messes up easy stuff while perfecting hard stuff. A human might struggle to draw a hand, but they don’t massively smudge and distort what should be straight lines. I saw dall e spit out basket balls where the black lines look like spaghetti.

Cherry-picked images, I guess.

Have you tried Dall-E 2? I've used a bunch of (less capable) systems and I don't think the cherry picking is that significant.

There's always a few duds and some definite limits (https://www.lesswrong.com/posts/uKp6tBFStnsvrot5t/what-dall-... )

but on the whole most of the output is very good - the images on this website aren't particularly taxing for it and I don't think cherry picking would be needed to replicate this level of quality.

My comment was not about cherry-picking for quality. It's about cherry-picking for not being able to tell the difference with human output.

I still largely stand by my point. But there are definitely subjects that Dall-E is worse at and would give the game away. I guess it depends what you mean by cherry-picking - it's certainly not the case that most Dall-E output would be easy to spot - more that it has some specific failure cases.

There was a post a few days/weeks ago on HN showing Dall-E and human images, and it was quite obvious which was which.

I think the current post really gives the wrong impression.

> There was a post a few days/weeks ago on HN showing Dall-E and human images, and it was quite obvious which was which.

Maybe it was cherry picked?

Not likely because there was quite a lot of work involved by humans. They demonstrated the output of AI and humans for the same prompt. The setup seemed more scientific than the current post (which seems more like a website trying to make a point).

I wonder how hard the jump to 3d models would be. It would be great if we had an AI thatyou can just prompt to generate a high quality 3d model. I know there are already Models that can create 3d models from a collection of images from different angles so I think it could probably be possible.

Gave it partial attention while in a meeting. Failed pathetically. One less excuse for meatbag existence.

The first picture that I saw [1] happened to be a building I used to work in. That gave the game away a bit.

[1] https://thisimagedoesnotexist.com/images/130.jpeg

28/30

Those 2 I got wrong were heavily stylized.

It's very easy to spot DALL-E artifacts, mostly geometric lines and shaped being distorted in a way that humans wouldn't draw, small details such as fingers and so on being deformed and so on. And it struggles with symmetry.

23/30. I thought I'd do better. Almost all of the machine-generated ones that I got make me feel like I'm crosseyed though. I wonder if other people get that. Also a lot of them make no damned sense...

Hi all, the author of ThisImageDoesnotExist here - thanks for all the great feedback! I've just added a few dozens of human-generated images and improved the randomization logic.

Looking forward to getting further feedback.

Hi,

Awesome project!

You might want to consider cropping all the human images to be square as well. It's a dead giveaway when all the AI images are square and the human ones have varying aspect ratios.

click multiple times on mobile, get free points until the tickmark loads in

DALL-E is a massively underused tool. This could transform the meme creation scene a lot. Instead of using Google images or DuckDuckGo or Yandex image search we can get DALL-E to dream up images.

I got 24/30, maybe DALL-E could work as an advanced captcha?

Unusable on mobile, the button behavior is all over the place

All of the "Does Not Exist": https://thisxdoesnotexist.com/

Not sure if it's because I'm on mobile, but the "fake" images were not easy to spot. I wonder how different platforms compare

This is awesome, but they should really crop every image to a square.

All the AI generated images are square which kind of ruined it for me.

Quite easy to tell the AI images still, especially those with eyes...the eyes are a dead giveaway on most AI images.

When will we be able to write our own prompts to generate new images? Will there be some sort of API for that?

I wonder if data from something like this could be used to improve the output quality of things like Dall E?

I got 5/7 before I stopped. I feel like this would be even easier if the images were higher resolution.

27/30

I struggled with some of the stylised paintings, but it's clear that DALL-E cannot do detail.

We're the AI generated images from scratch or they are edits to some human image ?

DALL-E 2 does support 'inpainting' and 'uncrop' (zoom out, synthesizing edges) functionality which can be used to edit AI or human-generated images. I don't think OP used any (at least wittingly): one of the ones I looked at in OP were ones I recalled as inpainted/uncropped, those are relatively unusual compared to pure generation, and it'd be a bit unfair to include any.

From the videos I have seen of it in use, there is no way the prompts had existing material. One prompt was “painting of a goat in the style of the Mona Lisa taking photos with an iPad”. And it spat out 10 images which showed exactly that.

The only human input is the text prompt.

We are all blade runners now

Who owns a copyright to these images? Can I just use them on my website?

https://labs.openai.com/policies/content-policy

https://unsplash.com/license

I don't think its legally established if these images can be copyrighted or not.

I scored 28/30. Wonder if it can influence a new type Turing test.

No idea what the UI expects of me. The clickable boxes have no text.

Hint: the robot has difficulty rendering reflections.

Would love to see this without cherrypicked images.

i loved this one

https://imgur.com/a/hF6R5Sx

The fact the images are curated make this assessment kinda BS. It’s way easier to identify synthetic art directly from DALLE

Human-created images are not curated?

Reply on news.ycombinator.com