cross-posted from: https://programming.dev/post/3974080
Hey everyone. I made a casual survey to see if people can tell the difference between human-made and AI generated art. Any responses would be appreciated, I’m curious to see how accurately people can tell the difference (especially those familiar with AI image generation)
The scenery ones are near impossible to tell because you have no context. The pencil art one is impressive though.
I got all the landscape ones correct—except for one—by applying my limited knowledge of art technique.
I was missing a “don’t know”/“can’t determine” option.
For photographs specifically and some types of paintings/artificial stuff, there are things you can look for. But for other things, I feel like, or at least to my knowledge, you can’t.
Like the pencil drawing. There’s not enough things it could be doing wrong. It’s a sketch. With simplistic but “error-excusing”/diffuse/transformable content.
The goal isn’t really to be a quiz, but rather just to see how susceptible people are to AI generated art. Many of the images I chose are intentionally vague, 80% of people so far got the line art sketch wrong, and that’s with knowing that many of these are AI generated. The results are definitely interesting to see.
A “don’t know” option would ruin the point since most people would just choose that. I want to see where people lean towards.
The line art one has a stockphoto watermark still visible in the bottom right corner by the shoe (subtle white cross pattern). That’s the only thing that gave it away as being real.
The back left leg of the bench in the pencil drawing is in the wrong place - at least that was what I considered the ‘tell’.
But I found it really hard to spot the AI.
13/20, I work in AI. The paintings were the hardest for me, because the art style obfuscates some of the AI artefacts that can be tells.
I basically came here to say this. The paintings are hard but real life photos are easier.
Yeh, I don’t work in AI but got 12 because the art was difficult. It’s still a while off until it becomes impossible to tell.
8/20. I am pretty good on photorealistic images, but the random drawings… honestly a lot of the ones by people I tagged as AI generated because i thought they kinda sucked.
Got 7/20, the second photo really did surprise me!
I got 17 out of 20. I pegged the bezerk drawing as generated because the bottom part of the armor lacked symmetry and didn’t make any sense. I got the other three line drawings incorrect.
I have spent WAAAAY to much of my freetime generating images and apparently have picked up an eye for the weird types of artifacts that these generators produce. The hardest one to articulate is that generated images have a very specific type of noise. Images create a very nice grainy type noise while digital images get more of the blocky jpeg artifacts and banding. Generated images get this weird hybrid of the two that isn’t consistent across the whole image.
10/20.
You can only fool me half the time! Joke’s on you, robots!
Got 10/20. The second photo really threw me for a loop. All the texture on the skin and and hair led me to believe human; I noticed the weird patch on the shoulder and the unnatural shine on the ear but excused it as technical flaws or something, chose human in the end. I really thought that corporate logo style drawing of the avocado was human, like it wasn’t even a question for me and yet the fact that it was AI really surprised me.
I also got 10/20. The second one is fairly obvious, though, in my opinion. Look at the shape of the glasses – the lenses are uneven and don’t match.
10/20 and most of it was just guess work. It has become pretty much impossible to tell when you pick real images that are in the style of AI images, you really have to go pixel hunting to find artifacts and even that is getting difficult, when things get blurry or you get JPEG artifacts when you zoom in.
That said, there are still plenty of images AI has a very hard time creating. Anything involving real world products will always look wonky in AI, interesting framing where you don’t have one object in the center is hard. Unusual aspect ratio are hard, as AI is trained on squares. Complex scenes with multiple characters rarely work. Facial expressions are hard. And generally just normal everyday photos don’t really work, AI stuff always looks like people posing for a stock photo.
9/20 rip
14/20, huh. i seem to have learned a lot from all of the ai generated pics on rule34.
[Survey] Can you tell which Surveys are AI generated??
deleted by creator
Regardless of score bragging, it requires some technical knowledge and pixel peeping to really be able to tell, and even then I can’t guarantee you can. I would imagine your average Joe wouldn’t even know any better.
Average Joe here, 7/20 from just guessing
The avocado had real text. Is Dall-E 3 capable of creating legible text?
Yes, it’s the only model that manages to get text right, and the results are usually pretty consistent. It’s a big step forward.
Kind of. It can generate readable text, but not all the time. It will frequently turn parts of your prompt into text that aren’t meant to be text or mix perfectly readable text with AI gibberish:
It is, yeah
