Social Question

Dutchess_III's avatar

Is there any electronic way to see whether a picture is AI generated?

Asked by Dutchess_III (47126points) 1 month ago

Other than a sneaky suspicious. Some tag or something?

Observing members: 0 Composing members: 0

16 Answers

ragingloli's avatar

You can usually tell by looking at the background. Unreadable text, people without faces. Objects blending into each other.
And the there is the general impression. AI generated pictures are usually too smooth. The colours too saturated. Too much contrast.

gorillapaws's avatar

Not in the way you’re talking about. Ironically I think they’re training AI models to detect AI images from real ones.

elbanditoroso's avatar

I read recently (this past week) that ethical image generators are embedding certain machine-identifiable tokens into the image that can’t be seen by people, but that are ‘visible’ to image processors.

jca2's avatar

Sometimes the shadows make no sense. I see room photos that are AI generated that have lots of plants and it’s just not realistic.

Think also of Princess Kate’s photo in the spring with her children. The hands weren’t right.

Blackwater_Park's avatar

Ultimately, there is no sure fire way unless one of the above defects or telltales are present.

Dutchess_III's avatar

I just get suspicious of oh so perfect pictures of nature. There should be a law. It’s false and misleading.

Tropical_Willie's avatar

Sometimes the people will have 6 fingers, different side of clothing is different textures and shading.

Call_Me_Jay's avatar

A little googling shows a few sites claiming the ability to recognize AI.

1)
Here’s one that looks experimental and not like advertising bait.
https://huggingface.co/spaces/umm-maybe/AI-image-detector

2)
Here’s a legit looking product, a tool for sale for web sites. They tout it for detecting insurance fraud, fake news, fake IDs, marketplace scams, etc.
https://sightengine.com/detect-ai-generated-images

3)
Also the googling shows a lot of research, but it seems like much is devoted to deliberately tagging content at creation with hidden “this is AI” info.

Google searches will now detect origin of AI-manipulated images

jca2's avatar

@Dutchess_III the perfect photos of nature are often photoshopped or other photo edit programs, where the colors are so bright they’re almost fake, the skies look like they’re painted, the trees are so bright and vivid. I hate that, because everyone will comment that it’s beautiful and “Wow!” but meanwhile, it’s not real, it’s photoshop.

Dutchess_III's avatar

@call_Me_Jay your second link is close to perfect! Thanks.

Dutchess_III's avatar

@jca2 I know. That’s why I said I was looking for a definitive way to tell.and not just rely.on my suspicions.

RocketGuy's avatar

Fingers and text are big giveaways. Sometimes weirdly long arms and extra legs can be seen.

Dutchess_III's avatar

Well, forget people. What about nature scenes?

snowberry's avatar

Bright green leaves (the kind that should be dead) in a winter scene.

ragingloli's avatar

What could be indications from the tests I have run:
– lack of imperfections or debris, like no twigs or fallen leaves on the floor
– grass is too uniform in colour
– grass looks too soft/airbrushed
– patterns in trees that look somewhat repetitive
– leaves/branches that seem disconnected from the main plant

Blackwater_Park's avatar

All the little things that give A.I. away now won’t be discernable much longer as the tech improves. There won’t be a way in a year or two.

Answer this question

Login

or

Join

to answer.
Your answer will be saved while you login or join.

Have a question? Ask Fluther!

What do you know more about?
or
Knowledge Networking @ Fluther