That article made me realize how insane Image Generators already are.
(www.theverge.com)
from laserkaspar@lemmy.world to technology@lemmy.world on 25 Aug 2024 18:34
https://lemmy.world/post/19046964
from laserkaspar@lemmy.world to technology@lemmy.world on 25 Aug 2024 18:34
https://lemmy.world/post/19046964
Really good article I read today. I was already impressed by the first hype of image generators but since haven’t informed myself much. They got really good lately apparently. I can’t decide if I am concerned or impressed. Do you think this would actually be used for something other than memes and misinformation? I thought I might share it and hear your opinions.
threaded - newest
I can see a few useful use-cases, mainly deleting unwanted stuff / people from a photo.
That’s a very good point! I never thought about the power of removing things. Would be nice to see it save photos that are out of focus, in a bad perspective or with a finger over half of the selfie.
I like using the magic eraser for photos of my daughter and being able to remove other kids and send it to the rest of the family or frame it.
This feature already exists, it’s called Magic Eraser and it’s on most Pixel phones. I feel this is something different and needs to be considered differently.
Yeah, it can do other stuff, I was simply stating a use-case I consider valid. Doesn’t matter that others have come up with the feature before. This is presumably better at detecting the object and removing / replacing it.
it’s the same technology in the background afaik
I find the author’s reasoning strained. They use flat Earthers as an example of the power of photographs to “prove” reality?
I question the central premise that photographs were ever the foundation of reality. Haven’t filmmakers been fooling us with photography for over a hundred years?
Yeah, clearly the author doesn’t know about Stalin
From the article you clearly didn’t read:
k
Yeah, clearly the author doesn’t know about the Loch Ness monster
Did you read the article or comment on the title
i think its more about the ease.
you used to need a team, then a qualified professional, now any moron can tell a machine to do it.
That all comes up in the article. The core idea the author is getting at is the general ease of fabricated situations is coming in a new way that previously hasn’t been a couple clicks for the average user. Think less about political turmoil (propaganda has existed as long as there as been politics) and more about how your Karen aunt can add a worm to their Google review for spaghetti. Most people won’t learn Photoshop, most people can click a few buttons.
I think it’s still important to consider the tomorrow we’re being thrust into even if we could do this on a smaller scale yesterday.
Yeah, we finally start to get accountability from public officials via bodycam and now here comes technology that will make it trivial to skew the narrative
It’s still rather easy to identify AI generated pictures, especially of people. There’s still way to go untill we get video that’s sufficiently good that it’s difficult to tell it’s fake. It’s absolutely going to be a problem sooner or later but I doubt we’re anywhere near.
Also, the one benefit this all comes with is the plausible deniability when you’re accused of something even if it really was you. Say you have nudes leak online for example. You can just say they’re not real and it would be really difficult to prove otherwise.
What, AI image generation? Sure. Come visit at !imageai@sh.itjust.works. Or look at civitai.com for a large collection of people working with them.
No thanks.
Before photographs or audio recordings to prove something, I understand that it was common to have trusted individuals or entities attest to something.
That’s a technology that we haven’t used for a while, but I imagine that we could dust it off and use it for many of the same roles as we do recordings. Could probably have services that just deal with that.
Testimony has always existed in the court room, it hasn’t gone away. The relative difficulty of believable photo manipulation to add in new subjects, has up until now been out of reach for the average person so generally speaking it’s been a nice enhancement of authenticity to testimony.
I think chain of custody will be more important than eyewitness testimony in the future. In the case of CCTV or body cam footage, it will be a combination of the footage being digitally signed by the recording device, and being able to prove the footage could not have been tampered with along the way.
So, a record of where and when the footage was recorded, when it was transferred from the device, where it was held after this, and that the storage location was also secure.
fyi that is a photo by Dorothea Lange not a photo of Dorothea Lange
This doesn’t even begin to scratch the surface of what you can do with tools like ComfyUI and stable diffusion. It’s sad to see how all the hate around AI image generation prevents people from being educated about it, which in the end is beneficial to bad faith actors using these tools to create misinformation. If people weren’t afraid to learn how it works they wouldn’t be fooled as easily by it. The days of 8 finger hands are over.
Have you seen flux.1? It blows stable diffusion out the water and it just came out, boomers and millennials alike might be unable to confirm anything anymore