We’re cooked.
Google’s AI model is getting really good at spoofing phone photos
https://www.theverge.com/report/837971/google-nano-banana-pro-realistic-phone-photos
https://www.theverge.com/report/837971/google-nano-banana-pro-realistic-phone-photos
We’re cooked.
12 Comments
stoy@lemmy.zip · 17 pts · 252d
So how long untill digital photos are inadmissble in court?
Longylonglong@sh.itjust.works · 12 pts · 252d
I mean of course it is. It got basically infinite training data
calliope@retrolemmy.com · 8 pts · 252d
Seems like very thinly veiled advertising for a new version of google’s ai image generation.
“We are cooked, fellow kids!”
The author also pretty much says “all other AI was slop before this, right guys?”
Hackworth@piefed.ca · 4 pts · 252d
Yeah, a more honest take would discuss the strengths & weakness of the model. Flux is still better at text than Nano Banana, for instance. There's no "one model to rule them all," as much as tech journalism seems to want to write like that.
Hackworth@piefed.ca · 6 pts · 252d
Nano Banana Pro's built into Photoshop now, as is Flux Kontext pro.
bonenode@piefed.social · 4 pts · 252d
Sinaf@lemmy.world · 3 pts · 252d
Most of these images have really shitty resolution as well. Can't they generate higher res stuff or would inconsistencies otherwise be more obvious?
Hackworth@piefed.ca · 2 pts · 252d
Directly, generating higher res stuff requires way more compute. But there are plenty of AI upscalers out there, some better, some worse. These are also built into Photoshop now. The difference between an AI image that is easy to spot and hard to spot is using good models. The difference between an AI image that is hard to spot and nearly impossible to spot is another 20 min of work in post.
Sinaf@lemmy.world · 1 pts · 252d
Yeah, I don't doubt it, but you still need human labour. Not anyone can simply fake a photo on a level that's believable.
TheWonderfool@lemmy.world · 1 pts · 249d
I would say that it mostly depends on the complexity of the photo. Random Instagram model posing for the camera? You can get it out of the box with a press of a button in your machine. Complex photo with multiple subjects and cluttered background? That would need lots of human work.
Sinaf@lemmy.world · 1 pts · 248d
I am definitely impressed by what it can do, but what real world demand does it fulfil? You know, apart from misinformation campaigns and the like. There may be some use cases, but enough to justify the huge investments?
TheWonderfool@lemmy.world · 1 pts · 246d
Porn, advertising, entertainment. The first one seems to be mostly ignored for now by western model training companies (though it must be difficult to do when your paying processor can destroy your business overnight). For the other two, I doubt the market is so small that it can be ignored.
Prepare in a few years (months?) to see movies advertised as no AI (like they are now advertised as no CGI, as ridiculous as it is).
1984@lemmy.today · 2 pts · 251d
Nothing will feel real, and we will own nothing.