We're all doomed: AI generated video is extremely realistic now

[ChrisFL]

We will be getting a ton of unintended consequences out of this

 

[Beezle2]

Utterly doomed. It is scary

 

[tony67]

Saw that yesterday myself - its pretty impressive how far its come in a year.
I had mentioned here recently how far AI Photos have come and many argued it was not all that great.

Anyway - yeah its pretty crazy. With 100 decent photos of a person you can do some very impressive stuff with photos...and a year from now it will be that much better. Even today the way people scroll photos on their phones they will never see the imperfections or even know they are commenting on an AI photo.

An influencer can create a model of themselves and put themselves in any city in the world and most wont even know its AI - the influencer is just sitting at home doing nothing and collecting the money.

Video, for the average user, is not that far off I'd say.
Already with just one photo you can do face replacement in a video, and its surprising good.

Luckily many of the companies are using AI to keep from creating things that are too explicit, but even with those efforts they are not doing that great a job and it wont be long before some company charges charges a little more with no restrictions.

That is where it gets concerning to me, creating models of a real everyday person, not just an AI generated person.

 

[mikehn]

Be careful who facetimes you

 

[T00dles]

Generative AI models produce patterns based on patterns they are trained on. Gen AI are essentially Galton Boards. They are data processors -- user inputs data (text, image, video, audio, etc.), which goes through the AI "peg board", and forms a pattern which the AI outputs. They are calculators for data forms beyond numbers. They are very much creator's tools.

As for deep-fakes...this is a human activity. Almost as soon as photography was invented, people began manipulating images. Double exposure fakes, supermarket tabloids...I mean, even the ghosts in the Haunted Mansion are created to make you see something that is not.

 

[MamaMermaid]

It's wild.

 

[_19disnA]

Most articles I read about AI are all over the map. They vary from the 'sky is falling' to 'no big deal'. I guess you can decide where you land on that subject.

 

[LuvOrlando]

So we can't trust anything we hear because people lie openly, audio can be altered. Can't even trust supposedly "credible sources" because they are under no obligation to be truthful and noone even goes to jail for making things up anymore, if a story or charges don't land it all just gets ghosted

We can't trust what we see because digital images can be easily altered.

What is left? Do we slide back to only believing in person experiences? Is real film more trustworthy? Things written on paper? Actual landline telephone conversations less likely to be altered?

How do we all counter how far the crazy has drifted? Don't we all already pay people to authenticate our information? Oh wait, not a line item in the multi trillion dollar budget

I just can't believe what our world is now, if the idea was a movie in the 80's we would have thought it too far fetched.

 

[BrianL]

Most articles I read about AI are over the map. They vary from the 'sky is falling' to 'no big deal'. I guess you can decide where you land on that subject.

I have been assured by someone at Cyberdyne Systems that it is no big deal. They were quite adamant about that.

 

[mikehn]

Skynet has come a long way

 

[ChrisFL]

So we can't trust anything we hear because people lie openly, audio can be altered. Can't even trust supposedly "credible sources" because they are under no obligation to be truthful and noone even goes to jail for making things up anymore, if a story or charges don't land it all just gets ghosted

We can't trust what we see because digital images can be easily altered.

What is left? Do we slide back to only believing in person experiences? Is real film more trustworthy? Things written on paper? Actual landline telephone conversations less likely to be altered?

How do we all counter how far the crazy has drifted? Don't we all already pay people to authenticate our information? Oh wait, not a line item in the multi trillion dollar budget

I just can't believe what our world is now, if the idea was a movie in the 80's we would have thought it too far fetched.

Right, and while most of this stuff is relatively harmless, when I scroll through facebook (which I still use because some of my friends/relatives are only on that platform), there's already TONS of AI generated images with commentors who have no clue at all. I think FB and other social media platforms should make it a requirement to notate something is A.I. and allow us to report if there's evidence that it is...but that is becoming more and more difficult to distinguish.

 

[Boopuff]

Skynet has come a long way

Yes!! Didn't anyone pay attention to Terminator? This is how it starts..... LOL

 

[MillauFr]

Doomed? Says who? The same people who said Y2K was going to plunge us back to the stone age?

 

[ChrisFL]

Doomed? Says who? The same people who said Y2K was going to plunge us back to the stone age?

Not sure how that relates at all...completely different situations. This is going to give everyone a tool with zero actual skill or time required to create fake videos for all kinds of nefarious purposes...you can't close pandora's box again.

 

[tony67]

Not sure how that relates at all...completely different situations. This is going to give everyone a tool with zero actual skill or time required to create fake videos for all kinds of nefarious purposes...you can't close pandora's box again.

Yeah this is what I see as the real issue.
Using it for deep fakes of famous folks and politicains...that will be less of an issue..and they will sort it out because it affects them.

The real issue is the everyday person being a target.

Take a look at seaart.ai

They have a face swap tool
You can simply upload a video and an image and it will replace the face and the results are pretty good, still not perfect.
I had used one of the boyband samples they had and replaced the face with Iger, it took 5 minutes at most to process.
Unfortunately I cant upload a video here.

Thing is you don't really need any sort of skill for that. For now it's mostly harmless (Don't panic ).

Using LORA models and generating images of a specific person still takes a bit of effort, especially to get it to be perfect.
It's not simple for the average user, but it's getting better an better every month.
With a service like seaart you don't need any kind of hardware - its all on their backend- you can use a phone.
Hands are still iffy, but again improving.

It would not take a lot of effort for someone with the motivation of revenge to get a bunch of photos of someone, generate explicit images, and share them to their work or post them on adult site and share to their work.
You could even create the background of their specific office.

Yes its this type of stuff has always gone on with photography, but it was not point and click that anyone could do and the output usually had flaws on deep inspection.
As this gets better and better, and easier, finding those flaws will get tough. Especially if its just some rando in an office trying to explain to HR its a fake image going around the office or a site they have nothing to do with.

 

[Neapolitan Ice Cream]

AI worries me. As a study aid it's great, but it errs on the side of infallibility all too often. I have been trying to catch mine out and so far it has only ONE blind spot - trying to spell words in pictures correctly!

 

[tony67]

AI worries me. As a study aid it's great, but it errs on the side of infallibility all too often. I have been trying to catch mine out and so far it has only ONE blind spot - trying to spell words in pictures correctly!

True - I think some of that is on purpose, but never looked into it. They use odd characters instead of actual words.

There are Lora's for making that better, and again if you use your own photos for the background I don't think that's an issue.

Regardless, it's just going to get better and better.

 

[ChrisFL]

I wanted to add the other side of this coin.....now that this is as realistic as it is, basically any REAL video we see can be challenged as being "fake" by others. I don't know how we get out of this, other than developing tools that can identify real vs generated videos.....a very challenging thing.

 

[Brett Wyman]

If thats what they show to the public imagine what they have behind closed doors!

 

[Beezle2]

Most articles I read about AI are all over the map. They vary from the 'sky is falling' to 'no big deal'. I guess you can decide where you land on that subject.

I am at least 1 Susan Collins of concerned. It will be interesting