Algorithm allows video editors to modify talking-head videos as if editing text

https://news.stanford.edu/2019/06/05/edit-video-editing-text/

by jonbaer • 7 years ago

558 201 7 years ago

news.stanford.edu

Just imagine people using this tool to spread false claims and propaganda. Can we also determine if the video was actually edited or forged?

quocble• 7 years ago

My thoughts exactly.

Datenstrom• 7 years ago

Just imagine people using this tool to spread false claims and propaganda.

That was done in The Expanse novels, I wonder if that is where people get the ideas to create things like this from.

Can we also determine if the video was actually edited or forged?

Detecting fake videos is actively being researched and there are working methods available, governments are very much aware of the danger.

mathieuh• 7 years ago

There will still be some lag between initial release and govt response. C.f. The War of the Worlds broadcast

DonHopkins• 7 years ago

Running m-x dissociated-press on a Trump speech makes it still sound like a Trump speech.

https://en.wikipedia.org/wiki/Dissociated_press

ceejayoz• 7 years ago

It'll be used to debunk true claims, too.

RandallBrown2• 7 years ago

Their mouth movement wasn't particularly natural in the parts where the speech was edited.

It's good enough that it will fool some people on Facebook/Twitter, but it's pretty far from being able to stand up to any scrutiny.

SketchySeaBeast• 7 years ago

Isn't that all it takes to spread propaganda? We're living in an age where if it feels right, it is, even if it's blatantly false.

latexr• 7 years ago

I agree that their mouth movement wasn’t particularly natural, but I disagree that it isn’t enough to fool most of us if we’re unaware. We’re not fixing our eyes on a person’s mouth while they talk.

Think of the scenario where someone edits a few seconds of a 30-minute interview. They make the interviewee go from saying they hate drugs to saying they love drugs. Even if you weren’t expecting that claim from that person, would you go back and recheck their mouth movement, to be certain if it was edited or not? Unlikely. Even if you would, I’d wager most wouldn’t, including most of us that could detect it.

underwater3• 7 years ago

The tech is only going to get better, faster, and easier to use. And that will happen incredibly quickly. Being able to spot artefacts in the output is not a viable defence.

r00fus• 7 years ago

How do we know it's not already in the wild? Ethical release of this would focus also on some way for easier forensic telltales that are hard to avoid.

Also, rule34 implications.

iamnothere• 7 years ago

GANs can be used to detect this sort of thing if trained well enough. Any manipulation tool is going to leave some kind of signature. Eventually you'll end up in an algorithmic arms race, and you'll be able to fool some people, some of the time. (Pretty much the same as things have been since the Spanish-American War.)

Smaller economies are going to get screwed by this, though, as they won't have the resources to fight yet another technological battle.

jalgos_eminator4• 7 years ago

You know, I thought that deepfake videos would be politically weaponized when I first heard about them. However, after doing more thinking on this, we have had photoshop for 30 years already! We see photoshopped images all the time and while some people can be fooled, many others remain skeptical of an image then try to verify it hasn't been altered. I don't think photoshopping has really been a big problem yet, which makes me think that deepfakes won't be one either because it is fundamentally the same kind of deception but in video form.

loceng• 7 years ago

Audio-video is much more stimulating and convincing because it mimics real life more than still images.

cwkoss2• 7 years ago

I feel like we just need some comedians to make a bunch of entertaining, realistic but labelled-fake content with deepfakes. Production cost of making a show where each actor is deepfaked into a world leader or dead historical figure or whatnot is not a significant hurdle.

jalgos_eminator4• 7 years ago

Yes, I realize that and photoshopping got a bunch of press for being used on modelling pictures to "enhance" the models before putting them in magazines. These deepfakes will probably be used to do some other similar things as well. My point was that deepfakes aren't anything new and we already have the tools to analyze them. Those tools just aren't computer programs, they are people posting videos on youtube going pixel-by-pixel to show how a certain photo was doctored. After all, a video is just a series of photographs.

I think writing computer programs designed to spot these deepfake videos would be very helpful as the volume of doctored videos increases, but this isn't some disruptive technology (at least for people trying to deceive others).

mamon• 7 years ago

This scepticism is itself a problem. There's a whole branch of philosophy that claims that the objective truth is impossible to know. With deepfakes that's even more true, and might drive a lot of people into despair and apathy.

hanniabu2• 7 years ago

Exactly, I have shown my friends media of their favorite politician contradicting himself and their response is fake news

lopmotr4• 7 years ago

We'll still have provenance and trusted organizations which is what we mainly use to verify important and easily faked things like written words, photos, and videos that might be taken from a different context to what their descriptions claim. There are other techniques people have developed to verify things too, like the group recitals that transmitted the old testament orally for multiple generations. You couldn't just edit it and repeat a fake version because the change would have conflicted with the consensus.

Society survived before videos and photos, when all information was easily edited. I think we'll be fine. Maybe we're just in a brief decade or two where we became complacent at believing all videos were real without taking any of the care that we used to take with grainy films of alien autopsies or spoken testimonies of people who claimed to have seen bigfoot.

umvi• 7 years ago

Yeah the bigger problem is taking video footage of a politician, and then going frame by frame to find the most unflattering possible depiction of them (usually right after a cough or a sneeze) so you can use it to "support" your trash click bait headline. No deep fakes needed - you can make anyone look like a raving lunatic if you take the frame right before a sneeze.

peacelilly• 7 years ago

The problem with deep fakes is a hostile nation taking over your phone/facetime calls and sounding exactly like your own parents. How can you tell if they can do it in real time? A bad actor could get you to do some really bad things.

mtgx• 7 years ago

I remember there was some oil companies-backed anti-Tesla propaganda image a while ago showing the "environmental disaster a lithium mine creates," which went viral for a bit. It was, I think, a tar sands mine.

There's no way deepfake videos won't make the propaganda situation worse, at least for a while.

VikingCoder3• 7 years ago

All of these things make it easier to mass-produce bullshit at low cost.

I'm pretty sure I know people who have been convinced by meme quotes. A headshot of a politician they don't like, with text overlaid, which they never said. People are outraged! And never bother to inspect the source.

Anything that makes it easier to lie about what someone said or did, or makes it harder to disprove... They're all politically weaponized, already.

Look at the "drunk pelosi" video.

babypuncher• 7 years ago

The thing is deepfakes don't really make this easier. They are a lot of work to produce and ultimately aren't a whole lot more effective than a good old fashioned photoshop or crappy meme.

AJ007• 7 years ago

Politicians say outrageous things all of the time. The ones that don’t, usually have trouble raising money much less winning elections.

mistermann5• 7 years ago

I'm pretty sure I know people who have been convinced by meme quotes. A headshot of a politician they don't like, with text overlaid, which they never said. People are outraged! And never bother to inspect the source.

I really wonder which type of meme has the most influence on average, straightforward and outright lies like the one you've noted, or the more subtle, subversive social commentary style. I'm a big fan of the latter, I think they're very interesting and underappreciated.

For example, this one - nothing more than a simple screenshot of Twitter, but to me this seems very persuasive: http://magaimg.net/img/80rb.jpg

Pointing out hypocritical grandstanding: https://i.redd.it/74n4exoy2g131.jpg

Media bias: https://i.redd.it/gtrq4xgemtx21.jpg

If Biden runs will be interesting to see how much airplay this meme will get: https://i.imgur.com/tqzGS6E.mp4

Laughing at the silliness of popular narratives: https://i.redd.it/a9wrex3ijc231.jpg

Just for a laugh: https://i.redd.it/i5bvml5m92131.jpg

Laughing at logical inconsistency: https://i.redd.it/rp9ydf0s88y21.jpg

An interesting way of looking at Brexit: https://i.redd.it/cs6o72p3dp031.png

Historical hypocrisy on border control: https://i.redd.it/983sm9wjley21.png

All of these are from t_d so obviously one-sided, I'm sure a similarly impressive collection from the other perspective could easily be assembled, and it's not that uncommon to encounter otherwise intelligent people who have obviously had their beliefs shaped by those memes.

fspeech• 7 years ago

This may actually make brand names in media valuable again.

ehsankia• 7 years ago

Isn't the drunk pelosi video literally just slowed down audio? If anything, it proves that we don't really need these insanely advanced machine learning techniques to make bullshit. Something as basic as slowing down a video 20% will do just fine in misleading people.

TravelAndFood• 7 years ago

Do these transformations leave any discrepancy or signature in the video or audio that would be detectable by a machine? (So, tiny, tiny discrepancies might work.) Someone could make a browser plugin to alert the user when video/audio has a good chance of being fake.

newsbinator• 7 years ago

The best way to fool someone isn't to do an indistinguishable Photoshop job. It's to do a passable-enough fake of something the person wanted to believe anyway.

I see this technology as no different.

contravariant• 7 years ago

Although in most cases doctoring an image isn't even necessary, you just need to put it in a misleading context.

mortenjorck• 7 years ago

We’ve had photo manipulation without the need of a darkroom or skilled optical retouchers for 30 years, but weaponized photo manipulation has been a thing since Stalin’s censors airbrushed out Trotsky a century ago.

jalgos_eminator4• 7 years ago

Yes, this is true. However, good quality photo editing became relatively cheap and convincing in more recent years. So now you don't need to be a nation-state to be able to convincingly pull something like that off. You can just be some guy in a basement with $1000 worth of computing gear and experience using photoshop.

1PlayerOne• 7 years ago

I agree, it was fake news articles on Facebook that spread false information but it took desperately ignorant people to believe it for the consequent chaos to ensue. So while deepfake videos are scary, I think what we have to really worry about is the deep ignorance of the voters.

maxxxxx2• 7 years ago

From my observation I believe that a lot of educated and less educated people don’t really care if something is fake as long it confirms their opinions. They don’t want to let go of their “facts” although they know they are wrong.

dylan604• 7 years ago

Isn't the deep ignorance of the voters the entire point of a deep fake? It's knowingly a fake, so by creating it you are already trying to pull one over on someone. When the president of the US can say "I never said that" even with video/audio evidence of him saying it while also screaming about deep fakes, the slippery slope is being greased. It's not hard to believe that viewers that only get information from a single source will fall for it. Even if they hear arguments it is fake, they will not research it on their own because their single source is never wrong.

soulofmischief• 7 years ago

The problem is that just like "fake news", deep fakes will provide alibis for people caught doing real crimes.

And their fanbases will listen because it's easier than accepting their idol could be a bad actor.

el_benhameen• 7 years ago

I don't think that fakes (photoshop, deepfakes, whatever) need to be absolutely believable to be effective. The long-game purpose is to erode trust in institutions, media, politicians, etc. Fakes accomplish this goal by being just believable and just frequent enough that more and more people start deciding to believe whatever it is they want to believe because "who knows what the real truth is!".

crazygringo• 7 years ago

Totally agreed.

Writers have been able to write nonsense for a long time... and photo manipulation we've gotten quite used to. All we do is add video to the category of things that might be lies, and so need independent verification.

Skepticism is good and healthy, and verification in the age of Google isn't that hard.

You can trust that if the NY Times or CBS publishes a video, they verified its authenticity, or else will be publishing a big retraction within a few days that will also make the news because it's so rare.

Whether your uncle sends you a random photo or a video of a politician that seems too exaggerated or weird or unbelievable... you assume it might be manipulated... as you already do now. Making Nancy Pelosi seem drunk didn't take a deepfake, just slowing it down.

It's not any kind of big change. Just applying the same skepticism we already automatically apply to so many other things.

HeWhoLurksLate2• 7 years ago

You can trust that if the NY Times or CBS publishes a video, they verified its authenticity, or else will be publishing a big retraction within a few days that will also make the news because it's so rare.

This may be, or become false, due to political motivation to seriously damage the "other side of the aisle".

oefrha• 7 years ago

In fact, often times you don’t even need to lie to skew the “truth”. Cherry picking facts or even just highlighting certain facts over others, plus an optional bit of extrapolation or subtle misinterpretation, is often enough to fit whatever narrative you want to push.

and verification in the age of Google isn't that hard.

It’s hard because publications often parrot each other. You walk away confident of your “verified” truth due to echo chamber effect, which might be worse than not verifying at all.

You can trust that if the NY Times or CBS publishes a video...

I can’t. Again, you don’t need to make factual mistakes to push an agenda.

siidooloo• 7 years ago

Mostly people seem to be sceptical of things they already don’t believe. If someone repeats something you believe, how much research are you going to do. So photoshopped images that reinforce your beliefs slip by, and the one that challenge you, you catch. Or worse, the ones that challenge you get labeled as photoshopped regardless of their provenance.

nihonde• 7 years ago

We’ve come to learn that images and audio can’t be trusted. It was inevitable that video would fall as a trusted source.

Next up: the person you’re sitting across from at dinner.

Izkata• 7 years ago

Some people already experience this: https://en.wikipedia.org/wiki/Capgras_delusion

fiblye• 7 years ago

Photoshopping is done by hand and generally has mistakes. Good photoshops are still believed.

Videos are being altered by machines. They’re being optimized for natural looking results. It’s harder to notice small mistakes when frames are going by at 24FPS vs poring over a static image for 30 seconds until you finally notice the one region with mismatched shadows or odd clipping.

tuesdayrain2• 7 years ago

I've literally never been able to identify a photoshopped image (except for immediately obvious work), without someone first pointing it out. I have a feeling that the percentage of the population that can spot edited images is in the single digits.

naniwaduni2• 7 years ago

On the bright side, video gives a lot more opportunities to make small mistakes for conspiracy theorists, er, fact-checkers to find!

kzzzznot• 7 years ago

Yes but you can pore over each frame if you’re taking time to study as you are suggesting someone is for the photo in this situation

empath75• 7 years ago

Not being able to believe realistic looking video is in itself a problem. People are simply not going to believe news anymore.

dsfyu404ed• 7 years ago

God forbid media outlets are forced to rely on reputation.

maxxxxx2• 7 years ago

I think the problem is that if we can’t trust video then there is really nothing visual left we can trust. Until now you could at least trust video recordings to some degree. Not sure if that a good thing or not.

ruytlm• 7 years ago

There are also many useful tools for the skeptic with regard to edited images: http://fotoforensics.com/

It will be interesting to see if such tools are developed for video as well.

Paul-ish• 7 years ago

Doesn't a lot of the "fake news" use photoshop?

pergadad• 7 years ago

The REAL issue here will not be the fake videos themselves. They will cause many messes, but the real issue is an acceleration of what we see today: a loss of trust in information and in particular the established media. More societal rift, easy to dismiss any negative news about your favourite politician/rapper/.. as fake video; more difficult court cases even where there's video evidence, ..... Terrifying.

creaghpatr2• 7 years ago

a loss of trust in information and in particular the established media.

Yeah, the trust is not coming back, regardless of fake videos.

ACow_Adonis• 7 years ago

You don't think that photoshopping has really been a problem and you think people remain skeptical?

I guess if you think that epidemics regarding images of male/female body image, body dismorphia, self-harm, anxiety, and using celebrities to sell products aren't connected to it, but I've found exactly the opposite.

I love photography and I am utterly unable to talk to non-photographers or convince them about what happens in the production of most images they see in most forms of commercial media.

It goes something like this:

"Hey ACowAdonis, how much of that photo do you think was retouched?"

looks at photo

"All of it".

"All of it? What do you mean?"

"I mean all of it."

"But that's Reese Witherspoon! (or insert popular celebrity here)"

"Yep, and you can see how her eyes have been adjusted, her skins been adjusted, they've changed the shape of her arm, taken a few pounds off the mid section, increased the boob size, changed the colour of her hair...and i'm pretty sure that's not her hand".

"Nah, you crazy..."

"You want crazy...pretty much every photo in every fashion magazine and every media item involving that celebrity has been adjusted to a similar extent"

"Nah mate, you're having me on. You're nuts."

bryanrasmussen• 7 years ago

Here's how it will work, someone will make a deepfake of a political opponent and then publish it on a forum where like minded people gather using a dummy account.

Other dummy accounts will take the deepfake and start making a narrative around it, sending chain emails to their real world contacts.

Real world contacts will start passing around deepfake chain mail they were sent.

Some of these emails will take the deepfake as true, some will talk of it as being a funny parody, but "funny because it's true" anyway.

Major news organizations can now address the issue as news because people are passing it around, maybe it will be something like 'Well Bob, I think the X have a real image problem on their hand, if the video is true or not..' You don't mean to say you think it's true!? "I didn't say that Bob, I'm frankly not qualified to judge and I haven't done any research what I'm worried about here is that there is a perception that it is true or if it is not exactly true in this particular instance that it might be true, and that is what I mean by a real image problem"

mk89• 7 years ago

Video is way worse, because it's connected to audio. Which means "did you hear that? He just said he wants to go to war...".

adrianN• 7 years ago

People know about photo manipulations and get suspicious because we've had Photoshop for thirty years (and analog photomanipulation even longer) and see them all the time. This wasn't always true. When photography was new, manipulations that wouldn't fool anyone today were taken as proof by many people. See for example https://en.wikipedia.org/wiki/Cottingley_Fairies

goshx• 7 years ago

Photoshop has been used for years to successfully fool millions of men and women that consume magazines showing people with smooth skin and sexy bodies.

donkeyd• 7 years ago

This is exactly what I wanted to post. People are becoming insecure about their bodies because of fake images of famous people. It creates high expectations that can never be met in real life and seriously ruins lives.

mendelbot• 7 years ago

The "Ethical concerns" section in the article feels like a punt. The author quoting "this technology is really about better storytelling" is aspirational -- the technology's story will be written by those who use it, and you can bet people will use this maliciously.

vernie• 7 years ago

And that's loads more self-aware than other researchers' I've seen completely blindsided by obvious ethical questions at end of their paper talks.

notaboutdave• 7 years ago

See this? This hand grenade I'm leaving in the schoolyard? This is about better games of hot potato.

okrad• 7 years ago

"Fixed film dialog without reshoots" and "better story telling" seemed rather compelling to me.

Perhaps the malicious use cases are more obvious due to how trustworthy a video can appear.

eatbitseveryday3• 7 years ago

I had this idea that devices which record content like images or video should have an unforgeable key internal to their hardware, like we have with PGP / GPG. Content that comes from the device would be signed, and allow users to validate whether it originated unmodified from the hardware source.

Granted, derived content will fail validation, but it will motivate tracking down the original, until validation can be performed. Maybe you can take pictures of fake imagery printed onto large high-def paper, but at least you eliminate one stage in the process...

Honestly, we should not trust digital content these days.

dooglius4• 7 years ago

unforgeable key internal to their hardware, like we have with PGP / GPG

PGP involves a private key, and if you have the private key you can "forge" any message. If you put the key in hardware, it can be read by an adversary with access to a powerful microscope.

thefreeman• 7 years ago

I mean how is this any different than the "secure enclave" on iphones or other forms of hardware security modules. Yes a sufficiently advanced adversary with an electron microscope could possibly extract the key but it still greatly raises the bar for 99% of other would be abusers.

surye2• 7 years ago

But each device could have it's own unique signing key, signed by a manufacturer master key that is not distributed. Then you could revoke just that one device if the need arises (or even a manufacturer). Isn't this a fairly solved problem in the TLS/CA space?

avmich• 7 years ago

Content that comes from the device would be signed

Wonder if that is possible. One can always convert a picture to a 2-D array of RGB values, so signature can't be in the video (or image) file container. So it has to be a watermark of a kind. If the algorithm is known, it's interesting if the signature can be unforgeable. If algorithm isn't known, then other issues (like with DeCSS) can appear.

surye2• 7 years ago

A signature doesn't have to be protected from malicious removal. If a bad actor removes it (or modifies it), then the video can no longer be validated and is considered compromised/unknown (which may be the same thing depending on how important validity is). DeCSS was removal of encryption, not signature, which is a whole different set of concerns and goals.

elliekelly• 7 years ago

Could it be something even more low-tech? Kind of like those faint yellow dots that all printers must print. Maybe a distinct pattern of pixel coloration that a person wouldn’t even be aware of unless they were looking for it.

eatbitseveryday3• 7 years ago

The algorithm which validates would need to be resistant to attacks, e.g., positioning of these pixels.

There are techniques for steganography, which may apply here, I wonder. Like adding a digital fingerprint in someway directly into the media itself.

Findus23• 7 years ago

This idea already exists: https://en.wikipedia.org/wiki/Machine_Identification_Code

pavanky• 7 years ago

Alternatively, we need to see how well these algorithms can fake videos from different angles in a consistent manner. If we have enough enough people recording, it may just be enough to prove something is not a fake.

w1nst0nsm1th• 7 years ago

In that case, video editing softwares and materials should also conceal a private key, in order to trace the origin of manipulation.

The whole chain of video production should be signed in order to trace filming and edition altogether and, all intermediary signatures from each stage of production process contained in a final signature.

That may be a business opportunity or at least (may be preferably) an interresting open source project.

anticensor• 7 years ago

You would need to sign a video with at least four keys: a public key from you, a private key from editor software, a public key from editor software, a public key from editor software generating the previous version of the video. And you would generate signatures for all public-private and private-private pairs and a seal value from a signed hash of those signatures, which is signed by a combination of two public keys in the scheme.

DoctorOetker4• 7 years ago

you might be interested in this discussion: https://news.ycombinator.com/item?id=16947652

hanniabu2• 7 years ago

Where is the audio coming from? If that was computer manufactured that was pretty good because it sounded very natural

chefandy• 7 years ago

The video mentions that the audio was recorded separately, and shows a few other options like text-to-speech (which obviously doesn't match the voice) and some smarter voice matching audio generation (VoCo) which could pass for the original voice sent over heavily compressed, low-bandwidth video conferencing or something like that. I'm guessing that if this is used for actual disinformation, finding a voice actor/audio engineer to try and match the speech would be most effective.

jakob223• 7 years ago

Most of the examples in the video they had the subject record the audio separately from the generated video. In one or two of them they cite some audio-generating thing.

hkon• 7 years ago

Looking forward to the autogenerated youtube videos about various topics. Perhaps they would be interesting to watch rather than a few images and a robotic voice.

scott_s• 7 years ago

I imagine the most above-board use for something like this would be in scripted tv and movies. Basically an enhanced form of ADR (https://en.wikipedia.org/wiki/Dubbing_(filmmaking)#ADR/post-...). Of course, I anticipate plenty of nefarious uses.

A place where I don't think it will be used much is actual facing-the-camera-talking-head content. Something we have learned from YouTubers is that audiences don't care if there are discontinuous cuts during a monologue. YouTubers don't try to pretend they did it all in one take, and will happily edit their video as if editing text. The cuts are obvious in both the audio and video. And still it works.

keiru• 7 years ago

The danger is not in false positives, but false negatives. The very existence of this kind of things erode trust and sow paranoia.

A simple morph cut in a John Pilger interview of Assange made a sizeable portion of nutjobs believe Assange has been long dead. Don't think this kind of behaviour can't eventually extend to the mainstream.

astrodust5• 7 years ago

It's the slow erosion of video evidence being trustworthy.

anigbrowl8• 7 years ago

Cryptographic signing of video footage is a useful blockchain application, but it will presumably be subject to the same flaws as domain security certificates.

sdinsn• 7 years ago

I agree. This technology will give everyone plausible deniability.

falcolas• 7 years ago

Forget deepfakes and such for a moment...

Think of the impact of this on dubbing movies between languages. This seems like an incredible tool.

Of course, we can’t just forget about deepfakes and such, but this particular usecase kind of excites me.

guelo• 7 years ago

I've watched many dubbed movies and tv shows and the slightly-off lip movement never bothered me, you stop noticing it after a bit. It wouldn't be that big of an improvement.

bradlys• 7 years ago

Well, it bothered me. I never watch dubs - if possible. I watch subs instead.

babyloneleven• 7 years ago

Well, it is if you are deaf.

drilldrive• 7 years ago

What would we call such a category of movies? Vubs?

jonplackett• 7 years ago

At least they can’t synthesise the voice automatically. But pair this with Lyrebird.ai and you can basically just stop trusting all video right now.

lopmotr4• 7 years ago

You can go to the official website of whoever is supposed to have said it, or a trusted news organization who recorded it and see if they publish the same video. Internet commenters will be quick to point out problems like that. In case it was recorded secretly or it's a crime occurring or whatever, then you should never have trusted it anyway because you don't know the context. Taking things out of context is routinely used by the news to fool people already. Stop uncritically believing whatever appears on your screen if you're concerned about what's true.

fooey• 7 years ago

Adobe was showing off some amazing tech for editing voices several years ago, called Adobe Voco

https://www.youtube.com/watch?v=I3l4XLZ59iw

doctoboggan• 7 years ago

Science Fiction author Greg Egan wrote a novel called Distress[0] where the main character is a science journalist who makes documentaries. He uses software exactly like this. The book was published in 1995. It's a very good book and I highly recommended it and basically any other book written by Egan. (My personal favorite is probably "Diaspora" followed closely by "Permutation City".)

[0] https://www.goodreads.com/book/show/19328253-distress

lainon• 7 years ago

Project page: https://www.ohadf.com/projects/text-based-editing/

imgabe• 7 years ago

I suspect in the not too distant future we'll need a way to produce provably true videos. I'm thinking something like the subject, a politician giving a press conference for example, carries something that emits a signal that the cameras encode into the video in a way that any alterations could be detected, something like a cryptographic signature. I don't really know enough about cryptography time be sure how / if it would work.

underwater3• 7 years ago

More likely that trusted sources become more important. A politician can share their own stream of a speech.

britch• 7 years ago

Right... What about when someone posts a video of them saying a racial slur backstage.

The politician says it was doctored.

The poster says it is unedited.

How do you verify who is telling the truth.

lopmotr4• 7 years ago

Multiple adversarial journalists recording it, like we currently have, would protect against that.

slimsag• 7 years ago

news agencies could just use an encrypted video stream, since only they have the private key you'd know it came from them and could be trusted as from them.

hoekit• 7 years ago

Or perhaps a return to analog, non-digital, older media such as paper film and video rolls will serve as "probably true".

yokohummer7• 7 years ago

I believe you already know this, hence the "probably" part, but such analog media can easily be produced from (modified) digital media, thus no credibility.

jazzyjackson• 7 years ago

I've been trying to think of a way to popularize digital signatures so when a video is purporting to be a CSPAN clip, you could check that the authorship is really CSPAN, but this always relies on trusting your sources, and who's to say CSPAN won't get hacked? Maybe they already were. [1]

I like your idea, it reminds me of something I stumbled upon reading about the 60Hz hum of AC electricity.

"...this side effect has resulted in its use as a forensic tool. When a recording is made that captures audio near an AC appliance or socket, the hum is also incidentally recorded. The peaks of the hum repeat every AC cycle (every 20 ms for 50 Hz AC, or every 16.67 ms for 60 Hz AC). Any edit of the audio that is not a multiplication of the time between the peaks will distort the regularity, introducing a phase shift. A continuous wavelet transform analysis will show discontinuities that may tell if the audio has been cut."

[1] https://www.nytimes.com/2017/01/12/business/media/cspan-russ... [2] https://en.wikipedia.org/wiki/Utility_frequency#Audible_nois...

jeffrallen• 7 years ago

They all wear those stupid flag pins, so why not have them flashing a digital signature in IR? This is almost so bizarre it's genius...

polygot• 7 years ago

Reality TV is probably going to have a hay-day with this

dpau• 7 years ago

Interested to see if counter-measures begin to be deployed in order to make a deep fake more difficult and buy time. Incorporating dynamic backgrounds and body gestures like touching one's face while talking.

nickjj• 7 years ago

Seems really cool but I wonder how well it will handle a case where you want to swap a phrase for a phrase, but have the new phrase have a "human specific" emphasis or variant to it.

Example: "That was a short trip" vs "That was a reaaaaaalllly long trip".

Language is so much more than words. When you deliver the variant message, your whole facial expression might change. So much would get lost if that doesn't carry over. Your facial expression and tone in that context also completely changes the meaning from you enjoyed the long trip to not enjoying it, but how can a machine know which one to pick.

ionwake• 7 years ago

Is there any open source software like this available anywhere ?

readingnews• 7 years ago

“Unfortunately, technologies like this will always attract bad actors,” --- uhmmm, what is the "good actors"?? I want to take video, and delete what you said and make you say something else. uhmmm, can't actually see the "good actor" point of that.

aidanlister• 7 years ago

Read the article?

jldugger• 7 years ago

Well, now that scene in movies where the newscaster narrates the aftermath of an event can be done realistically without having to hire actors?

w-ll• 7 years ago

It could be used to bleep movies for tv so that the reading the lips of the actor match the dubbed audio. You can easily tell what curses people are saying.

pault• 7 years ago

Making edits to scenes in movies during post production, cutting out mistakes in broadcasts... I'm sure video production professionals could easily come up with dozens of use cases.

sabujp• 7 years ago

youtube and other video sites need a framework for content verifying videos that haven't been digitally produced using steganography or keys injected into every frame. This could software could be embedded in video recording devices (keys could be updated OTA so hard hacks don't matter). If videos haven't been "reality" verified then people can just enjoy them as fake/fiction works. Video editing/compression software would need to be aware of the location of the key bits and maintain them within each frame.

devin• 7 years ago

people choose to believe things. The kind of people who will be duped want to be duped as it serves their own ideology. You can’t change that.

mopsi• 7 years ago

Everyone seems to think of nefarious uses, but I can't wait for this tech to appear in video calls, combined with translation. This could enable two people without a common language to have a conversation while appearing to each other as native speakers of their respective languages.

castis2• 7 years ago

Wow, that would be something else. I assume whoever gets that to market first would make a killing.

html5web• 7 years ago

S*. It’s scary

MichaelEstes• 7 years ago

It's strange to me that people are so concerned about these deep fakes when the National Enquirer has been around for so long. It's been easy to lie to people in mass for awhile now. I don't think this changes the number of people that are open to these suggestions, I think people in general are smarter than a lot of people give them credit for.

meribold• 7 years ago

On the flip side, it may become harder to convince people of the truth when there is such a convenient way to reject unwelcome video evidence. This could amplify echo chambers.

visarga• 7 years ago

Now we need cameras that watermark (using steganography) original videos so they can be authenticated and a blockchain solution for registering originals. Video sharing sites will need to process all their uploads to check for modifications and serve the original file (not re-encoded) as well, for third party checking.

hasahmed• 7 years ago

In regards to all the talk of not being able to verify your video source; perhaps we'll go back to film for things that need to be probably legitimate. Though perhaps that has the same issues.

ollifi• 7 years ago

I find it worrying that only now people start to be sceptical about visual information. There is a huge difference between real world and the framed and curated view photographer or documentarist gives you. If you go to school to learn about this stuff it's mostly about how to convey your view through these tools and the ethical implications of it.

I think people vastly underestimate how much editing and framing change the perceived truth of what happened. It is more subtle than manipulating the contents of video, but I think it can be in many ways more effective as most of this stuff bypasses your cognition and is not straight up lying.

It feels the same as in written news changing the quote vs. changing text around the quote.

I think we would be better of looking at video like it was picture drawn or text written by someone. It's an artistic rendition of the events.

drawnwren• 7 years ago

I completely agree. The hardest part of returning from military operations overseas was that Americans inevitably had their views shaped by one of two prevailing video viewpoints. Trying to discuss what happens over there in any meaningful way with someone who hasn't been inevitably devolves into an emotional reaction (usually roughly pro or against) instead of a conversation.

rjf72• 7 years ago

This [1] image is a perfect embodiment of everything you're saying. But I'd add that I think there's a dissonance in our belief in other people's naivete and the actual numbers. In particular trust in media has fallen off a cliff. Only 31% of people trust in mass media to "to report the news fully, accurately and fairly" at least a fair amount. [1] If you restrict the question to those 18-49 years old, it's 26%. People may not be able to precisely point to why they feel this way, but it's clear that people have become deeply skeptical of media in general. The reason this is relevant is because media, in turn, relies heavily on what you're talking about - images framed to convey a narrative.

I think what we really take issue with is something related but different. And that is people voluntarily believing things that confirm their biases, while ignoring or even denying things that challenge them. Pew did a nice piece on that here [3]. These [4] are just their poll questions, which are quite interesting. For instance, "spending on social security, medicare, and medicaid make up the largest portion of the US federal budget" - 41% of Americans incorrectly labeled that as an opinion. And of those that labeled it an opinion, 82% further incorrectly labeled it as false. Going the other direction, "increasing the federal minimum wage to $15 an hour is essential for the health of the US economy" - 26% of Americans incorrectly classified this as a factual statement, and of those 83% claimed it was accurate. The same is of course true on both sides of the aisle, with the questions affirming as such.

People have a tendency of believing what they want to be true, while challenging (often quite aggressively) everything that goes against that, or even simply denying it. This can create the perception of a naive people being misled by malicious actors, but I think reality is that people tend to pick the views that they want to be true often for entirely subjective reasons that cannot be clearly qualified, and then work to find evidence to support that.

[1] https://www.washingtonian.com/wp-content/uploads/2017/01/gar...

[2] https://news.gallup.com/poll/195542/americans-trust-mass-med...

[3] https://www.journalism.org/2018/06/18/distinguishing-between...

[4] https://www.journalism.org/2018/06/18/distinguishing-between...

_Codemonkeyism2• 7 years ago

Every news is to some degree manipulated.

Together with a friend I was one of the geocaching and confluencing pioneers in Germany.

Some large papers and TV stations reported about this "phenomenon" and wanted to make an article/ documentation about us. Every one of them came with a story in their head which we had to fill with our pictures and quotes. No one was interested in "reality". For a news clip we had to shoot situations several times, I remember leaving a house 5 times until they shot was done. Up until then I thought news would be unstaged.

DoctorOetker4• 7 years ago

the moment they essentially ask you to cooperate in staging a scene, why would you cooperate?

kingkawn2• 7 years ago

This tech allows the state or corporations to quietly adjust the historical record of their representatives words and statements to fit their ambitions at any given point.

castis2• 7 years ago

How is that any different than the bullshit they do right now.

kingkawn2• 7 years ago

It’s not at all, that’s how you can be pretty sure it’s being done basically as soon as the tech gets invented.

vinayms• 7 years ago

I actually love this ongoing cat and mouse game. I don't follow the events in this field keenly so I don't know if it exists, but the challenge is to find antidote to this concoction, created by mad scientists just for sake of science, that will be weaponized anytime now.

ozychhi• 7 years ago

Was I only one who thought of Jim Carrey in God almighty?

mnw21cam• 7 years ago

Bruce almighty?

sbhn• 7 years ago

The algorythm demonstrated in this video, that moves the cats mouth in real time to the naration, was written in javascript and html5 audio.

https://youtu.be/qdN__7C5kl4

apsharma• 7 years ago

This is gonna revolutionize the animation industry.

program_whiz• 7 years ago

All human evidence rests upon the shaky foundation of "because I believe its true", at the bottom of which rests the shaky foundation of your personal experiences. Don't believe me? Just ask a schizophrenic how hard it is to disbelieve your own experience.

Reminds me of the HP Lovecraft novel "The Call of Cthulu" Page 1, Paragraph 1:

The most merciful thing in the world, I think, is the inability of the human mind to correlate all its contents. We live on a placid island of ignorance in the midst of black seas of infinity, and it was not meant that we should voyage far. The sciences, each straining in its own direction, have hitherto harmed us little; but some day the piecing together of dissociated knowledge will open up such terrifying vistas of reality, and of our frightful position therein, that we shall either go mad from the revelation or flee from the deadly light into the peace and safety of a new dark age.

mzs• 7 years ago

lead author's page with links to other researchers' pages, the paper itself, and nearly 200MB of supplementary materials: https://www.ohadf.com/projects/text-based-editing/

Algorithm allows video editors to modify talking-head videos as if editing text

Related Stories