Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

nein.

or there was that time a german variety show 'false flagged' the germans by doctoring a video of yanis varoufakis to have him 'stick the middle finger to germany' during a talk he gave (https://www.youtube.com/watch?v=Vx-1LQu6mAE). There were witnesses; but they didn't stop the confusion and outrage that ensued due to the effects of viral media. Would this pass a sophisticated technical analysis of the video? No - is there time and public understanding for the results of said sophisticated technical analysis to prevent the damage before it is done? no.

The notion that witnesses or forensic analysis can combat the speed of information, especially across borders, is a bit quaint.



It's of course technically possible but it's also much more difficult. You basically have to photoshop a long series of frames and make sure that it looks natural and consistent in sequence. You have to take into account changes in lighting, focus, motion blur, things passing in front and behind the object etc...

In your example they modified a very clean source video, image is perfectly stable, very little motion and they edited a very small portion of it for a very short duration. TFA is about replacing one person by an other while they're talking and moving on a potentially lengthy video clip. Doing this convincingly by hand would be hard work.

I think a good example of the difficulty is with the infamous "moustache removal" for reshoots of Superman in the Justice League. It cost a fortune and the results were not exactly seamless. Also the re-created Princess Leia in that Star Wars movie, not bad but definitely still in the uncanny valley.


Its something entirely different to fake a face instead of a finger in a video. Then again, the deep fakes I have seen so far arent really that convincing either. But one can only wonder how much you can improve such a thing. With politicians of whom a wide array of video material exists and were you have the option to create an incident, instead of altering existing ones, the possibilities are really interesting.


did we really post that link at the same time, or did you edit your other response to match?


I posted the Böhmermann incident 2 minutes prior and see it more as an example for the state of the art in video manipulation. I think its hard to see the response back then as an example of future interaction with the topic of video fakes. Which when it comes to altering faces in videos is still in the future.

I edited the post you replied to with "in a video." in the first sentence to clearify


His response was generated by a GAN.


Could you elaborate? I dont understand what you are talking about


It's ironic that on a discussion about copying images of parts of someone's body over to images of another person's platz would ask "did we really post that link at the same time, or did you edit your other response to match?" i.e. copying over parts of one person's comment to another person's.

(And the GAN is the neural network model which does these transformations)


"Falsehood flies, and the Truth comes limping after it."

—Jonathan Swift

https://quoteinvestigator.com/2014/07/13/truth/




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: