16
Oct

Novo algoritmo EDITA VÍDEO como se fosse TEXTO (deepfake)


We’ll see an algorithm that serves to edit videos as if you were editing text! (I.e. “Deepfake” is a technology that
really impresses me! They had 3 deepfakes that
they messed with me. The first was made by the Corridor class,
simulating Tom Cruise talking to them. They even hired a guy who did imitations of Tom Cruise to stay as faithful as possible. [Tom Cruise – Deepfake]: “You know, I’m here
to help. I’m a helper … (laughs) … That’s what i do! Help me help you! (laughs)” The link is in the description. The second deepfake was also made by
Corridor crew, but this time they did a simulation of Keanu Reeves avoiding
an assault on a convenience store. It was a little less faithful, but to have done a script
Specific to the deepfake was phenomenal! (screams) (sound of falling) And the link is also in the description. And the third deepfake was made with my face
for a college job! Look at that dough! The video on the left is the original video of a vlog I made when I traveled there to Blumenau, is including in the channel this material, and on the right is a deepfake made
by Daniel Andrade. “If you also think something is missing,
or I should lapse some detail of the video, Leave a comment. It will help me a lot,
closed? A big hug!” Sensational! I was sailing one day there on the urge, then suddenly
someone marks me in a story … ok, beauty, normal … when I went to open it was a
deepfake mine! Out of the blue! The link to this material is not in the description, but I want to bring Daniel in to explain and teach us how he did it. Good, but this time we’ll talk about
a joint effort by Stanford, Adobe and other institutions to create a deepfake with an extra utility, let’s say so. And they put themselves the following challenge: “Imagine an editor that you could modify a video using a text transcript. In which you could easily add
new words, remove wrong words, or rearrange them just by dragging and dropping
them where you want to form the final video, and that it was almost perfect on the
perspective of an untrained eye. ” And that’s what they did! The algorithm takes the transcript
of the audio, tries to understand, within the video, what are the other parts that he can
reuse to assemble the new words, and then put it all together with artificial intelligence
and render the final video. Looks like revolutionary startup pitch
nowadays… “It’s like Uber, but to edit video!” But it was cool, just feel … In summary, what they did was the following: They take the audio transcript
and align this as video. Then they create a 3D version of the
face of the person. And then from a text edit they
scan the entire video to understand which is the best segment you can reuse. After that they mix a
information in the other, and render the underside of the person’s face. And with a last algorithm they render the final version of that information on the person’s face back. Let’s see an example of them
adding a new word: And I’ll tell you guys, okay …
This is fantastic for post-production! I would have used it countless times! Because it’s only at the time of the edition that you realize
Did you say something wrong? “seje” to have spoken a word in the singular
when it was in the plural, “seje” when it was to have said “be”, in fact … In a simpler production, like mine,
It’s easy to rerun the take the next day. and you would not even notice
so much the cut. But in a more complicated production, in which you often have to rent the place, or even depends on the team, then screwed! Or “seje”, you’re chipped! And now comes a funny part: There are 3 ways you can fill out
the text that was changed. The first way is the person re-recording the audio of the altered parts which is what we have seen so far. The second way is using a type voice …
level Google Translator. And the third is using an algorithm that
synthesizes your voice in a frightening way! Let’s take a look now at an example using
the voice level Google Translator, but hey … Can not laugh, okay? (laughs) I told you that … [female voice
robotized]: “I could not laugh, hahaha!” But the mass is that, in this example, it shows
the excerpts from other parts of the video which he used to reconstruct the visual part. Even the algorithm needs 40 minutes of recorded material to work. But then it shows an example
using a system called “Voco”, which is the case that synthesizes a real voice. Feel alone … I did not know this “Voco”, there I
went to search and it was a system set up roughly in 2016, and it also serves to edit
audio as if it were text. Look how interesting … The link to this “Voco” is also in the description. But now the most mass! You can create a new video with the person speaking a language they did not originally speak! Look at this example of the person speaking German! Very good! Imagine all the
possibilities this opens! And if you like that kind of subject,
This time I want to indicate a video another technology channel here from the
Brazil called Fish Babel. In this specific video, Camila
explains using examples of jaw-dropping, how these generative models work. So just click here! Closed? Thanks! (I.e.

Tags: , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ,

100 Comments

Leave a Reply

Your email address will not be published. Required fields are marked *