Deepfake videos are getting better
 

Armed with a GPU cluster and some clever code you can do amazing things.

The transitions in this video are stunning.



Github repo here: https://github.com/iperov/DeepFaceLab

That was really good and the face shifts were mind-bending. Impressive....

That was harrowing . . .

He´s got some skills :2 cents:

I love the smooth transition. I expect the software to only get better with time. Maybe we won't need actors at some p in the future.

I see it happens but it's not attractive like the previous one with Jim Carrey.

many people fear deepfakes will reform fake news and destroy the current world order.
pretty bold of an assumption given that the majority of humankind already believes everything that's written in ALL CAPS AND WITH ENOUGH EXCLAMATION POINTS!!!!!!!!!!!!!!

this tech is about to explode. Seen a few unreleased stuff lately that are mind blowing.

man is this going to get abused for fake news...it is only a matter of time before somebody rolls out an easy to use app...

Camera manufacturers are going to have to come up with some new tech to properly authenticate footage which "needs" to be real (ie more than just fun like this). Maybe something like the source camera digitally signing every frame. But it only takes one person to reverse engineer the firmware to find the key (or the method used to generate it) and then that could also be faked...

far out, psychedelic

Well, that's not both really neat and extremely unsettling.

… FakePorn.com domain for sale if interested :) message meh

voice can't be faked that well. therefore most of the successful deepfakes will be mute.

Yet... But the visuals are pretty good...

Once the visuals are 100%, the audio will quickly follow...

item # 9!

I’ve been playing around with this for about a month now. Most of it’s done with free open source software. I’m in a FB group with a bunch of these folks. The key is having a decent Desktop CPU / GPU and some patience. The best results can take up to a week to produce. The longer the program “learns” the better the results.

This excites me for the very last reason he mentioned in the video - Audiobooks.

Im a great lover of audiobooks and it will be fantastic to be able to choose your own favourite narrator for whatever book you want to listen too, and also the fact that no one would actually have to read the book itself - Just a text to speech from the already written text, in whoever's voice floats your boat - Superb !

THat's amazing!!!

And we're all fucked.

:1orglaugh:1orglaugh:1orglaugh:1orglaugh:1orglaugh :1orglaugh:1orglaugh:1orglaugh:1orglaugh:1orglaugh :1orglaugh:1orglaugh:1orglaugh:1orglaugh:1orglaugh :1orglaugh:1orglaugh:1orglaugh

A large tech firm that we all know has something better.

A Japanese research team will be releasing code that runs on NVidia Tesla cores to train speech, the early demos are frightening.

In the next 12 months this space (deepfake videos and voice manipulation) is going to explode.

In a few years we will be at a point where you can't be sure what is real anymore.

I Cagified one of the videos I did for a client. This was done with a Ryzen 2nd Gen Threadripper and a GTX 2070. I let the software process and learn for about 12 hours.

https://i.imgur.com/LUSH0G4.png

Just curious if you can tell the difference running a 2070 vs something like a 1080gtx?

You wouldn’t see a difference visually, but you’d probably have to let the computer process and learn for a bit longer. The longer you let it run, the better the results.

I’m working on onet that puts Chris Cuomo’s face on Fredo and his brother’s on Michael from The Godfather.

Wow, that's crazy and a little scary!