(no title)
dubin | 8 years ago
It was shockingly easy to do. We're technical, but neither of us know a lick about machine learning. It took a couple hours to collect training data (we turned speeches/interviews from each of them into thousands of photos), and 20 hours to train the model.
In the future you could automate the data collection for a person even more, to the point where they just need to film a selfie of themselves for a couple of minutes in a couple of different lightnings, and boom, after you train their decoder, you could put them on any celebrity.
EDIT: This is what we used https://github.com/deepfakes/faceswap Would be happy to walk anyone through how to do this
michaelbuckbee|8 years ago
some-guy|8 years ago
[1] https://www.hollywoodreporter.com/thr-esq/back-future-ii-a-l...
berberous|8 years ago
As you might expect, California leads the way in this area in the U.S., given their interest in protecting Hollywood / celebrities.
Someone|8 years ago
Law works similar with copies of physical objects. For example, you are free to make a 100% copy of an iPhone, the Mona Lisa or a pop song and enjoy them at home, but you can’t display them in public or sell copies, or even give them away (did I say I’m not a lawyer?)
gspetr|8 years ago
Without any precedents YMMV.
rmc|8 years ago
joshschreuder|8 years ago
https://www.youtube.com/watch?v=t-6IRuCaDw0
Nice work, it's a pretty amusing and somewhat worrying video of what's to come.
howToLearnSpark|8 years ago
dubin|8 years ago