Deep Fakes
Recently someone revealed that he managed to use deep learning libraries to generate fake videos after training with celebrity images.
A few days ago, there were multiple stories mentioning how there is now ready-to-go apps that can do this, and even come with pre-trained neural nets for specific celebrities. And groups on Reddit dedicated for the resulting videos.
While this was obviously going to come sooner or later, it means that very soon video evidence is going to become meaningless. There are already other papers about how to sync mouth movement to generate audio, and how to create audio that matches a voice. That is pretty much all that is needed for very high quality fake videos.
The ones generated by the app are not that convincing yet, but I'm pretty sure that is just a matter of time, and probably not a lot of it, either...
A few days ago, there were multiple stories mentioning how there is now ready-to-go apps that can do this, and even come with pre-trained neural nets for specific celebrities. And groups on Reddit dedicated for the resulting videos.
While this was obviously going to come sooner or later, it means that very soon video evidence is going to become meaningless. There are already other papers about how to sync mouth movement to generate audio, and how to create audio that matches a voice. That is pretty much all that is needed for very high quality fake videos.
The ones generated by the app are not that convincing yet, but I'm pretty sure that is just a matter of time, and probably not a lot of it, either...
Comments