We all know that you can do silly face swaps with snap chat and other apps. Previously face swapping faces on videos was terrible because you simply sequentially swapped all the faces frame by frame. As a result, the quality was very poor and did not handle head movements well. A Reddit user by the name of Deepfake created a recurrent neural net which accurately face swaps videos. Deepfake(the user) initially gathered a following by swapping celebrity faces onto porn videos.
Reddit has banned Deepfake and Pornhub has banned deepfakes from their platforms. Despite the attempts to remove fake deepfake videos, they are currently all over the internet.
Deepfake created FakeApp which allows anybody to use his technology to make deepfake videos. The website FakeApp contains both downloads and tutorials for anybody who is interested in using this technology. Recurrent neural nets are very computational heavy and will take you about a day of GPU power to generate a good deepfake.
There internet is currently in an uproar over this topic. In most cases is obvious which videos are deep faked, however, that can change in a few years – or even months. How do we prove that a video is credible if it could be faked convincingly? It is not unforeseeable that in the future someone will try to use this technology to blackmail a politician and smear their campaign.
Where do you draw the line between satire and freedom of speech? Most people argue that we should ban the use of deepfakes on any video that does not have the consent of the two people being swapped. However, is it fair to apply that rule to memes involving Nicholas Cage?
It is worth mentioning that there are very little laws that can be used regarding deepfaked content. If the content is seen as harassment, damaging, or violates copyright it is clearly illegal. But, besides that, there are very little laws regulating this new realm.
This technology may seem creepy and offsetting – especially when you are only considering it for face swapping celebrities on porn. The applications of this technology are limitless. Consider the movie industry, recurrent neural net technology may be used to enhance CGI. 3D modeling in video games could also use this technology to create more realistic graphics.
Don’t blame or ban the technology. There will always be people who misuse the technology, but, the benefits of such technologies outweigh the negative consequences.
Deep recurrent neural nets are here to stay. I foresee that in the near future we will have a neural net which can generate audio which sounds like someone else.