Machine learning does much of this hard work for you — if …
The first thing that came into my mind when I first heard about Deep Fakes, is what would happen if we could create DeepFakes in realtime and not just for existing videos or photos? If you want to see some examples of the fake videos generated with the presented approach, check out the full Supasorn’s talk below. collecting several photos and videos of a person; reconstructing a fine-detailed 3D face model from the collected images and videos without performing 3D scanning on the person; generating sharp facial textures and colors for a wide range of expressions; showing the machine the video footage of a particular person to teach it how to imitate unique mannerisms of this person; synthesizing the movements of the mount points by using a neural network to convert input audio into those mouth points; synthesizing the texture, fine-tuning teeth and other facial details, and blending it into the head and background from a source video. Suppose we can go online with someone else’s face, would this be funny or would this push the ethical boundaries even further? With this technique we can create a very realistic “fake” video …
“Print propaganda is as old as the printing press, video propaganda as old as TV. Such possibilities of AI video generators can be successfully leveraged in different business applications, especially in marketing and advertising. But AI helps automate the process, making it quicker and more accessible. It was very amusing to get online as John Oliver and invite friends to see their reaction. If you want to make a fake video using Photoshop, you have to edit every individual frame.
To assist in preventing the misuse of this technology, Supasorn is currently involved in developing tools that utilize a combination of machine learning and human moderation for easily detecting fake images and videos. Deep fakes is a technology that uses AI Deep Learning to swap a person's face onto someone else's. The trick happens when we encode the picture from the first person with their encoder, but decode it with the second persons decoder! Snapchat has been doing face-swapping for years, and computer engineers have long used machine learning to mash visuals together. As first reported by Motherboard, a Reddit user named “deepfakes” has been applying these tools to create content for the pornographic subreddit r/celebfakes. The danger if these tools become widespread is self-evident.
This became a reality after researchers started using an auto-encoder neural architecture to accomplish this. The Best of Applied Artificial Intelligence, Machine Learning, Automation, Bots, Chatbots. With nothing more than a few photographs and some free software, it’s now possible to quickly graft anyone’s face onto pornographic images, videos, and GIFs. He even created a full video with the face of Wonder Woman star Gal Gadot inserted into an incest-themed porn scene.