I know you like neural networks, so today I bring you to have fun with Vid2vid . Developed by Nvidia researchers, this neural model of video synthesis is capable of reproducing a “talking head” from a simple photo. Very practical for visios where a simple profile photo could become much more lively.
To illustrate their techno, the researchers have put a demo online that allows you to change the angle of rotation of the head, but also the position of the eyes, from a photo. You upload the image, you frame the head, you play with the sliders and then you click on the “Rotate” button and you will get a really stunning new image.
For example, here is my essay with Edward Snowden. On the left there original photo, on the right Edward probing your soul.
The rendering is not bad, don’t you think? A little bug with the glasses, but other than that, we believe it.
Even if we only touch the adjustment of the position of the eyes, we can have a funny thing.
To see now what it will give when it is implemented in a software that will make video and that we can test. In the meantime, here’s what it looks like and everything we can do with it:
