I have seen quite a bit of talking avatar, animation that sometimes look like the real thing and some other times are clearly intended to represent an artefact. What surprised me when I saw the video clip (watch it) was the amazing capability of endowing a life into a still image, like the one of the Mona Lisa portrait.
If you take some time to watch the video you can learn how a pool of researchers at the Samsung AI Centre and Skolkovo Institute of Science and Technology have managed to apply AI to animate in a very credible way still frames. You can read all details in their published paper, and it is surely worth reading.
I won’t go into the detail of the technology an techniques they used, if you are interested you will get all of them in their paper (and a quick overview in the video clip) and if you are not interested in technicalities you will find my recap boring…
What I like to share is my surprise in the capability of a machine to create an animation that gives me a feeling of looking, and discovering a person. As you might expect, the researchers are pointing out that the more still images they can get of a face, showing it from different perspectives, the better the quality of the animation. Yet, for me it is exactly the opposite. I wouldn’t have been surprised to see a good animation of a face if the software can start with a significant number of images. In a way the problem (not a trivial one, I know) would have been to connect the various images making small alterations to match what was being said.
With Mona Lisa (or Einstein, another character they chose) it is different. Starting from a single image the software had to invent a person and in doing that it provided that animation with a specific character. I was surprised seeing Mona Lisa talking because what I saw was not what I would have imagined as a Mona Lisa talking. I always thought of Mona Lisa as a sort of shy and reserved character (no reason for that, it is just how I felt looking at her portrait) but the Mona Lisa created by the software looks to me a much more easygoing person, a person that is open and likes to talk to me (and anyone else). Of course this is just my impression and yours can be quite different (like to know what it is … if you like to share).
The point I am making is that the software developed by these researchers is taking decisions and infuse a soul into a inanimate image, creating something that is completely different, that has a life of its own. I find these fascinating.