Remember, Deepfakes? Until now, you might have seen AI creating fake videos of Barack Obama or some other celebrity. But mostly it requires lots of previous footage related to that person to be fed into the AI.
Now, the researchers at Samsung’s AI Center in Moscow and Skolkovo Institute of Science and Technology are working on a new AI model that will surely blow your mind.
While it creates life-like animations of a person’s face, it can even do so just by using a single photo of that person – called single-shot learning. The model can make it speak, give expressions and even turn around the head like real humans.
So, this brings the possibility of making fascinating animations from paintings. And the researchers already have a surprise for you. Their model has made an animation of the famous painting Mona Lisa. Also, adding to the list are Marylyn Monore, Albert Einstein, and so on.
The initial training of the model does require large amounts of data to be fed. But after that, it can even give convincing results from less number of frames of people whom it has never seen.
It collects the various facial landmarks which are used to verify the face in the output to the source. But right now it is only limited to animating the face part and the results aren’t that consistent.
To achieve higher accuracy, it uses a technique called Generative Adversarial Network where two AI models are put against each other. Their task is to trick each other into believing that what one creates is real. And so the researchers can get the desired output they want.
You can read more about the Few-Shot model by reading the research paper here.