It’s one of the most recognisable works of art in history, and now the Mona Lisa has been brought to life by a new AI .
Researchers from the Samsung AI Centre have developed a new algorithm that can create fake videos of people, based on a single image of their face.
To train the algorithm, the researchers fed the system over 7,000 videos of celebrities from YouTube videos.
This allowed the algorithm to learn the movements of ‘landmark’ features of the face, including eyes, mouth shapes, and the length and shape of the nose.
When the researchers then fed the algorithm a still image of a face, it could pinpoint the landmark features, and apply the same movements to them.
- 1 What we need to do to raise number of women in STEM careers
- 2 Ethics should guide use of artificial intelligence
- 3 Why ethics should guide artificial intelligence use
- 4 Youth hold the key to growth of information technology in Africa
In their study, published on arXiv, the researchers explained: “It performs lengthy meta-learning on a large dataset of videos, and after that is able to frame few- and one-shot learning of neural talking head models of previously unseen people.”
To test the algorithm, the researchers fed it still images of well-known faces, including Mona Lisa, Albert Einstein and Mona Lisa.
And despite being fake, the resulting videos are eerily realistic.
While the algorithm could have a number of helpful applications, such as in the film industry, it could also be abused to create ‘deepfake’ videos.
Last year, several deep fake videos appeared online, appearing to show celebrities in explicit situations.
Celebrities who were targeted in the videos included Emma Watson, Melania Trump , Maisie Williams and Gal Gadot.