We've seen many examples of AIs doing image manipulation tasks and recreating things.
From the infamous deepfakes, to AI in creating nude arts, AI in creating artworks, to then AI in creating the faces of people that don't exist, to the latest, the controversial but fun FaceApp. Computer vision has become a lot smarter in each iteration.
And going after them, is where researchers at the MIT-IBM Watson AI Lab created an AI capable of turning people's selfies into artistic portraits.
Called simply as 'AI Portraits', its an AI artist capable of recreating arts by following some of the world’s greatest masters
The AI is powered by a Generative Adversarial Network (GAN) to rebuild users' selfies from scratch, by rendering them in faux oil, watercolor, or ink.
GAN as a popular AI technique, involves pitting two neural networks against each other to produce an acceptable result: a 'generator', which looks at examples and tries to mimic them, and a 'discriminator', which judges if they are real by comparing them with the same training examples.
In this case, the researchers used 45,000 portrait images to train the program, including paintings by Titian, Vincent van Gogh, and Rembrandt, as well as other works from various media.
This alone, is essentially a step forward beyond existing AI technologies used to 'paintify' existing photos.
None of the portraits the AI creates include smiles or laughs, because it was uncommon for such facial expressions to be painted in the era of the training examples.
The AI is primarily focused on 15th century art, which should explain all the straight, high noses and much paler skin tones.
The results can beautiful, amusing and mesmerizing, but at the same time, harrowing and haunting.
"Portraits interpret the external beauty, social status, and then go beyond our body and face," its creators wrote. "A portrait becomes a psychological analysis and a deep reflection on our existence."
When users use the AI, the tool automatically determines which style to employ based on the specific portrait users have uploaded. It then reconstructs the image in an approximation that proves very faithful to the originals upon which they’re based.
It should be noted that the certain elements within any selfie may prompt the algorithm to use a specific style.
In an example provided by the site, the creators said that the model settled on a "Renaissance style" and recreated its subject by "highlighting the elegance of the aquiline nose, the smoothness of the forehead."
If users worry about the privacy implications, like how the face-aging FaceApp is giving, the researchers at IBM noted that users' source images are deleted immediately once the AI is finished generating the results.
User photos are sent to the creator's servers, just to make the AI work. The researchers promise they won’t use this data for any other purpose.
What this means, there should not be any privacy issue here.
Its website at aiportraits.com however, does have a steady stream of generated photos uploaded by other users. But the digital paintings aren’t really photo-accurate - which is actually the point of this AI.
Furthermore, the showcased pictures don't have any identify information attached.