online tool AI portraits is developed with the intent to generate realistic images — the level of resemblance of the resulting portrait is close to the user’s face but deliberately not too similar to the starting image. using generative adversarial networks (GANs), the website scans your picture and creates a resembling face by overlapping millions of actors’ photos. the goal — as the developers explain — is to share the experience of being portrayed by an AI algorithm, discover how AI sees you, while there is no aim to improve or deform the starting picture.
AI portraits always generates a frontal close-up — the quality of the portrait improves with a frontal view.
‘we use frontal images of actors to activate portraits with our GAN,’ explains mauro martino, the co-developer of AI portraits. ‘the result is an image that examines the concept of identity, pushing the boundaries between the individual that recognizes herself/himself and the collection of faces that are sedimented in the neural network.’
the tool finds a way to portray the parts of your face missing at the photo
what you see at the generated portrait is a fictitious recreation of yourself — the training data of the actors vaporizes your image and creates a dialogue between you and the celebrities. the pictures generated by AI portraits are completely new and inspired by the face identified in the photos. there are no filters or visual effects — the portrait is generated by the company’s artificial intelligence algorithms.
AI Portraits continues to generate a frontal close-up by reconstructing the missing parts of the face
the GAN is composed of two neural networks, one learns to recognize portraits of the actors (discriminator), and the other learns to generate them (generator)
edited by: maria erman | designboom