What makes a human face? What makes your face? See a version of yourself as made by an AI system, and take a synthetic-selfie. Use your synthetic-selfie for profile pictures and see if people notice: deepfake-selfie. How does it work? A generative model is trained to create portraits by looking at tens of thousands of photos, the software seeks to find you in its vast internal space of human faces. It then creates its own impression of you.
How are your unique features captured by this machine learning technology? Are you well represented by the images used to train the model? Technology is a mirror of culture, and with a project that is literally mirror-like, the data sets employed are directly reflected. Diverse representation in training-sets is one key aspect of an AI model created to mimic any human face it sees. Another is data privacy, specifically in relation to face-tracking. One could use a synthetic-selfie as a mask to confuse tracking online. On the other hand, how will the software for this project use your likeness? Your image will not be retained, although the generative model of your face will be available online.
In an exhibition scenario, visitors see a screen with a continuously-morphing series of synthetic-faces generated by the software. Upon detection by webcam, the deep learning model creates a generative model of your face based on its training data. This uncanny "reflection" follows your movement and mimics your facial expressions. To capture a "selfie" you simply press a button and the image is saved to a website where it can then be downloaded.
Synthetic M Wingren in action
User tests w/first "Strange Mirror" iteration, no selfie-button
Synthetic Andrzej Pisarek is generated