Early in October we released Portrait Light for the Pixel Camera and Google Photos. This feature, gives you the capability of adding an additional light source in your portraits and to enhance pictures in post-capture.
Today we released a Google AI Blog Post that explains the details of the tech. Our method builds on top of our Relightables system to generate training data and then to train a machine learning network to infer geometry as well as illumination of the current portrait.
Using this information as input, a very efficient neural renderer generates a quotient image which is then multiplied to the original image, rendering the final result while preserving high frequency details.
To learn more about the method, take a look at the blog post!