I run a website that makes use of AI to digitally recreate images of a variety of people from the past. To quote my own blurb ..."Bringing to life historical figures using a blend of source artifact, AI/Machine Learning neural-net tool, photo editing software and other historical input to produce a likeness.".
The initial inspiration to make these digital recreations came from having read a book on ancient Rome and subsequently coming across some work online to digitally recreate Roman emperors from various historical sources such as busts, statues, coins etc. Fascinated by the recreations I tried the method on Horatio Nelson and a face mask taken five years before his death at Trafalgar and was pleasantly surprised with the results, enough to refine the figure and then continue with a series of other figures. One of these figures coincidentally was another member of the Mary Rose crew, the Vice Admiral himself, George Carew (right). Further inspiration often comes from a book, museum visit, podcast or in the case of the Mary Rose crew from a TV documentary.
The excellent TV documentary "Skeletons of the Mary Rose: The New Evidence" featuring the Trust's own Dr Alex Hildred included the brilliant images of the Mary Rose crew created by the facial reconstruction expert Oscar Nilsson. The images of the crew seemed like the perfect fit to run through the AI process. It's often the story behind the figures that makes you want to get closer and the story of the Mary Rose, her crew and the world of the Tudors that they came from is such a fascinating one.
The method used to produce the recreations employs one of a number of online tools that use image generating Machine Learning models based on a type of Neural Network called a GAN. A GAN (Generative Adversarial Network) improves on it's initial training by utilising two neural networks that compete with (hence the 'adversarial), and encourage improvement of, each other. One of the features of a GAN is it's ability to produce portraits of fake but realistic human faces based on a supplied source image.
The end results can vary but the AI is usually very adept at producing an image. The tool has some inbuilt ability to adjust a limited number of features however to break free from the inherent constraints and limits of what the tool produces requires the use of digital editing software and no small amount of creativity.
The heavy lifting of the AI tool is done on its own remote servers so no special equipment is needed. For the digital editing a standard laptop or similar will suffice.
The time required to produce the images from start to finish can vary enormously particularly depending on how much digital editing is required to get the AI output closer to the to the required final image, obviously a very subjective process.