Disney researchers have announced that they have developed an AI system called FRAN that makes it much easier to make actors look younger or older.
Until now, it has been possible for artists to hand-tune and uses CGI to make things look as realistic as possible, but this new AI tool could do most of the heavy lifting. This AI takes only 5 seconds to apply the ageing effect to one frame.
Re-ageing an actor, is generally an expensive and laborious process, requiring an artist to go through a scene frame by frame and manually change a character’s appearance. There have been previous attempts to automate this process using neural networks and machine learning. The Disney researchers noted that while such attempts may have worked for still images, other systems generally lose face identity, have lower resolution, and have less resolution between subsequent video frames. The results were unstable. The newly developed solution is the first practical, fully automated production-ready method for reworking faces in videos.
The researchers built a database of thousands of randomly generated faces to train a FRAN (face re-ageing network) neural network using a dataset of actual individuals. Then used, existing machine learning ageing tools to age these synthetic faces and input the results into FRAN.
Neural networks can analyse images of faces, predict which parts of the face will be affected by ageing, and apply effects such as wrinkles and skin smoothing as layers on top of the original face. The researchers argue that this approach allows FRAN to rework a performer’s identity and appearance as they are, even if the performer’s head or face moves around or the lighting changes during filming. Unlike other methods, FRAN also does not require an extra step of facial alignment.
It’s simple to understand why Disney would want to create such a tool. For example, it could reduce the workload of visual effects artists and potentially speed up the process. It helps keep budgets in check.