While watching a movie, you might have noticed actors looking younger in some scenes. Unfortunately, there is a limit to where the makeup artist can aid. Ever wonder how this happens? 

This is a technique called de-aging. It is a visual effects technique that makes an actor look younger, especially for flashback scenes. It is a common method used by filmmakers worldwide by digitally editing the image or using computer-generated imagery (CGI) overlays or touch-ups. Some movies create de-aged digital actors from scratch or with a mixture of stand-ins and CGI. The method can be used to make actors look older as well. 

X-Men: The Last Stand, The Curious Case of Benjamin Button, The Hobbit: The Desolation of Smaug, Ant-Man etc., are some familiar movies that have used the de-aging technique. Over the years, there has been an exponential use of digitally aged actors in film, TV series and advertising. 

Conventional method 

 There are primarily two methods used to digitally re-age actors. The first one is traditional 3D computer graphics pipeline which involves the process of rigging, animating and rendering. The second is through editing the actor’s face. The actor’s appearance, shot using the camera, is edited without requiring 3D geometry. 

 However, this process is time-consuming and requires honest effort from a VFX artist. Deep Learning algorithms were introduced to tackle the image-based re-aging problem. However, the results needed to be more convincing as many of those images were low resolution and did not satisfy the user. 

Disney’s AI 

 Understanding the struggles of conventional methods, Disney’s researchers found a new algorithm that produces high-resolution images under various situations and lighting. To ensure the result is as life-like as possible, artists can still make tweaks by hand, but the AI tool might do much of the work for them. 

 Disney’s FRAN (face re-aging network) is a neural network trained using a large database containing pairs of randomly generated synthetic faces at varying ages to avoid the need to find thousands of images of real people at different ages depicting the same facial expression, pose, lighting and background. 

How does FRAN work? 

 Each pixel to be re-aged is given an input and output age. FRAN takes this input as part of the 5-channel tensor. To construct the final re-aged image, the U-Net predicts per-pixel RGB deltas (offsets) that are superimposed on the top of the original image.  

The network may anticipate the re-aging output as RGB on top of the input picture, preventing severe loss of the input identity rather than learning how to produce faces of multiple identities. 

Analysing the results 

Applying FRAN on video frames yields reliable re-aging results, and it can adapt fluidly to changes in head attitude, lighting and depth of field. FRAN reliably and convincingly re-ages supplied images while preserving the target identity. 

 Compared to other algorithms, FRAN is superior in preserving the input identification and the specific skin detail of the given person. The real-world photographs perform nearly as well as simulated ones for FRAN’s processing. 

 According to Disney, FRAN is not anticipated to displace many industry jobs at the moment as manual VFX work and even actual prosthetic makeup application have limitations. However, the researchers found that FRAN wasn’t ideal for drastic changes like re-aging to and from extremely early ages. Also, greying of scalp hair wasn’t reflected when aging up an actor. 

 However, algorithms like this could help visual effects artists save time by reducing the work they must put into their projects. In addition, it might keep rising production costs in check and assist low-budget movies in artificially aging their performers.