Disney unveils neural network ‘re-aging’ software

The process of using digital effects to age actors is an increasingly common practice in Hollywood (despite many criticisms), but for every mysteriously young Robert De Niro IrishIt takes hundreds of hours of painstaking work from behind-the-scenes VFX artists and engineers. However, as neural networks become more common and accessible, entertainment industry heavyweights are finding the opportunity to leverage technology to streamline the process while simultaneously creating more photorealistic results. There are few entertainment companies as influential as Disney right now – so it’s probably not surprising that House of Mouse announced that it’s using cutting-edge AI to “re-age” actors.

[Related: Disney is turning to robots to pull off dangerous aerial feats.]

Yesterday, Disney Research Studios unveiled the Face Re-Aging Network (or FRAN), its latest advancement in VFX work that uses neural networks to create “the first practical, fully automated, production-ready method to re-age faces in video footage.” According to Disney’s own account, at least 12 movies and television series in 2022 used re-aging technology; this number has grown steadily since its debut just a few years ago. “Photorealistic digital re-aging of faces in video is becoming more and more common in entertainment and advertising. But the dominant 2D painting workflow often requires manual work, frame-by-frame, which can take days, even by skilled artists,” the team explains in the summary of their research report. FRAN’s solution to this problem relies on a multi-step neural network process that reduces labor time and costs while increasing the realism of the effect.

Watch Disney’s video synopsis below:

Aspect Gizmodo and Ars Teknik Described earlier today, FRAN uses a series of steps to create its re-aged subjects. First, the team used a program called StyleGAN2 to randomly generate thousands of synthetically aged faces aged 18 to 85 years. After this database was created, machine learning tools aged and re-aged these artificial portraits; another neural network, FRAN. FRAN can apply what it learns to videos of real people by adding or subtracting years regardless of angle, position, movement or lighting. Artists can then manually retouch frames and transitions as needed.

[Related: Machine learning can measure mood on social media.]

The results are still not perfect, but arguably as persuasive as those seen in movies like this: rogue one or Ant-Man and the Wasp. At least, the end products will only evolve over time as they get smarter with neural network techniques like FRAN.

Leave a Reply

Your email address will not be published. Required fields are marked *