Facebook and Google's Battle to Make Cool Video Filters Rages On 

Mark Zuckerberg / Facebook

Facebook CEO Mark Zuckerberg teased on Wednesday a new artificial intelligence project that turns video into animated portraits like you might see in The Wizarding World of Harry Potter. While the videos aren’t sentient — yet — they’re the latest example of how A.I. could be used in the near future to make everyone an artist.

Zuckerberg explained that the video is created using an A.I. technique called “style transfer.” It works by taking a subject, like the family dog, and applying a recognizable artistic style, like impressionism, to it. The end result of this process is an image or video that depicts a unique moment in time with the characteristics of a timeless piece of art.

“Looking forward to getting this in your hands soon!” Zuckerberg said.

This is the video.

In a separate post, Facebook director of A.I. research Yann LeCun explained that the style transfer tech “works on video in real time on your phone” through “deep learning methods” that have been optimized for mobile devices.

“More information on this will appear over the next couple of weeks,” LeCun writes in the post.

This type of A.I. isn’t unique to Facebook. A smartphone app called Prisma has a similar effect when it’s used on photos and videos, although it’s not applied in real-time. Researchers at Google also announced Wednesday that they’re working on style transfer tech.

Google also chose a dog to show off how its tech works:

Style transfer works by teaching A.I. the features of a particular art style, and telling it to apply the same effect to new content. Google’s gone a step further by creating A.I. that can learn from multiple styles, which will allow people to mix-and-match different effects to create something new. At that point, creativity will no longer be constrained by talent.

“Our method enables style interpolation in real-time as well, allowing this to be applied not only to static images, but also videos,” reads the Google blog post.

These techniques will also help Google’s Magenta Project learn how to create new art. A.I. learns best when it has access to many different sources — encouraging people to make art via style transfer technology could provide that inspiration.

Whether A.I. uses those abilities for good or for, say, a website that generates horrifying faces is up to its creators. (At least for now.) But currently, advancements in A.I. and other computer graphics tech lend themselves to making everyone a decent artist.

Later on Wednesday, Zuckerberg uploaded three more videos that showed the A.I.-powered video filters in action. This time he used his office as the subject, and said that he was “coding [his] home A.I., in videos drawn by another A.I.” Welcome to the recursive world of A.I. art.

Related Tags