If she ever abandons acting, Kristen Stewart might have a gig in artificial intelligence. The Twilight actress has just published a research paper on a machine learning technique called "neural style transfer" and its use in film, reports Quartz. Style transfer is a "fun technique," according to Tech Crunch, though it's a bit complicated to explain. Essentially, an algorithm is used to apply the style of one thing to something else. In this case, Stewart—with help from Starlight Studios producer David Shapiro and Adobe research engineer Bhautik J Joshi—sought to transfer the look of one of her own paintings to scenes from her short film Come Swim.
Using computer programs known as neural networks, the group was able to add "blocks of color and texture" to a test frame until it matched the look of the painting, which depicts "a man rousing from sleep," reports the Verge. Once they had achieved the look they wanted on the test frame, they applied the same technique to the opening and closing scenes of Come Swim, described as "a poetic, impressionistic portrait of a heartbroken man underwater," which premiered at the Sundance Film Festival on Thursday, per the Guardian. The Verge notes you might have used this same technique yourself: Facebook introduced a style transfer filter last year.