Google recently published a blog post describing a new feature in their Arts & Culture app. The app is called “Art Transfer,” and when used on your smartphone, allows you to take a photograph (or use an existing photo in your library) and “apply the characteristics of well-known paintings to your own images.”
After taking or uploading a photo, one can choose from numerous different masterpieces by famous artists and the app will transfer that particular artist’s style onto your image. They explain specifically that “… Art Transfer is powered by an algorithmic model created by Google AI. Once you snap your photo and select a style, Art Transfer doesn’t just blend the two things or simply overlay your image. Instead, it kicks off a unique algorithmic recreation of your photo inspired by the specific art style you have chosen.”
In theory, this sounds pretty cool, but how good – and how accurate – is it? I used a photographic “self portrait” and put it to the test. Google does us a favor and imprints text as to which artist and which masterpiece is being integrated into the algorithm that alters my photo. Here’s a handful of results from my first go with this new app:
My first pair includes use of the styles of two of my favorite artists: Andy Warhol and Jean-Michel Basquiat. What do you think? The Warhol image (below left) uses this painting as the basis of its algorithm; the Basquiat image uses this painting … and to be honest, maybe I’m expecting too much, but I don’t think my “inspired by” self portraits resemble the respective styles of these artists at all.
Next up, I ran the same photograph through their algorithm for Vincent Van Gogh (below left) and Edvard Munch. These get a little closer, although in my opinion, it’s primarily the colors that work well as opposed to any accurate mimicking of brushstrokes or painterly style.
The following two are based on works by Yayoi Kusama and Keith Haring. The specific Kusama art work doesn’t lend itself to a portrait, but personally, I think this is my favorite of the algorithmic interpretations so far … I like how my face is made out of Kusama dots. The Haring piece doesn’t feel like Haring at all to me – I wish had been transformed into one of his cartoon-like characters, rather than these random markings.
The last pair that I tried was using Frida Kahlo and Roy Lichtenstein. Again, I don’t think these interpretations bring to mind the original artists, although I do enjoy seeing them as new digital art works. The Lichtenstein version actually makes me think of Italian Futurist painting more than it does American Pop Art, but so be it.
I had fun doing this, and will probably use this app some more. Maybe I’ll try using photographs without the goal of mimicking the famous artist, but rather seeing what new and unusual things I can come up with. If you have any thoughts or feedback on the intersection of algorithmic AI (artificial intelligence) and fine art, please share in the comments section below.