You’ve probably already heard of Prisma, the app that turns photos into paintings (like the ones that flooded our news feeds all last summer), but now it looks like there’s something new on the horizon that will be able to do the complete opposite.
As French website Numerama reports, researchers in Berkeley have just created an AI program called CycleGAN that can turn a painting into a photo.
If you ever dreamed of seeing the actual places Claude Monet or Paul Cézanne painted, now you can. The program can take any painted landscape and turn it into a photo of the place depicted in the image, like so:
And it doesn’t stop there: the AI is also apparently able to do things like change oranges into apples and horses into zebras (see photos below). While you might be wondering what the point of all this is, this type of technology opens up a whole new world of possibilities for photo editing software.
As Engadget explains, this type of tool allows you to create a very weak depth of field, so you can add blur and retouch the backgrounds of your images in post.
What’s impressive is that the app works both ways. You can change a painting into a photo, or a photo into a painting. With this double function, the researchers have worked to develop what they call “cycle consistency.”
If you transform a painting into a photo and then convert the same photo back into a painting, the result should be very close to the original artwork (kind of like when you translate from English to French, then from French back into English for fun).
The team in charge of the project explained that, for the moment, the app is relatively new and still makes mistakes. But certain results have proven to be particularly promising.
There seem to be a lot of AI projects popping up that can automatically create pictures from large databases (like the image translation tool that turns drawings into photos), so it seems like these types of tools will only get better from here.