This is amazing. The researchers at UC Berkeley who came up with the automatic colorizer algorithm we first shared back in March, 2016 have released a major update. The software now lets you team up with the algorithm to colorize complex black and white photos accurately in seconds.
Both the original app and this update were created by Richard Zhang and his team at UC Berkeley. But while the original often got many of the colors wrong, the update lets you give the algorithm “hints” so that it knows what color goes where.
The fact of the matter is, the neural network-powered algorithm is really good at applying color to black and white photos, but it’s often bad at guessing what color any particular item of clothing, inanimate object, or even person should be. Letting you play “middleman” in this scenario is a simple solution to this problem.
Here’s a demo of the new program in action. As you can see, the app starts by trying to colorize the image itself, and then it gives you a list of “suggested colors” you can use to help correct mistakes. As you add markers, the software continues to update the image, following your instructions to create something (hopefully) more accurate:
Here are a few examples from the UC Berkeley website, showing how Zhang’s team colorized some iconic black and white photos from history.
On the left is the original, black and white photo; in the middle is the photo plus some points of color added by Zhang and his team, telling the computer what shade that area should be; and on the right we have the final colorized photograph:
Of course, you can also mess with the algorithm, giving it “bad hints” as it were and watching the computer colorize a pink beard onto Earnest Hemingway… for example:
Fully automatic colorization feels like science fiction—something we will probably see SOME day in Photoshop, but not someday soon. This update application, on the other hand, feels like it could be included in the next version of the software.
Black and white photo + a few colored dos = fully colorized image. That would be a game changer.
We hope Adobe is listening, and we hope they’re working on this right now. But even if they’re not, you can download Zhang’s creation from Github and give it a shot for yourself.
Image credits: All photographs via Richard Zhang/UC Berkeley.