CNET también está disponible en español.

Ir a español

Don't show this again

Christmas Gift Guide
Mobile Apps

Google's Allo morphs your selfies into custom emojis

Chat app's new feature taps neural networks and the work of artists to turn your selfie into a personalized sticker pack.

image2.gif

Put yourself in your emojis with Google's Allo.

Google

The next time you want to use an emoji to show your friends how you feel, you might think about actually showing them how you feel.

Google wants to help you do just that with a new feature for its chat app Allo that creates custom emojis using your face as its base. The auto illustration feature relies on neural networks and the work of artists to turn your selfie into a personalized sticker pack.

The new feature is the latest to come to Allo, which was the first product to showcase the Google Assistant, a digital helper offering automatically generated responses called Smart Replies and other computer-generated suggestions for your everyday life.

Turning selfies into emojis isn't as simple as it might sound. While the human eye can easily recognize qualitative features such as eye color, regardless of available light, it's no easy task for computers. So Allo turned to neural networks, a computer system modeled on the human brain that can learn from past actions to solve new problems without being specifically programmed to do so.

The aim was to avoid the reaction of creepiness people often feel when seeing at human replicas that appear almost, but not exactly, like real human beings. If you ever saw the 2004 animated film "The Polar Express," you are already familiar with this phenomenon, which is called "uncanny valley."

"In machine learning, this could be compounded if were confronted by a computer's perception of you, versus how you may think of yourself, which can be at odds," Jennifer Daniel, the expressions creative director for Allo, wrote in a blog post Thursday.

"Rather than aim to replicate a person's appearance exactly, pursuing a lower resolution model, like emojis and stickers, allows the team to explore expressive representation by returning an image that is less about reproducing reality and more about breaking the rules of representation," she wrote.

Google also worked with a team of artists to create illustrations that represent a wide variety of features. In one example, the artists designed a set of hairstyles they thought would be representative and used them to train the network to match the right illustration to the right selfie.

In the end, the feature offers more than 563 quadrillion combinations.

Virtual reality 101: CNET tells you everything you need to know about what VR is and how it'll affect your life.

Batteries Not Included: The CNET team shares experiences that remind us why tech stuff is cool.