03 Real Tone

For enabling more realistic, inclusive imagery

Years before the 2020 tidal wave of global anti-racism protests prompted companies to scrutinize their offerings through a lens of equity, Google had quietly launched a project to make smartphone images more inclusive. 

“Going back decades, cameras have centered light skin—a bias that’s crept into many of our modern digital imaging products and algorithms, especially because they’re not being tested with diverse enough groups of people,” says Florian Koenigsberger, the company’s image equity lead. “We knew that building for the community meant we had to acknowledge our own gaps and learn from the folks who know this issue best.” 

The result? Real Tone, an AI-powered post-processing software that Google rolled out in October 2021—integrating what it dubbed “the world’s most inclusive camera” into its Pixel 6 and Pixel 6 Pro phones. 

First launched in 2017, the project brought together a roster of image experts: photographers, cinematographers and multidisciplinary artists and colorists hailed for their stunning and accurate representation of people of color—including Adrian Octavius Walker, Dana Scruggs, Deun Ivory, Kennedi Carter, Kristian Mercado Figueroa and Zuly Garcia. The experts tested Google’s cameras in a wide range of tough lighting conditions through thousands of portrait shots, helping to retrain and re-engineer Google’s face-detection software and improve its performance in a wider spectrum of lighting conditions.

“One of the big lessons in this process is that if you want to build something for someone, you will always do a better job of that if you build it with them,” Koenigsberger says. “Real Tone is the expression of a lot of different technologies and tuning adjustments to make sure that people of color are showing up as they are and as they ought to be seen in our camera and imaging tools.”

The project radically increased the diversity of Google’s image data sets—by a factor of 25. The updated auto-white balance and improved auto-exposure ensure a subject’s skin tone actually looks like them—not unnaturally brighter or darker. Project engineers also developed an algorithm to reduce the effects of stray light, preventing darker skin tones from looking ashy or washed out.

“It’s important to think of images as facts,” says Carter, an artist and photographer. “Whether it’s an image of your family and your grandma cooking dinner or it’s something like a protest, this will be a point of reference for a very long time.” 

In May, Google expanded the reach of Real Tone, incorporating its 10-shade skin tone scale (developed in partnership with Harvard professor Ellis Monk) into Real Tone filters, Google Photos, Google Search and other products and services. The scale will train AI models to build more representative datasets, and Google has publicly released it for other R&D teams to use. 

“Photos are symbols of what and who matter to us collectively,” Koenigsberger says. “So it’s critical that they work equitably for everyone.”