ipopba - stock.adobe.com
This story was updated Dec. 20.
For many people, paying for the Lensa AI image editing app to generate "magic" avatar images of themselves is about following a cool trend on social media.
For author and podcast host Elizabeth Leiba, creating an avatar of herself with the controversial, widely used new app was more than a trend. It was about representation.
Risk versus benefit
For Leiba, the representation Lensa provides is worth the privacy risk of handing over facial data to a company that can then profit from that data.
"My images are across social media," Leiba said, adding that she has been using tools such as Facebook, Instagram and Twitter since their inception. "We know that already information about us is constantly being collected and being used and misused. So even though I think about that, it's almost like a risk-reward calculation."
She said the reward of seeing herself represented in a different light as a Black woman balances the risk associated with the app.
As the author of the newly released book I'm Not Yelling: A Black Woman's Guide to Navigating the Workplace, Leiba is aware of the lack of representation for Black people in corporate America, on social platforms and in the general media.
"Black folks in general, we haven't seen ourselves reflected in art," Leiba said. "If you think about the history of art and the legacy of art, it typically has always been framed through the lens of the majority and the Western lens of beauty."
Over the past month or so, avatars created by Lensa have flooded social media pages. While Lensa has been around since 2018, Lensa's parent company, Prisma Labs, introduced the Magic Avatar feature last month.
The feature enables users to turn themselves into fairies, princesses, astronauts or even anime characters for as low as $3.99. The app also offers membership and other options. To make the images possible, Lensa uses a text-to-image deep learning AI-generating tool called Stable Diffusion. Stable Diffusion is trained on data publicly found on the web.
While many users found Lensa cool and alluring, it didn't take long for critics to find problematic details about this new AI trend.
The fine print
A key problem is that millions of people jumped on the internet meme to provide Lensa with their face without reading the details about how their data would be used.
"People are turning over, in many instances, more than five, perhaps even 10, if not more, images of themselves," said Michael Bennett, director of the education curriculum and business lead for responsible AI at the Institute for Experiential AI at Northeastern University. "Meanwhile, the policies, terms and conditions are very aggressive in terms of the company."
Prisma Labs updated its terms and conditions to provide clarity and reassurance to customers, a representative wrote in a statement to TechTarget Editorial on Tuesday.
The company noted that users' images are used "solely for the purpose of [users] creating their own avatars." Prisma Labs also shares the data with cloud providers such as Google Cloud Platform and Amazon Web Services for the same purpose. User data is "automatically deleted within 24 hours after being processed by Lensa," according to the company's statement.
"Criticism aside, 'democratization of access' to cutting-edge technology like Stable Diffusion -- which is now packaged in the shape and form of an app feature -- is quite an incredible milestone," the statement read. "What was once available only to techy well-versed users is now out there for absolutely everyone to enjoy. Limited technical skills are required."
Elizabeth LeibaAuthor and podcast host
Prisma Labs updated its terms and conditions on Dec. 15 to describe the rights granted as "time-limited" and "revocable" in an apparent concession to critics of the previous terms.
What Prisma Labs is saying to users is that the company has the right and ability to transfer these rights or license them to another company without paying for their subsequent use, Bennett said.
"They could certainly be paid by another entity for access to the data," he added. "Folks are probably going to be in for a rude awakening if they decide later that they want not to be involved in whatever ventures the parent company decides to enter into using their data."
While one could argue that Prisma Labs needs to be more transparent about its policies or that users need to be more educated, Bennett said this is more a case of people needing to do more due diligence and inform themselves about the app details.
"The policy is right there," he said. "If people are willing to read, they can learn what a parent company can do with that data if they turn it over."
"This language is not exclusive to Lensa," said Chirag Shah, professor at the Information School at the University of Washington. "This is what we always sign up for. It's just that in some cases, the potential for harm may not be as much as in others."
A space for representation
Meanwhile, when Leiba saw posts from Black counterparts posting pictures that Lensa generated of Black people as superheroes and fairies, she was intrigued.
"We've never seen ourselves as some of the artwork you see generated by AI," she said. "A lot of Black folks are leaning into using this app, and we are already exploited. We are already worried about just moving around the world in our regular skin.
"So, looking at ourselves reflected in an app in a way that's magical, in a way that really reflects how we feel about ourselves inside, and getting to see that image is something that's priceless," Leiba said.
But AI-generated art only addresses some of the problems of representation.
Many Black Americans who used Lensa also complained that the app generated images that changed their skin color.
Leiba also had that same problem and chose not to upload those images that produced results that made her look white, she said. The problem for her goes to the root of those creating the AI algorithms.
"I think that's something that's a part of the discussion that needs to be had," she said, adding that because there is a sense that the majority culture in the U.S. is white, then most of the art generated by AI would create that.
"If most of the people who work in AI and technology happened to be from that majority culture, then they're going to, in their mind, probably gravitate toward images or toward an algorithm that creates images that are representative of that," Leiba continued. "That's problematic as well."