What to know about Lensa, the AI portrait app all over social media

CTV National News: Privacy concerns over Lensa John Vennavally-Rao reports on the privacy concerns of Lensa and other apps that use AI technology to generate images in various styles.

If you've logged on to any social media app this week, you've probably seen pictures of your friends, but re-imagined as fairy princesses, animé characters, or celestial beings.

It is all because of Lensa, an app which uses artificial intelligence to render digital portraits based on photos users submit.

Lensa's highly stylized, eye-catching portraits have taken over the internet, but they have also been the subject of concern from privacy experts, digital artists, and users who have noticed the app making their skin paler or their bodies thinner.

Here's everything you need to know about Lensa:

How to get your own 'magic avatar'

The pictures making the rounds online are products of Lensa's "Magic Avatars" function. To try it out, you'll need to first download the Lensa app on your phone.

A yearlong subscription to the app, which also provides photo editing services, costs $35.99. But you can use the app for a weeklong free trial if you want to test it out before committing.

Generating the magic avatars requires an additional fee. As long as you have a subscription or free trial, you can get 50 avatars for $3.99, 100 for $5.99, or 200 for $7.99.

Lensa recommends users submit 10-20 selfies for the best results. The pictures should be close-up of your face with a variety of different backgrounds, facial expressions, and angles. Lensa also stipulates it should only be used by people who are 13 and older.

Lensa is a product of Prisma, which first reached popularity in 2016 with a function allowing users to transform their selfies into images in the style of famous artists.

The app explains in its privacy policy they use TrueDepth API technology, and the user-provided photos, or "face data," are used "to train our algorithms to perform better and show you better results."

We tested out the app to see what it's like

To test out the app, I curated 20 selfies I thought showed off a variety of expressions and angles and chose the 100 avatar option. It took around 20 minutes for Lensa to return my avatars, which fell into 10 categories: fantasy, fairy princess, focus, pop, stylish, animé, light, kawaii, iridescent, and cosmic.

In general, I felt like the app did a decent job producing artistic images based on my selfies. I couldn't quite recognize myself in most of the portraits, but I could see where they were coming from.

It seemed to recognize and repeat certain features, like my pale skin or my round nose, more than others. Some of them were in a more realistic style, and were close enough I might think they were actually photos of me if I saw them from afar. Others were significantly more stylized and artistic, so they felt less specific to me.

For some women, the app produces sexualized images

One of the challenges I encountered in the app has been described by other women online. Even though all the images I uploaded were fully-clothed and mostly close-ups of my face, the app returned several images with implied or actual nudity.

In one of the most disorienting images, it looked like a version of my face was on a naked body. In several photos, it looked like I was naked but with a blanket strategically placed, or the image just cut off to hide anything explicit. And many of the images, even where I was fully clothed, featured a sultry facial expression, significant cleavage, and skimpy clothing which did not match the photos I had submitted.

This was surprising, but I'm not the only woman who experienced it. Olivia Snow, a research fellow at UCLA's center for critical internet inquiry and professional dominatrix, told CNN the app returned nude images in her likeness even when she submitted pictures of herself as a child, an experience she documented for WIRED.

Snow said artificial intelligence technology like the one Lensa uses could be used to generate "revenge porn," i.e., making naked images of someone without their consent.

For Snow, the output was a sign of the "complete lack of content moderation" on the app. She also called for greater regulation of AI apps like Lensa.

Lensa did not reply to a request from CNN to comment on the app producing nude or sexualized images.

Other users have documented different forms of bias produced in their Lensa images, like Black users who are "whitewashed" and shown as paler than they actually are. Similarly, Aubrey Gordon, a writer and fat rights activist, wrote on her verified Instagram the app produced images which made her look much thinner than she actually is.

"Lensa is really working overtime to make AI me into a thin person," she wrote in the caption.

Digital artists say the app co-opts their work

Lensa's technology relies on a deep learning model called Stable Diffusion, according to its privacy policy. Stable Diffusion uses a massive network of digital art scraped from the internet, from a database called LAION-5B, to train its artificial intelligence. Currently, artists are unable to opt-in or opt-out of having their art included in the data set and thus used to train the algorithm.

It has raised concerns from some artists, who say Stable Diffusion relies on their artwork to make their own images, but they are not credited or compensated for their work. Earlier this year, CNN reported on several artists who were upset when they found their work had been used without their consent or payment to train the neural network for Stable Diffusion.

Several of the artists expressed concern the apps could also threaten their livelihoods. Digital artists cannot compete with the low prices and fast turnaround artificial intelligence enables for a digital portrait, they said at the time.

Lensa's owner Prisma has tried to assuage concerns about their technology eliminating work for digital artists.

"Whilst both humans and AI learn about artistic styles in semi-similar ways, there are some fundamental differences: AI is capable of rapidly analyzing and learning from large sets of data, but it does not have the same level of attention and appreciation for art as a human being," wrote the company on Twitter on December 6.

And "the outputs can't be described as exact replicas of any particular artwork."