Author: Perez Clark
This week saw social media users come across millions of AI-modeled depictions of themselves, all thanks to the Lensa AI app. With the service, users can provide a personal machine-learning model with their photos to recreate whimsical illustrations of their faces called “magic avatars.”
This article looks at the magic avatars, the companies behind them, data privacy on the app, racial and sexual challenges it currently faces, and the possibilities of stealing copyrighted art through the tool.
Let’s get started.
Keep Seeing AI-Selfies? Here’s All To Know About The Trend
What Are Lensa AI Selfies? How Are They Created?
Lensa relies on an open-source learning model named Stable Diffusion. Created by Stability AI, Stable Diffusion uses text-image pairs from the internet to generate avatars tailored to your text inputs.
Once you upload a picture, Lensa AI sends it to cloud storage and prompts a customized machine-learning model to make and recreate images in your likeness. According to Lensa AI’s Website, Stable Diffusion is trained using LAION5B, an enormous data set containing billions of image-text combinations.
On average, processing an image takes 10 minutes and requires over 120 million billion mathematical operations to analyze the source photos, the website reads.
How Was Stable Diffusion Created?
Stable Diffusion, the free-to-use machine model that Lensa uses, was initially released in August.
“The people responsible for training Stable Diffusion basically used billions of photos from the internet,” says Inger Yakon, cofounder and chief technology officer at Lightricks.
“They scanned the web and showed the model tons of images along with their text. As a result, Stable Diffusion can fantasize and generate new photos using a few data inputs,” Inger adds.
Should I Be Concerned About My Data Privacy
Prisma Labs, the company behind Lensa, says it permanently deletes the machine-learning model trained for your avatars and the photos you upload to generate them.
It also says data and insights generated from your photos aren’t shared with third parties. Although, its privacy policy leaves space for Lensa to do so in the future.
“Each time you purchase a new avatar pack, the creation process resets and begins afresh,” says Lensa on its website. “It is for this reason we prompt you to submit your photos each time you acquire a new package.”
Why Are Artists Warning Against AI-generated Portraits?
Ever since 2022’s upsurge in using Lensa AI’s avatar generator, there has been a heated internet debate regarding how it uses trademarked art.
For context, Lensa AI charges users to submit their photos and create magic avatars. That, undoubtedly, means it demands a fee on avatars created using image data sets from artists who didn’t give their permission.
Many digital artists now plead with the public to stop patronizing a service that uses copyrighted art and fails to compensate deserving artists.
“Our machine learning models recreate patterns learned from images,”Lensa AI states on its FAQ page. “Instead of replicating from the original dataset, it refers to the principles acquired in the learning process to provide your magic avatars.”
Inger from Lighttricks says AI portraits are the opposite of a threat to designers and creators. “I believe these machine learning models provide artistic power to creators, especially ones new to the space.”
“People now have the power to celebrate what other artists create. When you consider artists and their creations, you’ll realize that a vast majority of artworks are inspired by the productions of others,” Inger said.
Are Magic Avatars Possibly Racist and Misogynistic?
Another major issue against Lensa AI and similar AI selfie generators is that the images they produce are sometimes excessively sexual, race-preferential, and child inappropriate.
One user, for example, claims she uploaded her face images to Lensa and received several magic avatars characterizing her partly naked or with her cleavage exposed.
Other users with dark skin report seeing more glitches and abnormalities in their avatars compared to their light-skinned counterparts. Asian users, in addition, and women using hijabs, also complained more about avatars’ inaccuracies.
Lensa explains that its app and machine learning models rely on unfiltered web content to design the magic avatars. That means AI generators would naturally produce artworks that contain the biases we humans share.
It also claims that Stability AI, the company behind the stable diffusion model, has trained its model further, making it more difficult for the generator to produce racially and sexually aggressive avatars.
On NSFW avatars, Lensa says more progressions are being made to improve its app’s usability. However, the company advises parents to restrict minors from using the service.
“The entire industry’s goal is to completely eradicate gender or racial bias from our AI models,” Inger writes. “We’ve barely scratched the surface of this technology’s capabilities, and I’m sure it will only improve with time.”
Final Thoughts
Lensa AI’s app subscription charges $36 yearly, with an extra $3 to $12 for avatar packs. To get started with the app, submit a bunch of your photos that show your face clearly with no one in the background, wait a few minutes for your machine learning model to produce your avatars, and delight in the images of your face in your selected theme. But If Lensa AI’s art plagiarism concerns worry you, seeking a local artist, opting for a service that compensates traditional artists, or paying for a portrait from sites like Etsy are great alternatives.