A viral AI generator that lets users conjure up their own self-portraits has sparked new concerns over creators’ rights

If you ever feel hurt by the limited artistic offerings of street caricaturists, Lensa AI can take your discontent away. From cyberpunk Medusa to rococo garden nymph to Celtic gym warrior, the app outputs user-uploaded selfies via AI to conjure up highly stylized self-portraits. Costs? $3.99.

Lensa AI may not be new – it was launched in 2018 by Prism Labs – but the app rocketed to the top of the iOS App Store this weekend and has been downloaded 700,000 times in the past month. Its sudden popularity is due to the rollout of the Magic Avatars feature, which generates digital portraits based on user input.

To activate the feature, users simply upload 10 to 20 selfies (preferably with different backgrounds, face distortions and angles), indicate their gender (female, male, other) and check out by choosing between 50, 100 and 200 images, or sign up for a membership (or get 50 images for $3.99 as part of a “free” trial). The images arrive after 20 minutes.

The past year has seen the mainstreaming of AI imagers, largely driven by platforms such as DALL-E 2, Midjourney, and Dream Studio. Part of Lensa AI’s popularity rests on the simplicity of its user experience, but like its AI peers, the app renews its focus on privacy and ownership issues of algorithm-generated images.

First, there’s the wider issue that Magic Avatars are built using a dataset trained on the job with uncompensated performers. It then uses the submitted images to train its AI and, as stated in its privacy policy, users grant the company the right to “use, reproduce, modify, adapt, translate and create derivative works forever and royalty-free”.

Second, there’s the lack of nuance with which the AI ​​seems to treat the faces of people of color and the over-sexualized nature of many images uploaded by women.

Magic Avatars runs on Stable Diffusion, a free, open source image generator. Prisma Labs has created its own inputs (or prompts) to accompany the images it feeds into AI. Customers are essentially paying for the convenience of not having to work with software or code. Stable Diffusion itself has been trained on a dataset of more than two billion images, which are somewhat dubiously sourced from websites across the web, including Pinterest, Flickr, and Getty.

“This is not a filter or effect,” Prisma Labs explains via Instagram. “These AI avatars are generated from scratch, but with your face in mind.”

Follow Artnet News on Facebook:

Do you want to stay one step ahead of the art world? Subscribe to our newsletter to receive the latest news, eye-popping interviews and astute critiques that move the conversation forward.

Leave a Comment