Lensa AI and the Non-Problem AI Art Tries to “Solve”
This article was originally published on LinkedIn.
Lensa AI has held the №1 spot on the iOS App Store’s “Photo & Video” chart for over two weeks now. To put that into perspective, Instagram is №3 and Snapchat is №5 on the chart at the time this article went live. But the proof isn’t in the chart. It’s on your social media feeds. If you use social media at all, you’ve come across your friends’ posts of algorithmically produced portraits.
Although the photo editing app has been around since 2018, Lensa AI’s rise to prominence is the result of its “Magic Avatars” update, which launched just last month. By uploading 10 to 20 images of your face and forking over a few dollars, this new feature generates up to 200 portraits of you in a variety of styles.
Of course, Lensa AI achieving viral status, and its underlying AI technology, is not without controversy. Does the app steal artists’ work to train its AI model? Why does the app have a habit of sexualizing women? What does Prisma Labs, the developers of Lensa AI, do with your selfies?
Let’s start with a slightly easier question:
How Does Lensa AI Work?
While you’re paying Lensa AI, these “Magic Avatars” aren’t generated by a proprietary AI. Instead, you’re paying Prisma Labs to generate portraits using Stable Diffusion, an open-source machine learning model.
Stable Diffusion is trained on billions of publicly-available images. The way Stable Diffusion is trained is why it can mimic dozens of artistic styles, from sci-fi and fantasy to photorealistic and anime. In extremely simple terms, Lensa AI generates your portrait by combining your photos with art that already exists, though hundreds of images may contribute to a single resulting portrait.
Stable Diffusion may be trained on publicly-available images, but these images are sourced without consent. The Large-scale Artificial Intelligence Open Network (LAION) database that trains Lensa AI’s underlying AI model also contains stolen artwork, photoshopped celebrity porn, medical photos, and even stolen nonconsensual porn. There is no way for anyone to opt out of the dataset. In fact, there is no way to even know whether your image has been scraped and added to the dataset.
What Does Lensa AI Mean for Art?
AI models can be helpful creative tools. Tools like Stable Diffusion can generate reference images for visual artists or help writers visualize scenes in their manuscripts. However, the fact remains that Stable Diffusion produces images that contain traces of artists’ styles and even their signatures.
While the art industry is subject to stringent copyrights laws in the United States, the use of copyrighted material in AI remains legally uncertain. In fact, using copyrighted materials to train AI models has been found to fall under fair use laws. As a result, even when AI-generated art bears striking similarities to an artist’s work, if they don’t look identical, the AI art tool is probably in the clear.
The internet democratized art by providing new avenues for artists to promote their work and connect with other artists. For many artists, posting online and maintaining a social media presence is essential for landing commissions and reaching new fans and patrons. In other words, artists are more likely to find success by making a portfolio of work samples publicly available.
That’s what makes apps like Lensa AI so insidious and exploitative. These apps are powered by AI models trained on images scraped from sites like DeviantArt, Getty Images, and Pinterest, sites artists rely on to publicize their work.
On the bright side, some big players are taking a more cautious approach. For instance, Getty Images banned the upload and sale of AI-generated illustrations. Shutterstock plans to launch a Contributor Fund to compensate artists when content they uploaded ends up in an AI’s training dataset.
How Does Lensa AI Use Your Data?
While Prisma Labs’ privacy policy states that it “do[es] not use photos or videos you provide . . . for any reason other than to apply different stylized filters or effects to them,” the company’s terms of use tells a different story:
Solely for the purposes of operating or improving Lensa, you grant us a perpetual, irrevocable, nonexclusive, royalty-free, worldwide, fully-paid, transferable, sub-licensable license to use, reproduce, modify, distribute, create derivative works of your User Content, without any additional compensation to you and always subject to your additional explicit consent for such use where required by applicable law and as stated in our Privacy Policy.
Prisma Labs’ terms of use go on to define what it refers to as its “Company License”:
The Company License is for the limited purpose of operating Lensa and improving our existing and new products, including but not limited to training Lensa’s AI within your Use of Magic Avatars feature of the application or if otherwise implied by the Using of Lensa and its services, unless you have provided us your additional explicit consent for the different purpose where required by applicable law.
TLDR: you’re helping Prisma Labs train its AI.
Although your photos are deleted, they are converted into “face data” to train the AI model powering Lensa AI. Information like facial feature position and orientation and even face topology, are retained so that the app can improve the way it identifies those characteristics in the future.
The company claims the photos you upload are automatically deleted within 24 hours, but if you want to play it safe, you can send a request to delete all your personal data to: privacy@lensa-ai.com. Note that, at least according to the terms of use, Prisma Labs isn’t obligated to delete your data upon your request. If you upload any of the “Magic Avatars” portraits to social media, you’ve also given Prisma Labs permission to use them in advertising. To rescind that permission, you can email: contact@lensa-ai.com.
What Should You Do Instead of Using Lensa AI?
Lensa AI generates portraits in seconds, but does so by scraping the internet for examples of art that were born from years of practice. AI art is appealing because it is cheap and fast. However, art is often neither of those things. Quality paint, software, and cameras are expensive. Artists tend to create their best work when given the time to do so.
The problem with AI art tools like Lensa AI is they try to solve a problem that doesn’t exist. There is already so much art out there and it’s largely undervalued. After all, artists are infamously and egregiously underpaid and underemployed (please refer to the “starving artist” trope).
As I slowly get off my soapbox, please remember to support artists. They may take longer to create your portrait, but they deserve your support more than an app (and they won’t keep your face data).