Lensa AI, the app that creates “magical avatars”, raises red flags for artists

If your Instagram is flooded with algorithmically generated portraits of your friends, you’re not alone. After adding a new Stable Diffusion-based avatar generation tool, photo-editing app Lensa AI has gone viral in recent days, with users sharing their bizarre AI-created avatars (and gruesome duds) in stories and publications.

Lensa’s fun and eminently shareable avatars mark the first time many people have interacted with a generative AI tool. In Lensa’s case, it’s also the first time they’ve paid for computer-generated art.

Stable Diffusion itself is free and many people play with it for research or just for fun. But Lensa and other similar services — Avatar AI and Profilepicture.AI, to name a few — make money by selling the computer cycles needed to run prompts and spit out a set of images. It certainly changes the equation a bit.

Lensa is built on Stable Diffusion’s free and open source image generator, but acts as a middleman. Send Lensa 10-20 selfies and $7.99 ($3.99 if you sign up for a free trial) and the app does the heavy lifting for you behind the scenes, returning you a set of stylized portraits in a range of styles like sci-fi, fantasy and anime. Anyone with enough processing power can install Stable Diffusion on a machine, download some templates, and achieve similar results, but Lensa’s avatars are impressive and Instagram-ready enough that many people are more than happy to pay for convenience.

While the tech world celebrated advancements in AI image and text generators this year — and artists watched the proceedings with caution — your average Instagram user probably didn’t start a philosophical conversation with ChatGPT or fed nonsensical DALL-E prompts. It also means that most people haven’t grappled with the ethical implications of free and readily available AI tools like Stable Diffusion and how they’re about to change entire industries – if we let them.

In my weekend experience on Instagram, for every 10 Lensa avatars, there’s a Cassandra in the comments berating everyone for paying for an app that steals artists. These concerns are not really exaggerated. Stable Diffusion, the AI ​​image generator that powers Lensa, was originally trained on 2.3 billion captioned images – a massive sample of the visual internet. There’s all sorts of stuff in there, including watermarked images, copyrighted works, and a huge amount of images from Pinterest, apparently. These images also include several thousand photos pulled from Smugmug and Flickr, illustrations from DeviantArt and ArtStation, and stock images from sites like Getty and Shutterstock.

Individual artists have not opted out of appearing in the practice dataset, nor can they opt out. According to LAION, the nonprofit organization that created the huge datasets to begin with, the data treasures are “simply indexes on the Internet,” lists of URLs to images on the Web associated with the alt text that describes them. If you are an EU citizen and the database contains a photo of you with your name attached, you can file an opt-out request under GDPR, Europe’s groundbreaking privacy law, But that’s about all. The horse has already left the barn.

We’re in the early stages of figuring out what this means for artists, whether they’re freelance illustrators and photographers or huge copyright-conscious corporations getting sucked into the AI modeling process. Some models using Stable Diffusion take the problem even further. Prior to a recent update, Stable Diffusion Version 2, anyone could create a model designed to emulate the distinct visual style of a specific artist and create endless new images at a rate no human could match. .

Andy Baio, who co-founded a festival for independent artists, posted a thoughtful interview on his blog to dig deeper into these concerns. He spoke with an illustrator who discovered an AI model specifically designed to replicate his work. “My first reaction was that it was invasive to have my name on this tool,” she said. “…If someone had asked me if they could do that, I wouldn’t have said yes.”

Back in September, Dungeons & Dragons artist Greg Rutkowski was so popular as a stablecast guest used to generate images in his detailed fantasy style that he feared his real art would be lost in the sea of ​​algorithmic copies. . “And in a year? I probably won’t be able to find my job there because [the internet] will be awash with AI art,” Rutkowski told MIT’s Technology Review.

These concerns, echoed by many illustrators and other digital creatives, are resounding on social networks because many people are encountering these thorny issues – and the existential threat they seem to pose – for the first time.

“I know a lot of people have been posting their Lensa/other AI portraits lately. I would encourage you not to, or better yet, not to use the service,” comedian Jenny Yokobori wrote in a post. popular tweet thread about Lensa. In another, Riot Games artist Jon Lam share his own discomfort with AI-generated art. “When AI artists steal/co-opt art from us, I don’t just see art, I see people, mentors and friends. I don’t expect you to understand.

Personally, I was sick and stuck at home over the weekend, where I spent more time than usual scrolling through social media. My Instagram stories were a blur of flattering digital illustrations that cost pennies apiece. Lensa clearly tapped into something special there, appealing to both the vain impulse of effortlessly collecting 50 stylish self-portraits and the interactive experience of quizzing your friends on what your spitting image is (most, in my experience) and what are the hilarious mutations that only a computer doing its best to pretend to be a human could start.

Some friends, mostly artists and illustrators, pushed back, encouraging everyone to find an artist to pay instead. Some creative people around me also paid and it’s hard to blame them. For better or worse, it’s truly amazing what the current cohort of AI image generators can do, especially if you just logged on.

Soon we will all be paying attention. In the name of storytelling and vanity, I downloaded Lensa and gave the app a try. I had only paid once for an artist to do a profile picture for me in the past and it was just one picture, all the way back in 2016. Now for less than $10, I had a set of 50 epic avatars generated from my most me pictures, but these were in addition me. Me in various futuristic jumpsuits straight out of the pages of a graphic novel, me in purple robes resembling an intergalactic saint, me, me, me.

I see the call. A handful of friends commented on how the footage made them feel, hinting at the genre euphoria of being seen as they see themselves. I wouldn’t blame anyone for exploring this stuff; everything is very interesting and at least so complicated. I love my avatars, but part of me wishes they didn’t. I don’t plan to use them.

I thought about my own art, the photography that I sell when I think about stocking my online store – mostly mountain landscapes and night sky photos. I thought back to a handful of prints I sold and how much effort I had to put in to get the shots. One of my favorite shots involved special permission from the National Park Service and a five-hour backpacking trek to a remote fire watcher in Washington. Many involved lonely hours tending my tripod alone in the freezing cold, tracking the Milky Way as it circled above a dark horizon like the clock hand.

The AI ​​models already have enough training material to faithfully recreate photos of a nearby hidden mountain location that only local night photographers seem to know about. Three years ago when I took pictures there, I had to snag a competitive campsite and drive miles on a rutted Forest Service road waiting in the dark for hours. I cooked a pack of expired ramen noodles with a small camping stove to keep warm, stuffed the feathers back into my jacket, and jumped on anything that made noise in the dark.

I don’t live from my art. But it’s still like a waste to think that these experiments and the human process they represent – learning to predict a cluster of ancient stars in the dark, slipping on wet rocks and chipping my shoe, keeping extra batteries warm in a down pocket – might be worth less in the future.