If your Instagram is overflowing with algorithmically-generated portraits of your friends, you’re not alone. After adding a new avatar creation tool based on Stable Diffusion, photo editing app Lensa AI has been gaining popularity over the past few days, with users sharing weird AI-created avatars (and horrific true stories) in stories and posts.
Lensa’s fun and eye-catching shareable avatars are the first time many people have interacted with generative AI tools. For Lensa, this is also the first time that money has been paid for computer-generated artwork.
Stable Diffusion itself is free and many people play with it for research purposes or just for fun. However, Lensa and other services like it (such as Avatar AI and Profilepicture.AI) make money by selling the compute cycles required to fire prompts and spit out a series of images. That definitely changes the equation a bit.
Lensa is based on Stable Diffusion’s free and open source image generator, but acts as an intermediary. Send in a Lensa 10-20 selfie and $7.99 (or $3.99 if you sign up for the free trial), and the app will do the heavy lifting for you in return for a set of stylized portraits in a variety of styles, such as sci-fi, fantasy, and more. and animation. Anyone with enough processing power can install Stable Diffusion on their computer and download some models to achieve similar results, but Lensa’s avatar is impressive and ready for Instagram, so countless people are willing to pay for the convenience.
The tech world celebrated the advances of AI image and text generators this year, and artists watched the proceedings carefully, but the average Instagram user probably didn’t start a philosophical conversation with ChatGPT or give DALL-E absurd prompts. It also means that most people haven’t wrestled with the ethical implications of readily available, free AI tools like Stable Diffusion, and how they’re poised to transform entire industries if we let them.
My experience over the weekend on Instagram is that for every 10 Lensa avatars, in the comments there is Cassandra berating everyone for paying for an app to steal from the artist. These concerns are in fact not exaggerated. Stable Diffusion, the AI image generator powering Lensa, was originally trained on 2.3 billion captioned images, a vast slice of the visual internet. All of this includes all sorts of things, including watermarked images, copyrighted works, and tons of photos from Pinterest. These images include thousands of photos from Smugmug and Flickr, artwork from DeviantArt and ArtStation, and even stock images from sites like Getty and Shutterstock.
Individual artists have not opted in to appear in the training dataset and cannot be opted out. According to LAION, a nonprofit organization that has created massive data sets, the treasure trove of data is “simply an index to the internet,” a list of URLs to images on the web paired with alt text that describes them. If you’re an EU citizen and our database contains a picture of you with your name attached, you can file a takedown request under GDPR, Europe’s landmark privacy law, but that’s all. The horse has already left the barn.
We’re in the early stages of grappling with what this means for artists, whether it’s independent illustrators and photographers caught up in the AI modeling process, or large copyright-conscious corporations. Some models with Stable Diffusion exacerbate the problem. Prior to our recent update, Stable Diffusion Version 2, anyone could create a template designed to mimic a particular artist’s unique visual style, and create new images indefinitely at speeds humans couldn’t compete with.
Andy Baio, co-founder of the Festival for Independent Artists, wrote an in-depth interview about these concerns on his blog. He spoke with an illustrator who discovered an AI model specifically designed to replicate his work. “My initial reaction was that having my name on this tool felt intrusive,” she said. “… If they asked if I could do this, I wouldn’t have said yes.”
By September, Dungeons & Dragons artist Greg Rutkowski was worried that real art would be lost in a sea of algorithmic copying due to the popularity of the Stable Diffusion prompt used to generate images in a detailed fantasy style. “What about in a year? I probably won’t find my job there ’cause [the internet] It will be flooded with AI art,” Rutkowski told MIT’s Technology Review.
These concerns, echoed by many illustrators and other digital creatives, echo on social media Many people are faced with these tricky issues and the existential threats they pose for the first time.
“I know a lot of people have been posting Lensa/other AI portraits lately. I highly recommend not doing that, or even better not using the service,” voice actor Jenny Yokobori wrote. Top Tweet Threads Another about Rensa is Riot Games artist Jon Lam. share Your own discomfort with AI-generated art. “When AI artists steal or pick up our art, I don’t just look at art, I see people, mentors and friends. I don’t expect you to understand.”
Personally, I’ve been sick on the weekends, so I’ve been stuck at home and spent more time than usual just scrolling through social media in a daze. My Instagram Stories were flattering blurs of digital artwork that cost a few pennies a page. Lensa is an easy-to-use collection of 50 stylish self-portraits, an interactive experience in which you vote for your friends on which image you’re spitting on (mostly, in my experience), and a computer pretending to be human at best can spin. .
A few friends, mostly artists and illustrators, stepped back and encouraged everyone to find an artist to pay for. A few creative people in my circle paid off too, and it’s hard to blame them. For better or worse, it’s truly amazing what the AI image generator community can do right now.
Soon, we’ll all be paying attention. In the name of story study and vanity, I downloaded Lensa and tried the app. In 2016, we only paid a one-time fee to the artist to turn it into a profile picture. In 2016, it was just one image. Now, for less than $10, I have a set of 50 epic avatars created from my most photos. me photos, but these additional me. Me in various futuristic jumpsuits straight out of the pages of graphic novels, me in purple robes that look like intergalactic saints, me, me, me.
I see appeal A handful of friends told me how they felt when they saw the photos, alluding to the gender happiness they see in the way they see themselves. I won’t blame anyone for exploring this stuff. It’s all very interesting and at least that complicated. I like my avatars, but I wish some of them didn’t. I have no plans to use them.
I thought about my own art. These are the photos I sell when I need to stock up on my online store, mostly mountain landscapes and night skies. I rethought the few prints I sold and the effort I put into taking pictures. One of my favorite photos was taken while backpacking for five hours to a special permit from the National Park Service and to a remote fire lookout in Washington. Many have entailed lonely hours manning tripods alone in the bitter cold, tracking the Milky Way spinning over the dark horizon like clockwork.
The AI model already has enough training material to faithfully recreate photos of a nearby remote mountain range that only local night sky photographers seem to know about. Three years ago, when I was taking pictures there, I had to stop at a competitive campsite, drive several miles along a hole-in-the-wall Forest Service road, and wait for hours in the dark. I cooked a bag of expired ramen on a small camp stove to stay warm, tucked my feathers back into my jacket, and dove into anything that made noise in the dark.
I don’t make a living off of my art. But to think of those experiences and the human process they represent still feels like a loss. That means learning how to anticipate ancient star clusters in the dark, slipping on wet stones, breaking my hot shoe, and learning how to keep a spare battery warm in my down pocket. It may depreciate in the future.