Why did I do it? Why did I shell out $8 for 50 different abstract, impressionistic, mind-bending portraits of myself? Am I that much of a narcissist? Are we all?
Our Instagram and other social feeds are currently flooded with these high-resolution interpretations of our faces and various lewks, also known as Magic Avatars. They all come courtesy of the Lensa app, an AI-image-generation engine that creates some wild and often fantasy-driven interpretations of whatever images we feed it.
There have been other, similar platforms like Reface, which let you put your face inside famous movie clips. Everyone was doing it for a while. I turned myself into Tom Cruise.
The app was free (with lots of ads, if I recall) and eventually we all decided that maybe it wasn’t such a good idea to upload our images to some rando app developer.
Lensa fills that space but with a product that almost demands you pay for it, first with a hard sell on a costly subscription fee and then more casually with a come-on for a pay-per-AI-image batch offer.
What’s shocking to me is just the sheer number of people that are paying to have their images digested and then spit out as striking AI artworks. The FOMO is so strong here that everyone (even this now-embarrassed reporter) is succumbing. My adult son spent almost $15 for a batch of images of his girlfriend.
I'm not proud
Why did I do it? Two reasons. I got tired of seeing other people’s Lensa Magic Avatars on my feed and I also really needed something fresh for Instagram. I’m kind of that way about my social feeds, always trying to keep the pipe full for reasons I can hardly explain here.
Adding my AI images to Instagram’s growing legion of majestic Lensa-generated portraits wasn’t hard. The app is freely available on iOS and Android. Opening it presents you with a $49.99 subscription offer that you can “cancel anytime.” I personally hate apps like this, ones that tease you with an incredibly cool feature but demand exorbitant cash payments up front (I may also be cheap).
Like many other apps, the sub offer is kind of a front. If you ignore it, the app immediately dumps you into the pay-per-play section, where you can buy anywhere from 50 ($7.99) to 100 ($14.99) AI-generated Magic Avatars without subscribing to anything.
Now, I cannot remember the last time I paid $8 for the privilege of using a single app feature, but such was the siren song of those beautiful portraits. I rationalized the cost in my head, “Come on, you’d pay that much for two Krispy Kreme donuts and a water,” which is something I did buy last week.
After that, the app asks you to upload between 10 and 20 images of yourself in various poses and with a variety of expressions. I decided to peruse my iPhone’s “Selfies” folder, where I found a bunch of appropriate photos. I uploaded a batch, but Lensa rejected some as unusable (it didn’t say why), so I selected a few more. In hindsight, I should have chosen far fewer profile shots and images with me making silly faces. As with most AI image systems, you get what you put into Lensa.
In general, I’m pleased with the results (you can see some samples above) but I came away uneasy, feeling a little bit like I’ve been played. I should’ve asked some questions, a lot in fact, before uploading my photos.
A day after Lensa delivered 50 4K AI images of me (they stay on the app, but you can download your favorites in standard or 4K resolution), I sent Lensa developer Prisma Labs a list of questions:
- Do you store photos uploaded to you? If so, are they encrypted?
- When did you launch the Magic Avatar portion?
- How, exactly, is AI used to generate the images?
- How many people have uploaded their images?
- Why do you front-load the subscription fee when people can just pay $7.99 for 50 images?
- How are you addressing concerns about people's images being uploaded without their permission?
- Have you heard about some being used to generate adult content?
In some ways, you could take any – or all – of these questions and apply them to virtually an AI-image generative system. While all these AI systems are exciting, they do feel like black boxes. For some, we simply put in text prompts, but we still don’t understand how they’re generating art.
The heavy lifting behind Lensa's Magic Avatars, for instance, is carried out by the free, open-source image generation platform Stable Diffusion - one of the many that has been accused of appropriating artists' work to train its AI model.
But that's just one of my concerns. Another is what happens to the images we so willingly supply to it? How do we know, for example, that images we upload are not being used to further train the AI? I see no way to opt out of that possibility.
Nor are there any prompts warning that you shouldn’t upload someone’s image without permission. I’ve heard reports that some people are using the system to generate pornographic images of unsuspecting people without their consent.
Where do our original images go when Lensa’s work is done? Without information that promises Lensa is deleting them, I must assume they’re keeping them all on file. We can only hope they are at least encrypted.
Prisma Labs responds
I do have some good news. Even though Prisma Labs doesn't include a lot of what I consider necessary information in its app and didn't answer my questions directly, it did point me to a voluminous FAQ that does address some of these concerns.
On the question of what Lensa does with our images, Prisma Labs writes that they're deleted from Prisma Labs' servers once processing is complete. However, the FAQ also contains this language:
"We store avatars for as long as it takes to provide our users with the service. Please note, a new feature, which would allow users to permanently delete their avatars from our servers is currently underway."
Put another way, your original photos are gone, but those images that Lensa created? Prisma Labs still has them. There is no mention of data encryption, by the way.
The FAQ does not address considerations of image ownership: The image you upload should be your own or ones that you have the right to access. It does address "sexualized" images, but mainly the AI's tendency to sexualize female characters. More importantly, Prisma Labs adds, "The intentional use of the app to create explicit content is absolutely prohibited by our Terms of Use and obliges Lensa users to comply with that requirement."
It is not clear how Prisma Labs is policing that particular rule and, even less helpfully, Prisma Labs adds this note, "Unfortunately, all these efforts haven't yet made AI absolutely safe from biased content and explicit imagery. Therefore, we stipulate that the product is not intended for minors’ use and warn users about the potential content risks."
I can’t legitimately tell you not to try the app, especially not after I went ahead and did it. Before you do, however, make sure you're comfortable with Prisma Lab's answers (and lack of answers) to any of these questions.
No comments:
Post a Comment