Self-flattering is driving AI into the mainstream. But should you join the thousands of people experimenting with AI portraiture? This new trend, made possible by an app called Lensa AI, raises some very tough questions about craft, consent, and bias.
Table of Contents
How does Lensa AI work?
Developed by Prisma Labs, the Lensa AI app (iOS/Android) was launched in 2018 and offers multiple photo retouching features. Essentially, it’s a photo beautification tool – it makes your selfies prettier by applying filters and removing “imperfections”.
But the latest version of Lensa AI includes something called “Magic Avatars”. This feature allows you to commission up to 200 AI-generated portraits in a variety of styles. All you have to do is share 10-20 pictures of your face and pay a few bucks. The results, as you can see, are pretty impressive.
Alright folks, let’s have fun! @MKBHDwe created AI avatars that seemed too cool not to share. If you like them, feel free to use them. And guys, let’s know who to do next! pic.twitter.com/9CMtEZDqG5
— Prisma Labs (@PrismaAI) November 25, 2022
Well, these “magic avatars” are not generated by a proprietary AI. They actually pay Prisma Labs to create portraits using Stable Diffusion, an open source machine learning model. The Lensa AI app is a middleman and curator, but it’s easier than taking care of Stable Diffusion yourself.
Stable Diffusion is trained on millions of publicly available images. Because of this, it can mimic dozens of artistic styles, including manga, sci-fi, pop art, and traditional portraiture. (Put simply, Lensa AI combines your selfies with existing Art. The reality is a bit more complicated as hundreds of images can contribute to a resulting portrait.)
If you want to try Lensa AI’s Magic Avatars tool, you can install the app on iOS and Android and spend a few dollars for the feature. You can order at least 50 portraits for $2. But most people pay $4 for the 200 max portraits as only a handful of the images produced by this AI are really useful or appealing.
Copyright and privacy concerns are real
The Magic Avatars tool generates images through Stable Diffusion, a machine learning model trained on publicly available images. These images are sourced without consent and there is no way for any person, artist or company to opt out of the record.
This, of course, raises privacy and copyright concerns. A person might not want their selfies included in the AI’s dataset – it’s creepy! And with Prisma Labs raking in hard cash, many artists are concerned that their work will end up in someone else’s bank account. It doesn’t help that AI image generators, which are fast and cheap, threaten job security for professional artists.
There is also concern that someone could put your selfies in Lensa AI and create artwork of your face without your permission. This is important for a reason we will emphasize later.
I’m cropping these for privacy reasons / because I’m not trying to call a single person. These are all Lensa portraits, with the mutilated remains of an artist’s signature still visible. These are the remains of the signature of one of the many artists who stole it.
A 🧵 https://t.co/0lS4WHmQfW pic.twitter.com/7GfDXZ22s1
— Lauryn Ipsum (@LaurynIpsum) December 6, 2022
Here’s the problem; Today’s laws and regulations don’t define how machine learning datasets are supposed to work. We don’t know if this stuff violates privacy or copyright laws. And because of this, you would have a pretty hard time arguing copyright infringement in court. The images produced by Stable Diffusion contain traces of artist styles, original content and signatures (or watermarks) but they don’t look identical to all existing images.
On the positive side, well-known companies are cautious in this area. They are openly concerned about how AI can lead to copyright infringement. For example, Getty Images refuses to touch AI until the rules are better defined, and Shutterstock takes a unique approach to ensuring real artists get paid.
Dozens of companies could have invented “magic avatars,” but few are willing to take the risk. Even if you don’t see the Lensa AI as a privacy or copyright issue, it’s clear that this issue will end up on the desks of legislators at some point.
Where is your data going?
We often share selfies on Instagram or Facebook without batting an eyelid. But when using a strange app like Lensa AI, some people get cramped. How will this company use your photos and will this result in an invasion of your privacy?
Well, according to the terms of service, photos uploaded to Lensa AI are converted into data to train a machine learning model. The actual images are discarded while retaining information such as position and orientation of facial features. Additionally, images captured with an iPhone selfie camera (which uses TrueDepth technology to map your face) may contain data such as the topology of the face.
Note that Lensa AI is also a photo beautification app. And when reading the terms of service, it’s often unclear whether Prisma Labs is referring to its beautification or “magic avatar” features. Unfortunately, I’m not exactly sure about the details of this information. (Nonetheless, the Terms of Service specifically state that “Magic Avatar” selfies help train stable diffusion, and that these selfies are deleted after the AI generates their images.)
You can request that they erase all your personal information by sending an email to mailto:[email protected]
3. If you upload any of the images to social media, you have given them permission to use them for promotional purposes. You can revoke permission by sending an email to: [email protected]— Chanda Prescod-Weinstein (@IBJIYONGI) December 4, 2022
Since this AI is trained on publicly available photos, user privacy might not be a big issue for some people. If you upload a lot of selfies to Facebook or Instagram, your face may already be in the record. (Also, if you’re a fan of AI, feel free to contribute your facial data.)
But some people have a limited online presence. Privacy is priceless, and if you’ve done a good job of keeping your face off the internet, I suggest you avoid Lensa AI. After all, we don’t really know where this data will end up.
If you have tested Lensa AI and wish to remove your information from the record, please contact us [email protected]. Note that according to the Terms of Service, Lensa AI is not obliged to delete data upon your request.
Lensa AI can lean toward bias
Like all technologies, Stable Diffusion and Lensa AI are prone to distortion. Some people take this to mean “the AI is a fanatic,” which is funny (but technically wrong). Artificial intelligence is just an algorithm made by humans from a jumble of data.
This becomes clear when you leaf through it Alien Lensa AI portraits. The AI has an awkward habit of sexualizing women, likely due to the imagery included in its dataset (I suspect soft-core fanart makes up a decent chunk of the dataset). To be clear, I’m not trying to sound like a prude – this AI Yes, really has a soft spot for big boobs and such TechCrunch It reportedly spits out porn on occasion.
The AI also fights with race. There have been multiple reports from Asian women noticing that the AI has reduced or altered their facial features – as described by one user, the output reads “rather East Asian distorted.” Again, this is likely due to the dataset potentially containing too much illustration of fanart (which usually focuses on Japanese styles, ideals, and trends).
Well, that’s not just an offensive inconvenience. This is a problem that can easily lead to abuse. What’s stopping someone from taking your photos, putting them in the Lensa AI and producing porn or racist images? Is that something we have to ask with every AI image generator?
Should You Use Lensa AI?
As with all emerging technologies, AI imaging is a mixed bag. Tools like Lensa AI can create amazing portraits for a very low price. It’s faster, more convenient, and more accessible than any artist in the real world, but that convenience can come at a cost.
Unfortunately, we cannot look ahead in time to see the impact of this technology. We don’t know how this will affect artists, individuals or companies. And from a privacy perspective — well, can you think of a single way someone could use your facial data? This lack of knowledge can be very worrying.
Using Lensa AI is of course a personal choice. And I don’t blame anyone for testing this technology. It’s interesting, exciting, and often very flattering. But the potential downsides of this trend should not be ignored.
This article was previously published on Source link
