After adding a function that enables users to create gorgeous digital portraits of yourself in modern art styles, the Lensa photo and video editing software has gained social media notoriety in recent weeks. For a little cost and the effort of uploading 10 to 20 different pictures of oneself, it does that.
Lensa is basically prompt engineering on top of Stable Diffusion packaged as an app that’s rumored to be making $1M/day from in app purchases.
This is just one example of a new generation of apps that will turn prompt engineering into productivity and entertainment tools. pic.twitter.com/LiVxInklD5
— Dare Obasanjo 🐀 (@Carnage4Life) December 8, 2022
Lensa may be the most elegant commercial use of text-to-media AI to date. 2022 was the year that this technology left the labs and began colonizing our visual culture.
It has sparked a different type of fire among the creative community as well as among social media influencers trying to stand out. Kim Leutwyler, an Australian artist, told the Guardian that she recognized elements of other artists’ styles in Lensa’s portraits, including her own.
Since the early this year debuts of Midjourney, Dall-E from OpenAI, and Stable Diffusion from the CompVis group, the ease with which individual artists’ approaches can be imitated has raised red flags. Artists believe their intellectual property has been compromised, along with potentially some of their soul. though has it?
Well, not in the eyes of the current copyright laws.
I’m cropping these for privacy reasons/because I’m not trying to call out any one individual. These are all Lensa portraits where the mangled remains of an artist’s signature is still visible. That’s the remains of the signature of one of the multiple artists it stole from.
— Lauryn Ipsum (@LaurynIpsum) December 6, 2022
What is it, if not outright theft?
Text-to-media Although AI is naturally exceedingly complex, it is conceptually understandable for those of us who are not computer scientists.
It’s worth taking a few steps back to understand how artists’ unique approaches can find their way into, and out of, the black boxes that power systems like Lensa in order to fully appreciate the advantages and disadvantages of Lensa.
In essence, Lensa is a simplified and personalized front-end for the open-source Stable Diffusion deep learning model. It gets its name from the fact that it generates creative output through a method known as latent diffusion.
The crucial word here is “latent.” A latent variable in data science is a property that cannot be directly measured but can be inferred from things that can be measured.
Machine-learning algorithms were fed a huge amount of image-text pairs when Stable Diffusion was being developed, and they taught themselves billions of possible ways these photos and captions could be connected.
This resulted in a sophisticated body of information, none of which can be understood by people. In its outputs, Stable Diffusion perceives a cosmos of numbers and relationships, while we would see “modernism” or “thick ink.” And the numbers produced from the original image-text pairs were used in the intricate maths that led to all of this.
We can navigate the vast ocean of potential outputs by entering in useful prompts because the system digested both descriptions and image data.
As an illustration, consider the picture below. The text prompt included “digital art” and “artstation,” a website where many modern digital artists call home. Stable Diffusion developed an association between these terms and characteristics it discovered in the many works of art it was educated on. The end result is a picture that looks great on ArtStation.
What sets Lensa apart?
Since Lensa only accepts images rather than words, it seems extremely different from Stable Diffusion, which is a text-to-image system that lets us explore several possibilities. That’s because one of Lensa’s most significant improvements is the simplification of the textual inversion procedure.
Stable Diffusion’s existing knowledge base is updated by Lensa using user-supplied images, which teaches the system how to “catch” the user’s features so it can later stylize them. While this is possible with the standard Stable Diffusion, the procedure is far from straightforward.
Although you can’t manipulate the photographs on Lensa to look a certain way, you get a lot of possibilities that are almost always spectacular in exchange. These images incorporate concepts from the works of other artists but do not actually include any of their individual pieces.
According to the Australian Arts Law Centre, while individual works of art are protected by copyright, their underlying concepts and aesthetic aspects are not. Similar to this, it was determined that an art style is exempt from copyright law in the Dave Grossman Designs Inc. v. Bortin case in the US.
How about the creatives?
However, the fact that art forms and methods are now transferrable in this manner is incredibly unpleasant and disturbing for artists. There may be pressure on regulations to change when technology like Lensa become more widely used and artists feel more and more ripped off.
The future appears tough for artists who work on small-scale projects like producing digital drawings for influencers or other web businesses.
While it is simple to create an attractive piece of art using AI, it remains challenging to produce a very specific piece with a specified subject and context. Therefore, the artist’s personality continues to be a significant context for their work, regardless of how apps like Lensa transform the process of creating art.
It may be necessary for artists to put more effort into promoting themselves by taking a page from influencers’ playbook.
It’s still early, but the next ten years are going to be turbulent for both artists and art patrons. The genie, however, is no longer contained, and that much is certain.