7/17/21 | By: Matthew Lallas & Stefano Paolina
As you have probably seen, Artificial Intelligence (A.I.) and NFTs are taking the world by storm. These emerging technologies are becoming increasingly common and being implemented into all different types of companies and industries. Here, we want to highlight how a company called Mirage Gallery is using A.I.-powered art and NFTs to redefine its industry. First, however, it is important to understand what NFTs and A.I. art are at their core.
NFT stands for "Non-Fungible Token.” This means that every NFT is a completely unique digital "token.” Simply, an NFT certifies ownership of any digital asset (this could be a photo, video, GIF, etc). While the asset itself can be replicated (i.e. screen-grabbed or downloaded), the buyer of an NFT retains sole ownership, similar to the sale of physical art. This ownership is logged, tracked and maintained in blockchain so that it’s history and validity can always be traced.
The demand and hype for NFTs has grown exponentially in the past several months. There are a few notable moments that are attracting lots of attention: Beeple, a popular artist adopting NFT, sold an NFT backed piece, "The First 5000 Days" for $69 million); additionally, the rapid growth of crypto projects such as CryptoPunks and CryptoKitties add diverse implementation of NFTs and are gaining unique attention and traction from the public eye.
Though, where Mirage Gallery arrives is the emergence of A.I. art as a prominent subset of NFTs. In order to create A.I. art, Mirage Gallery uses new technology like Generative Adversarial Networks (or GANs) and GPT-3. These programs, which we'll explain in detail later, are the products of vast expansions and innovations in collaborative A.I. software in the past several years. Mirage Gallery then utilizes GANs to take "real" inputs of art pieces in order to create completely new and unique visuals that replicate a human rendering. For example, this A.I. can be trained on thousands of paintings (say, Picasso or Salvador Dalí) in order to create brand new paintings that incorporate a similar style. These technologies, along with a larger societal shift from physical to digital forms of art, is what creates the market for Mirage Gallery's work.
Generative Adversarial Networks (GAN) are a quickly growing mathematical and computer model that allows novel output from AI materializing from real world data input. To conceptually understand how GAN works, it seems easiest to begin with how we (humans) work when it comes to art or imagination. Let's start with a thought experiment. Let’s say I ask you to imagine a bicycle and then I print a copy of the first image in your head and we analyze that image. What we will find is that the bicycle in your imagination doesn't come from nothing, but rather, it stems from a vast set of information in your head: previous experience of various bikes, knowledge of colors and designs, styles of bikes, etc. So the human brain is generating an image of a bike that is based on all the bikes you have ever seen. But your mental image might not be 100% realistic- maybe the tires don't have treads, maybe the proportions are off, maybe the wheel frames are not proper for the bike, or maybe you didn't include a valve on the wheels to pump. When we compare your mental image to a real image of a bike, it's probably pretty easy to tell which is real and which is fake. Then we do round two and repeat the experiment. Now your mentally generated image is more detailed and specific. We repeat this until your mental image is so convincing of a real picture of a bike that it's indistinguishable.
What GAN does is exactly this. A GAN has real data that gives real outputs and generator model that generates outputs similar to the real data- this is considered a fake output. A GAN also has an analysis - this is called the discriminator model. The discriminator model tries to determine if the output is from the data or if it is fake. If the output is a fake bike and the discriminator knows this, the generator model updates itself to make the fake output more convincing. If the fake does convince the discriminator, then the discriminator model updates to be more precise. Founder of Mirage Gallery, August Rosedale articulates it, “the generator creates outputs and passes to discriminator which then determines if its from the actual data set or not. So the goal is to trick the discriminator in to thinking it is from the actual dataset." Ultimately, this allows AI to generate new (fake) outputs based on real data, updating itself to the point where the fake output is just as convincing as the real world data.
Dataset: the real-world data given to the system (all the bikes you have ever experienced)
Generator Model: generates outputs that re not from the real data but closely resemble (imagining a bike that is not specifically one you have seen) This model is assisted by noise or random input that allows for fake outputs. It generates a variety of close outputs that are not real.
Real Output: output that is directly from the real dataset (picture of a bike you have seen)
Fake Output: fake output that is not from the real dataset (your imagined bike)
Discriminator Model: differentiates whether the output is real (from real world data) or fake (generated, imagined bike)
Updates: revises generator model if the discriminator knows the output is fake-revises discriminator model if it is too easily convinced that the fake output is real (from dataset)
So, for Mirage Gallery, the AI uses real data from real artists, and then attempts to give novel outputs of art. The art becomes more and more unique and convincing as the model updates itself, generating art that isn't already existing, yet draws from the input data from many of our favorite artists.
Both NFT and A.I. art come together on marketplaces like OpenSea. OpenSea, along with other crypto marketplaces, allows creators and buyers to peruse different art pieces- most of them are bought and sold using Ethereum or a similar cryptocurrency.
Mirage Gallery is currently operating at the crux of NFTs and A.I. art. In order to learn more about this rapidly growing industry, and understand how individuals can break into this space with projects of their own, we sat down with Mirage Gallery founder and OmnyLaunch co-founder August Rosedale for an exclusive interview.
Q: "What was the inspiration behind Mirage Gallery?"
AR: "I saw people using different methods to create art and crazy filters, and I saw that and was always thinking: okay, this is super cool, people are able to train these models on huge datasets and create anything with them. So I was thinking, I could do this with art. When I saw what people were doing with art, and when I was playing with that, I was thinking that one of the things with art is that there is the meaning behind it and that there's a story behind the artist and the art piece, which is part of what gives it value. So, instead of just doing art, I wanted to create artists that create art. So, the important part of artists is that they have their own personalities and don't just create art. So, I started to combine things: I have art, and I have this possible way of creating textual outputs for these artists. I decided to put that together and create the face and the art with a GAN.
The blockchain side of things was what I wanted to do in order to add a way to authenticate it. It's funny because I got into this before the crazy NFT hype happened. I was planning on only selling physical pieces. Then the NFT hype happened and now it's targeted more in that way to sell them for Ethereum."
Q: "Why is A.I. art/NFT art important, and where do you see it going in the future?"
AR: "The thing I've always said about NFT art and its being important is depending on how you're implementing it. Because when I first saw this NFT hype happening, I saw people posting NFTs of them eating an apple. There's no value there. But, the things that I found to be cool are when you're not doing it to create an NFT, but when it makes sense for your project to do it as an NFT. For Mirage Gallery, I just wanted a way to verify ownership and origin of the pieces. So an NFT makes perfect sense because you can see it came from my contract. So, for the value of things like NFTs and where I think it's going to go, I think things are going to continue moving towards a place of dynamic NFTs. Companies like Async Art are doing some really cool stuff. They're doing something really creative which could only be done as NFTs. They're doing this with music now, though they started with just art: one person owns a master image. The master is a bunch of different layers, where different people can own each of the layers, and could change them with a few different options. That's awesome because that would be uncommon to have that done in any physical art piece. So, I think things are going to continue to go in that direction with NFTs, where it will become not as common for it to just be "I'm selling this picture as an NFT", because that only makes sense when the picture started in digital form. What I think is going to stop happening is people taking a picture of something and making that into an NFT. Because, in my opinion, that doesn't really make as much sense for why that would be done that way."
Q: "How would a company like OmnyLaunch help someone who is wanting to break into this space or who is wanting to create something similar?"
AR: "So, in this space, there's different ways to get into it. And some ways are much more technical than others. And when you get into the technical side of things, with A.I. and blockchain, it can get really complex really fast. Like I was saying earlier, there's an aspect of helping each other and building off of each other. So, if you're trying to get into this space and you don't want to just mint an NFT, because there are marketplaces like OpenSea where you can mint an NFT through their site and you don't need to do anything technical, and you want to go one step further and do something like creating a smart contract that does something dynamic, that's where we [OmnyLaunch] can help. That's where you're going to need the technical knowledge to do it. And you can have awesome ideas without having the technical expertise to implement them, and that's where it's super exciting for us to potentially have people with awesome ideas, partner with them, and then create something awesome."
Below is a brief look at the AI generated artists and their self-willed faces, biographies, and artwork. The system generates its own novel artist and their entire story, from personality to art style to twitter handle. Taylor and Alejandro are not real people; real people are put into the system and this is the product from that input. As you can see, everything about the artist is unique and personal. This advancement from just AI art to AI artists and art is what adds a personal layer to the innovation.
Alejandro is a realist visual artist born in Panama in 1974. He painted landscapes during his adolescence and received his formal arts education in Panama, Barcelona, and New York City. Since 2001, he has held over 25 exhibitions of his paintings throughout the US and Europe. His work has appeared in numerous publications as well.
Taylor is a young abstract artist born in Waco, TX and raised in Austin, TX. She studied Visual Effects at the Austin Career College from 2003–2005 and has been working as a freelance illustrator since then. Taylor creates vibrant images using a wide variety of techniques such as oils, acrylics, pastels, and spray paint and her art tackles human identity, experience, society, and the environment.