AIsneaks ‘crafts’ unforeseen virtual sneakers using artificial intelligence

There probably hasn’t been a more disruptive year than 2021 in terms of our ongoing shift to everything virtual. The concept, and our notions thereof, of what ‘ownership’ means, stand to be upended by NFTs and cryptocurrency, regardless of whichever direction the two (essentially) digital revolutions lead. With Metaverse in the mix, the true extent of how our lives and the entire foundation of what technology has meant to us may change remains largely unknown. As if transferring our meeting rooms to Zoom screens wasn’t enough, Bill Gates recently predicted future meetings taking place in the Metaverse itself. While these massive strides in technology may just amount to fairy dust, or conversely, may entirely transform living as we know it, there is no denying that their intersection with design, and virtually every creative field, has resulted in some unprecedented acts of ‘creation’. Case in point is this essentially endless collection of ‘unreal’ sneakers, generated in the virtual world using sophisticated code.

Globally, sneakers continue to be highly prized collectibles beyond their functional utility in sport. As an extension of that, sneaker design continues to evolve as a sophisticated discipline in product design, more often than not even intersecting with art. The prestige associated with them leads to exclusive designs nearing auction prices of close to a million dollars; the adidas x Meissen ZX 8000 Porcelain sneakers being a recent example. AIsneaks seeks to mine into the mania, transforming something physical, tangible, ownable, into a visual entity, with potentially limitless possibilities.

Designed and authored by Berlin-based programmer and web developer, Niels Garve, the AIsneaks program uses complex code that samples thousands of existing sneaker designs to generate vibrant, never before seen virtual sneaker designs. Garve terms his designs to be visually and technically “futuristic”, combining programming, artificial intelligence, and machine learning. The AIsneaks program learns to recognise patterns, motifs, and other components in the design of sneakers from its compiled repository of over a thousand existing designs, recreating them in original versions and different designs. The main technology employed to make this possible is a custom DCGAN (Deep Convolutional Generative Adversarial Network), inspired by Pytorch tutorials, save some samples that were produced using an open source NVIDIA PROGAN.

As visually distinct and stunning as AIsneak's design generations may be, it is the future that is a particularly interesting proposition for the program. While AIsneaks has already produced more than 3,000 sneaker virtual models, its planned outreach manifests in two ways: unsurprisingly virtual and physical. Virtual versions of these AI generated sneakers in 3D may be made available as NFTs, or made purchasable-wearable in the Metaverse; essentially translating to its growth curve being nearly proportionate with these digital revolutions. The unique designs may physically manifest through 3D printing, furthering the technology quotient by miles, or via a planned partnership with a brand or designer, involving a curatorial aspect.

STIR interacts with Garve to probe further into the “generated by AI” mania, and where he sees his potentially landmark intervention heading.

Anmol Ahuja (AA): Please tell us about the genesis of your idea to create a code that could in turn design sneakers. Where did the process begin? Where did it intersect with what you were originally doing?

Niels Garve (NG): As a web-developer, I am always curious about future technology. Besides that, it's almost necessary for developers to stay up to date. So, three years ago, I started participating in online courses from Microsoft and IBM about artificial intelligence. At the same time, NFTs started off and I also wanted to take a look at blockchain programming. After my online courses in machine learning, I felt it was time to start my own project to deepen my knowledge. Sneakers are collectible per se; NFTs are a way to collect, and they work pretty well with images. So, the idea was born: of AI generating something that is collectible as NFTs: sneakers.

AA: Some of the best sneaker designs from last year boasted artisanal craftsmanship and an intricacy in their design and craft. Do you think the AI-designed sneakers seek to counter that?

NG: I understand that, and it is good having a critical eye on AI in general. For me, AI in design works as a creative tool suggesting shapes and textures for designers to work them out. It is a source of inspiration. It feels a bit like the technology revolution of the 80s and 90s. I don't want to oversimplify it, but designers used to draw by hand on paper, and I am pretty sure that most of the designs from the last few years’ collections have been created using computers and programs like Photoshop or Blender. I would say artificial intelligence is an opportunity for taking the next step, but without losing the human touch.

AA: What do you think about the physical manifestation of your sneaker designs? How do you plan that, if at all?  

NG: Besides being collectible, sneakers are made to be worn and I think 3D printing is a huge opportunity for AIsneaks. The issue here is that I have to ‘teach’ my AI the concept of what makes a shoe wearable or comfortable or performant for running. Most likely, a designer has to take over the AI’s idea and bring it to life.

AA: Do you have some systems in place to avoid plagiarism or counterfeiting of the very idea to produce sneaker designs the way you do? 

NG: NFTs are currently the best way to prove the uniqueness of images, be they have to be hand-sketched or AI-generated. Additionally, my code is closed source, and my approach is not yet very common. Furthermore, there are a lot of high-performance image generation algorithms out there. First of all, NVIDIA's StyleGan got a lot of attention with thispersondoesnotexist.com, which is a website showing images of non-existent, AI-generated people. And there's a lot of art and design out there based on StyleGan. The reason why I chose to go a different way is that using these sophisticated algorithms, my images stayed within a certain contemporary design range and did not show something entirely new.

AA: What do you have in mind for the future of your own, if I may call it that, digital 'Sneakerverse'?

NG: I am planning to release sophisticated and futuristic 3D sneakers as NFTs that were originally created by AI in 2D, and then curated and finished by designers in 3D. Besides that, a big opportunity is making them virtually wearable in the metaverse or in augmented reality applications. I am already planning that. In the far future, I would like to also generate other items like furniture and clothes. I would like to dive deeper into the “created by artificial intelligence“ approach.