Introduction of My Top 5 AI Sneaker Designs

Asymmetrical Color Digital Sneakers

What stood out were the asymmetrical color variations and differences between left and right sneakers.
Looking closely, you can see differences in lace colors, decorations, positions, sizes, and the number of fasteners.
It’s a dynamic design that would be challenging for a human to create.

Evangelion-style Digital Sneakers

The color scheme and shape are on point.
Although I didn’t specifically use any Evangelion-related prompts, these sneakers were miraculously generated.
I’d definitely buy these if they were for sale.
They’re CC0, so how about it, Hideaki Anno and Studio Khara?

Illuminated Digital Sneakers

The design is great while still incorporating lights.
Maybe they’re night running shoes.
These were popular a while back.

Unique Digital Sneakers

I was amazed by these art sneakers.
They somehow look like living creatures, which is endearing. They could probably sell well in Harajuku.

Sparkly and Fluffy Digital Sneakers

These are my favorite sneakers from this project, combining futuristic elements with sparkles and fluffiness.
The AI even generated toe-up angles automatically, so there must be reference shots it’s learning from…

Other designs can be viewed as NFTs on OpenSea

👟AI Sneakers 1763👟
https://opensea.io/ja/collection/aisneakers1763

My Experience Creating with Generative AI

This was my first attempt at AI-generated images, and I was amazed by the high quality.
Many of these designs are so good that I’d want to buy them if they were sold in stores as fashion sneakers, and the designs are quite bold.
By creating a collection on OpenSea, I’ve made an enjoyable digital sneaker catalog.

What surprised me was that even when using the same prompt, not a single identical design was generated.
The models and camera angles seem to have learned from actual products, giving them a professional feel.
The initial feeling of generation was similar to displaying HTML in a browser for the first time – I was hooked.

Production Environment for This Project

For this project, I used the following environment:

• Image Generation AI: Stable Diffusion
• GPU: Google Colab Pro
• Consistent prompt: text2img (generating images from text)
• Number of generations: About 2500
• Size: 512px × 512px
• Deleted unsuccessful ones / Minor corrections in Photoshop
• Time taken: About 1-2 weeks

Key points in the creation process were to grasp prompt patterns before mass generation to get closer to my imagined image, decisively delete unsuccessful ones, and be prepared to make corrections in Photoshop when necessary for otherwise good designs.
Using Lightroom Classic for lighting adjustments could further improve quality.
In this regard, I think graphic designers and retouchers who are particularly skilled could achieve even higher completion levels.
For guidance on crafting prompts, I recommend checking out Takayuki Fukatsu’s article:

“AI Drawing Course Understood as Magic”
https://note.com/fladdict/n/n0f0be20e3e23

Personally, I found it challenging to control textures, angles, obvious inconsistencies, and text or logos.
This time I used text2img (generating images from text), but next time I want to try img2img (generating images from images) to increase variations by combining techniques.

Future Prospects

Although aratama 璞 has just started, we launched with AI digital content at our core, so I want to continue research and development to see incredible things with those who are interested.
So if you’re interested, please join our Discord. I want to have fun together, but it’s a bit lonely with so few people at the moment!

“Midjourney” and “Stable Diffusion” haven’t been out for long, so I expect various developments including updates and new products in the future.
Emad Mostaque, CEO of Stability AI which developed “Stable Diffusion”, said, “What we announced in August 2022 is the first in the series. We’ll support audio in September, followed by video, 3D, language, and code.” They also plan to reduce dataset sizes.

In the future, as AI becomes democratized, just as everyone can use the internet and web3 (especially cryptocurrencies), everyone will be able to use AI to combine, extract, create variations, and iterate on ideas multiple times, greatly expanding what’s possible.
It’s a revolution where new things are born one after another from data assets.
Conversely, those who can draw originals might be valued in a different way, but I think we’re heading towards a different world line of creation 2.0, distinct from traditional creation.
Of course, we need to properly address legal issues such as copyright. Recently, misleading AI-generated images on Twitter have raised moral questions. The economic impact also means different countries may respond differently.

As someone involved in design, I understand why some people might be pessimistic, but I believe the future is bright if we use our existing skills to enjoy expression and challenge new possibilities.
Rather than working alone, I want to research AI digital content generation while having fun with everyone, saying things like “That’s amazing!” and exploring the charm of secondary to hundredth creations.
Wouldn’t it be wonderful if we could build a token economy world here, including NFTs in a web3 style, where people could be rewarded for the projects they’re involved in?
It might also be interesting to AI-ify the DAO itself.