Discover more from Sneakerhead VC
Sneakerhead x AI
A return to my roots in sneaker design
I’ve been thinking a lot about the power of AI in creative tasks. So I went back to my roots in sneaker design and played around with prompts to generate images of shoes using Midjourney and Dream Studio/Stable Diffusion.
When I was a Creative Director at AND 1, I made creative briefs all the time and used different mediums to communicate my ideas depending on the designer I was working with. Sometimes I used text, I often used images, and occasionally, I shared music, videos, or an immersive experience (i.e. I once visited a high-end car dealership with a designer, where we found inspiration in the details of the cars’ interiors).
It’s been fascinating to see how the AI models interpret the text directions I give them, and in many ways, it feels similar to working with a designer. I started to see patterns as I modified the order of the words in my prompts, using alternative vocabulary to generate more nuanced results.
In this process, I also found some shortcuts. For example, when a word or concept carries a specific graphic iconography—like a famous designer’s name—the results represent that “look” in a way that is hard to describe. But you can feel it.
I sent this prompt to Midjourney as an initial experiment: “Basketball sneaker inspired by kobe bryant, audi RS6, Los Angeles Lakers, 35mm camera, magazine advertisement, realistic shot 3/4 view from the lateral front.” The system generated this image:
The image looks like something a board a designer would put together for concept art in the exploration phase of product creation. Crazy!
This was my first attempt at defining the visual language for the shoe based on the written design brief. Then came iteration and refinement. Next, I made a simple request to generate more versions of a specific shoe with no change to the prompt:
Modifying the prompt with words and concepts that are not visual didn’t seem to heavily influence the image. Words like “attitude” or “speed” translate into human English but didn’t work for the AI. In my experience, giving this type of direction to a designer might make a pattern more angular to depict an aggressive attitude or change the lines in the shoe to be more horizontal to incorporate the concept of speed. The generative models felt less influenced by this type of prompt: “Basketball sneaker inspried by kobe bryant attitude, audi RS6 speed, Los Angeles Lakers colors shot on 35mm camera, magazine advertisement, realistic 3/4 view from lateral front”:
Not surprisingly, creating variations on a design using AI is very fast. I wish there was a way to generate new versions with a single modification, but even without this, iteration leads to lots of options to quickly hone in on a look:
In an effort to be more effective in my descriptions, I looked for a common understanding of the text as a visual concept. I started to experiment with iconic objects that express a very specific visual language, as well as people who have a visual body of work that is well-documented or easily recognizable.
The following names and objects carry meaning, which you can see in the images the models generated.
Virgil Abloh was known for his "Off-White" collection, and the shoes below are easy to recognize in their deconstructed look. Adding “Virgil Abloh” and “Off-White” to a prompt for classic running shoes resulted in this in Dream Studio:
Chris Bengal designed the BMW 5 series, 7 series, and 3 series as they evolved the body style and look of the cars. He oversaw the creation of the X6 and Z4 and with the prompt, “Chris Bangle, running shoe, speed blur, photography by tyler mitchell” Dream Studio created this image:
Giorgetto Giugiaro is the embodiment of Italian car design—he worked on iconic brands like Ferarri, Alfa Romeo, Fiat, BMW and Maserati. “Giorgetto Giugiaro, classic sneaker, pastel, photography by tyler mitchell” generated this:
Charles Eames needs no intro, but his furniture designs defined the mid-century modern movement in many ways. The image generation model leveraged his work to create this in response to “Charles Eames, Future, Baseketball Sneaker” and below that is “Charles Eames, Collector sneaker.”
Jasper Johns, the American painter who was well-known for his depictions of the American flag:
Ansel Adams, the photographer known for his black and white images:
And last but not least, Frank Gehry. This one is a response to “Frank Gehry, classic sneaker, photorealistic.” Insane how good it is.
It's a powerful iteration tool—one where I can simply say "Frank Gehry" and trust that the AI as my design partner knows what I mean. I'd love to see a way to have the AI develop a 3D version of the 2D "sketch" to make sure the design is feasible through structural modeling logic. If you're working on something like this - or if you just want to talk about sneakers - feel free to reach out.