Pushing the boundaries of AI in footwear design to find where it is helpful and where its limitations are.
AI, like 3D printing and scanning, quickly produces images that look great in a thumbnail. Beautiful renderings that look impressive at a glance, but distorted on closer inspection.
Based on research since late 2022 (engaging many different tools and prompts, design conferences, tutorials, expert articles, and interviews) I came to the conclusion that current AI excels at renderings, but still requires a lot of human guidance.
The prompt engineering and source images users provide are critical to getting useful results. If everyone is using the same AI tools, trained on the same LLM and image data sets, then the results will tend to look the same.
The most important limitation I found is that AI image generators, and usually the chat AI also, really struggle to come up with original functions, mechanical features or materials. So the ideal workflow becomes a back and forth between man and machine.
SAMPLE DIGITAL DESIGN WORKFLOW:
use ChatGPT to create design briefs, summarize trends and market factors, write prompts (and key negative prompts)
input the prompts (fine tuned by experienced human) into an AI image generator like Midjourney or Stable Diffusion
include images along with the prompt that reflect the aesthetic style desired (play with the weighting percentages, I often use 15-30%)
choose the best AI results as inspiration or underlay reference images in Gravity Sketch
design a quick VR footwear 3D model in Gravity Sketch on a last or mannequin (so you know it fits)
take screenshots of your GS design back into AI to create realistic renderings for the clients or team
Gravity Sketch also makes it easy to 3D print a full size mockup shoe for in-hand inspection, and if printed in TPU, even allow an on-foot test
Experimenting with many different prompts, source images, and tools I tried to ideate concepts like: ways to move air and moisture through the shoe; new constructions that help the shoe flex, new lacing or ankle support structures for basketball; trail running shoes inspired by nature to repel water or grip; athletic shoes in the style of famous designers like Syd Mead.
But the results showed that AI’s power for designers is to generate cool renderings and many variations quickly. The creation of user needs/functional features/constructions/fit/materials/etc. are still up to the experienced designer. The AI tools available to us today can’t ideate mechanical features.
It’s also important to note that AI image generators tend to distort human bodies, laces and midsoles in strange ways with “uncanny valley” appearances.
Parts and pattern lines are often ugly and random, don’t align or end in aesthetically pleasing ways, or are simply impossible to manufacture. It’s the techno-babble gibberish of a technology still in its infancy, but I expect it to improve steadily.
You can’t deny that the rendering quality is amazing when you input great source images. AI can sometimes replicate hand drawn techniques but does struggle to create something like loose pencil sketches versus renderings.
AI tools make for excellent inspiration and reference images, but you couldn’t send them to a factory to make samples just yet. We still need plenty of human hands to create products that appeal to humans, ensuring comfortable fit, to make realistic midsole models that can actually be cut into molds, patterns for cutting or knitting materials, choosing marketable colorways, and so much more.
In summary, I think AI is great for inspiration and a digital workflow, but is not taking over just yet. It remains to be seen in the next decade how quickly AI and automated systems can integrate so many parts of softgoods, footwear design and development reliably and profitably.
AI FOOTWEAR DESIGN
Pushing the boundaries of AI in footwear design to find where it is helpful and where its limitations are.
AI, like 3D printing and scanning, quickly produces images that look great in a thumbnail. Beautiful renderings that look impressive at a glance, but distorted on closer inspection.
Based on research since late 2022 (engaging many different tools and prompts, design conferences, tutorials, expert articles, and interviews) I came to the conclusion that current AI excels at renderings, but still requires a lot of human guidance.
The prompt engineering and source images users provide are critical to getting useful results. If everyone is using the same AI tools, trained on the same LLM and image data sets, then the results will tend to look the same.
The most important limitation I found is that AI image generators, and usually the chat AI also, really struggle to come up with original functions, mechanical features or materials. So the ideal workflow becomes a back and forth between man and machine.
SAMPLE DIGITAL DESIGN WORKFLOW:
Experimenting with many different prompts, source images, and tools I tried to ideate concepts like: ways to move air and moisture through the shoe; new constructions that help the shoe flex, new lacing or ankle support structures for basketball; trail running shoes inspired by nature to repel water or grip; athletic shoes in the style of famous designers like Syd Mead.
But the results showed that AI’s power for designers is to generate cool renderings and many variations quickly. The creation of user needs/functional features/constructions/fit/materials/etc. are still up to the experienced designer. The AI tools available to us today can’t ideate mechanical features.
It’s also important to note that AI image generators tend to distort human bodies, laces and midsoles in strange ways with “uncanny valley” appearances.
Parts and pattern lines are often ugly and random, don’t align or end in aesthetically pleasing ways, or are simply impossible to manufacture. It’s the techno-babble gibberish of a technology still in its infancy, but I expect it to improve steadily.
You can’t deny that the rendering quality is amazing when you input great source images. AI can sometimes replicate hand drawn techniques but does struggle to create something like loose pencil sketches versus renderings.
AI tools make for excellent inspiration and reference images, but you couldn’t send them to a factory to make samples just yet. We still need plenty of human hands to create products that appeal to humans, ensuring comfortable fit, to make realistic midsole models that can actually be cut into molds, patterns for cutting or knitting materials, choosing marketable colorways, and so much more.
In summary, I think AI is great for inspiration and a digital workflow, but is not taking over just yet. It remains to be seen in the next decade how quickly AI and automated systems can integrate so many parts of softgoods, footwear design and development reliably and profitably.