Apologies in advance for this Artificial Intelligence (AI) story. I try to avoid talking about AI, Machine Learning (ML) and the like. A lot of the stuff coming out about AI is ridiculous, exaggerated and inaccurate. But amidst the quagmire of AI hype, there is some information that can be useful.
According to an article by the Barcelona-based architecture firm, they are prototyping with Midjourney. I recently developed some earrings with designer Roman Reiner. He quickly turned to DALL-E's text-to-image model to iterate on ideas. He used it to quickly see if he and I were on the same page, and then quickly came up with some completely unconventional designs to give me feedback.
A few weeks ago, my girlfriend took Footwearology's 3D Printing Shoes course where we used AI to quickly come up with and execute ideas for shoes. I used Wonder AI, a simple AI image generation tool, to bounce design ideas off of my friends and explain what they meant. I also used Wonder to generate images to explain the look and feel of the brand to a graphic designer.
A few weeks ago I came up with the idea for filter pipes. I made a rough AI image to make it easier to explain in emails, to save time on explaining. A few friends are experimenting with using ChatGPT to create STL files and then printing them. I'm sure these features will improve bit by bit over time.
Being able to iterate more closely to my idea would be very powerful. Fine-tuning the design and iterating better also makes a big difference. Right now, iteration is tedious and doesn't allow me to fine-tune a lot of the prompts more precisely to get closer to what I want. This precision step is very important because if I get close to what I want, I print the file every time and use it to continue the process. But I'm now using AI to quickly gather ideas, explore different options, and then have the most promising ones 3D modeled more precisely. Being able to go to that step right away with a more directional and precise process really speeds it up. For example, if I'm working on a relatively simple earring, the process basically starts with an AI image generation tool.
Now, I usually walk around and come up with my own ideas. Then I imagine the person who would wear or use the thing and narrow down a bunch of ideas. Then I generate a prompt with Wonder AI and tweak it a few times until it's close to my idea. Sometimes this works surprisingly well, but many times it doesn't turn out as I want it and I get frustrated. Finally, I create about five two-sentence descriptions of the product variants. Some of these have Wonder images and some don't. I use Wonder because it runs on my phone and is a little faster than the desktop-based ones.
Once you have about five ideas, let them sit for a few hours. Put the ideas in a Google doc, paste the images, and look at them again with fresh eyes. This is where you see if anything isn't authentic or fits you. If something feels wrong or personally wrong, this is where you reject it. But if the idea is stupid, doesn't work, or impossible, let it be. Then compare the idea to a use case, client, goal, or practicality. What will this bring to my client, their client? What will this bring to someone's life, day, or smile? Then think about existing ideas. Finally, brainstorm to generate the stupidest ideas possible.
This is a really free-form, playful process where I think great ideas often come. At this point, I have written out about 10 paragraphs, many of which are AI-generated images. I do other work, then I copy and paste all the good ideas, leave the terrible ones, and leave it at about 5. Then I share them with the people I work closely with and make a list with their ideas. We discuss and see which ones make it. Then I 3D model these and 3D print them using my desktop machine. I keep doing this until I find something I can test. Then I play around with FDM versions a little bit, maybe even make them in vat polymerization. I test them and ask for user feedback. Then I show the prints to users and interview them about what they want and what they need. I tweak them and make them in powder bed fusion or in the final technology. That's more or less my process. Now, the reason I'm sharing this is two-fold:
One of the reasons I share this is because I've seen more and more people recently share similar workflows from ChatGPT to 3D printing to the final part. For example, architects like External Reference are sharing their workflows from Midjourney to simulation to 3D printing. Nike is doing something similar, and many others are doing similar. To me, this is a growing trend of many people spontaneously turning to ChatGPT and similar tools to generate ideas and turn them into real products. When I asked several people if they were using AI for this article, all of them had tried AI and many had incorporated AI into their workflow.
This has several implications for us: if we can tightly integrate this process into the 3D printing workflow, it will empower many more people to 3D print and invent with 3D printing. This is a huge opportunity. But it also poses a risk, as further advances in AI could allow these tools to avoid 3D printing large parts.
At the same time, another interesting development occurred. When I started using Google, I realized that I was thinking in search terms, not in systematic thought. Instead of recalling the rise and fall of the Eastern Front of WWII to recall the major battles, I was thinking “major battles of WWII,” “important battles of WWII,” “most important battles of WWII,” and so on. Instead of searching for information or reading books about the Eastern Front to supplement my knowledge, I was collecting a small mosaic. Before I knew it, my knowledge had become a mosaic, not my own brushstrokes.
Now, when I think of an idea and try to explain it to myself and give it a name, I think in prompts. Before, I had ideas of visual images of my generation, but now I collect prompts. This change is very interesting to me. AI may be overrated, but in prototyping, text-to-image and text-to-STL conversions could really be the future.
