How I Used AI to Make A Fabric Pattern: From AI Image to Real Sample

I woke up one morning looking for a way to visualize an outfit design inspired by some beautiful ombre flowers I had seen.

A couple weeks back, I had already used Meta AI on my Whatsapp app to design a laptop case which I turned into real samples with a local Ghanaian seamstress and was curious to see how Meta AI would perform with an outfit. 

I quickly learned through experimentation that Meta AI’s output was only as good as:

1. my ability to describe what it needed to do, and 
2. my understanding of its limitations. 

This, my Scratchpaddies, is why it's important to learn about prompt engineering. Prompt engineering means you're giving AI instructions in a way that gets you relevant results. If, like me, you enjoy turning ideas into visuals, asking questions, experimenting to bring products into life, then this one's for you! In the case study below, I’ll walk you through key lessons and tips I found that will help you sharpen your own prompt engineering skills.

Crafting and refining to develop an effective prompt

I was a little creative with this one. I first thought about what I wanted to see: a bridal fit, (because who doesn’t love a good Ghanaian traditional wedding?) inspired by the yellow and pink ombre flowers I had seen earlier. I drafted initial details and then asked ChatGPT something along the lines of, “Here’s the plant I’ve seen, I want to use it as inspiration for a form-fitting bridal dress. Help me craft the right prompt.” ChatGPT then returned a draft for a prompt which I could use in Meta AI:

I tweaked my prompts a bit and with that, the visual experimentation fun began!

Unlike ChatGPT, the Meta AI model (Llama 3.2 specifically) doesn’t take into account prior details from your conversation in the chat box. It requires the necessary knowledge/context to be restated in each prompt, which is referred to as multi-turn conversations. Engineers developing with Llama have to pass along context each time if they want Meta AI to build upon prior knowledge. 

Upon realizing this, I reran the same prompt with the necessary context and was seeing different visualizations — including this one where our model has three hands (thanks, AI 😂).

In my final rerun of the prompt, I came across this visual below which I personally think is a cute pattern that captures what I was thinking and can possibly be adapted to a kente outfit.

I messaged a kente weaver in Ghana, sent him the picture, and two seconds later, I received a call. “Ei, did you use AI?!” he exclaimed. Will he bring the design to life as a kente sample? Stay tuned! We're working on it.

In the meantime, here are three key prompt engineering reflections for when you next decide to play around with your own ideas: 

• The more specific you are about what you want, the better the response will be.
• Experiment! I leveraged ChatGPT to help generate my initial prompt for Meta AI as a starting point. However, I could have been more intentional about the kind of techniques I used. One technique out there is role play prompting. For this technique, you can tell the AI model, for example, “Imagine you’re a fashion designer doing X, I want you to…” Prompting the AI to act like a specific character can result in more relevant responses. (There are a variety of prompt engineering techniques out there and I suggest you learn more and experiment away — check out Coursera if you haven't already! shameless company plug)
• Be aware of the various models and their limitations, e.g., Meta AI needs the context passed on each time, whereas ChatGPT builds upon your previous prompt without you needing to repeat the context. The Llama model I tried from Meta AI also still doesn’t have human images completely figured out (case in point-image with three hands).

Have you used AI to bring any visual ideas to life? I’m specifically interested in how people interact with various AI tools on their phones and use that to create real product samples. Will you experiment? I would love to hear about it. Feel free to drop any questions or comments below! Let me know if there are any skills you'd be interested in learning next.