I am calling my new design products “A.Icons”, a fun play on words that includes the acronym A.I., adding “-cons” at the end to make it one phrase. I stumbled upon the term’s usage in my recent work, HibbitsDesign – Project Visual ID which I have added to my portfolio and am describing in this post.

When a friend and colleague (now a client), Paul Hibbits approached me to create visual elements for his HibbittsDesign.org page on GitHub, I thought, “For sure, this sounds right up my alley!” It seemed like a straightforward project—a simple header image for a website. But as we started talking, it became clear that this was more than just a header. Paul had a deeper vision of what the imagery should represent. With his generous timeline, I had the space to experiment and dig into some creative processes I’ve been exploring lately, including using AI to generate textures.

Since I recently wrapped up a micro-course for BCcampus titled AI Images in the Classroom, I was already familiar with tools like DALL-E,  Adobe Firefly Copilot and the like. The course primarily focused on generating realistic images—people, places, and objects. You know, the kind of images that get attention because they’re so lifelike… until they aren’t (hello, multi-fingered humans). While photographic realism has its place, for this project, I did not want AI to generate this kind of image or replace my design process. I was interested in what it could do in terms of mock-ups and if it could tackle symbolic representations of concepts like icons, that I could manipulate and integrate into my style.

That’s the direction I prompted my AI assistant ChatGPT (who now goes by the chosen name “Alex”) to head in. Paul’s project had three lines of text that the visuals needed to reflect:

  • Open-Source Software to
  • Empower Digitally Curious
  • Open Publishers, Educators, and Creators

After briefly discussing visual metaphors and icons with Alex, including my love of the Noun Project, I asked for ideas to represent these key themes visually. Here’s a snippet of that conversation:

Alex then provided three very suitable icons from the Noun Project, including links and the designer’s names.  

This was a delightful surprise.

While the results were not profound visually, they were usable, and I appreciated the shortcut to find them without searching and scrolling. It reminded me of the vast collection of visual representations of the NounProject and that the latest premium subscription features could benefit my project. I wrote about that in my last post. I quickly subscribed and set up a shared folder or “kit” for Paul to review. After another client meeting, we decided to use the NounProject for icon selections, but I realized the final design would need a more exciting background than a flat color or gradient.

It was around this point I realized Alex’s value wasn’t just in executing ideas or generating images but in sparking new creative paths I hadn’t considered. For instance the ability to generate rich descriptions of abstract concepts when prompted.

The Shift to AI-Generated Textures

From there, I decided to take it further. Could Alex generate the abstract textures? What about Adobe Photoshop’s (beta) latest generative fill feature? While I experimenting with Photoshop already, I found it fell a bit short when it came to abstract design. It was great for filling in gaps or removing objects, but it didn’t have the creativity to handle complex abstract textures the way I wanted. Or did it? Again, I turned to Alex to help me create the prompts to use in Photoshop. I also asked it to include water, fire and earth elements to each concept to provide more detailed descriptions. 

For example, the water prompt turned into something like this:

A few things were wrong with this prompt. There are several tips about using the Photoshop generative fill, #1 among them, Be Specific. Use clear language with at least three words. Avoid terms like “ add”, “fill”, “change”, “make”, “generate,” or “create.” Instead, be direct about what you want. A little more back and forth prompting and editing myself I tried the next version. 

The results of this prompt were interesting but somewhat… lackluster. A bit more back and forth, prompting and editing and I asked Alex, who uses Dall-E for image generation to take a crack. Quite a different look as you can see in the images here.

Not that it was as simple as all that. It took several attempts, and I had to ask multiple times for it not to incorporate any symbols or text into the design, which seemed to keep wanting to do. However, I got to a place where I could share with my client, and although we were both impressed with the texture’s level of detail and sophistication, they were almost too much. Eventually, I took the less is more approach and developed a more subtle yet still descriptive prompt in Photoshop. I have included some of my other experiments below, some from Dall-E and others from Photoshop. You can also check out the final design(s) in my portfolio

AI-Generated Textures results

If I have learned anything here, I suppose this approach to generating textures may be effective, but it certainly takes some time and a lot of trial and error. Perhaps now that I have done this research, it will be easier and faster the next time. Alex was great at generating prompts to use in Photoshop but still needed to be trained and coaxed to provide some usable text. It is a good idea to review Photoshop Beta’s tips and use them in your ChatGPT or whatever to help with that. 

Tips to Write an Effective Text Prompt on Photoshop generative fill

Tip 1: Be Specific
Use clear language with at least three words. Avoid terms like “ add”, “fill”, “change”, “make”, “generate,” or “create.” Instead, be direct about what you want.
Example: Instead of saying, “Generate a landscape,” try “A lush green forest.”
Tip 2: Include Detailed Description
Describe your vision vividly. The more details you provide, the more possibilities you unlock. Let your imagination soar and see what comes to life.
Example: Rather than “Create a portrait,” try “A smiling woman wearing a red dress, standing in front of a sunset on a beach.”
Tip 3: Explain Your Ideas
Challenge the AI to produce unique results. Use natural language to describe feelings, styles, lighting, and more.
Example: Instead of “Generate a background,” try “A dreamy, mystical forest with shafts of sunlight filtering through the trees.”
Tip 4: Use Relevant Words
Use words like “love,” “gentle,” and “playful” for heartwarming images, or “powerful,” “strong,” and “uplifting” for inspiring ones.
Example: Instead of “Generate a family,”  try “A tender moment between a parent and child, radiating love and warmth.”

Wrapping Up: A.Icon technique, a work in progress

I am still working on this. The A.Icon technique I described here still needs some refinement, and I will be looking toward a simple step-by-step process I can share for the next project. Abstract textures are an element of visual presentation that I believe is one step removed from ownership and copyright. They may sidestep the ethical dilemma of using the dataset of human creatives. Perhaps. Perhaps.

Share:

Leave a Reply

Your email address will not be published. Required fields are marked *