Posted by Apple Podcasts ¦ Amazon Music ¦ Spotify
Hi, and welcome to Try AI for Growth, a baby podcast out of Make Space for Growth. I’m your host, Sara Vicente Barreto, and here I share short and sometimes surprising stories of how I’ve used technology to tackle everyday challenges—at home, at work, and in business.
Today’s episode is about visuals. More specifically, about finally getting somewhere with AI-generated images. If you’ve ever tried to get DALL·E or another image generator to do exactly what you want, you’ll know it can feel like more trial than triumph. And quite a bit of frustration asking it to work on a specific image, only to see your request bluntly ignored. Until, of course, something clicks. That’s what happened to me this week—and I want to walk you through it.
The Background: Images that actually work
Earlier this week, I attended an AI workshop for non-profit enterprises and, for the Q&A, I had my usual question. How do I get images to work? The trainer was humble enough to say he might not have a definitive answer for me, but he did share a few tricks. With them in hand, I decided to make some experiments!
For a workshop I am preparing, I wanted to try to add a visual touch—just a few cartoon-style images to illustrate some core ideas. But I didn’t want anything too busy, too polished, or too confusing. I decided to go for black-and-white or sepia, childish-style illustrations. Think simple, clear, a little naive. The kind of thing that would support—not distract from—the message.
Now, I’ve used AI image generators before. DALL·E, Image Generator, MidJourney… I’ve tried. However, if I did not like the original image (which I often did), I had no way of making small adjustments to an image. Sometimes, the image would be great—except for the words printed on it, or a small element that didn’t make sense. In the past, that’s where I’d stop. And nothing made the AI understand that I wanted to change that picture. Any prompt would just create an entirely different new image.
This time, things changed.s I have done extensive work on my messaging and refining of my value proposition examples with it.
Getting Started with the Right Style
I began by defining the style in the prompt:
I want to prepare a series of cartoon type of pictures, black and white or sepia style. I like them to look childish and to illustrate a few concepts in my presentation for my workshop. I don’t want the image to be over-crowded. Are you ready for the prompt?
This was one of the tricks that the trainer had mentioned. To provide context and a style that would then apply through the multiple pictures I was looking to recreate. So then I could move on.
I would like an image that explores the concept of “Stop to think”, to illustrate one of the key actions in my critical thinking workshop.
The image came back spot on with what I wanted, but here is the thing – it came back with text… And I did not want any text. That is when I applied trick number 2.
The Select Tool
When the image came back with text, I clicked on it, and this magic select tool was available. I had not even noticed it was there before. So I selected the part of my image that I wanted edited and asked
With this, I finally started seeing light at the end of the tunnel in my usage of AI for image generation. The text was gone, the image was just what I wanted.
Please remove the selected text from the image.
Game Changer Trick
The unwanted words vanished. The background filled in beautifully. I had my image.
Concept by Concept: It Got Better
I wanted to go a bit further to test the alignment in the image generation and also, whether it got worse or better along the conversation. I ended up testing the images for three concepts in different parts of my workshop. Honestly, not because I needed them right away, but because I wanted to test the possibility when I have a moment and not when I am knee deep in preparing slides for the workshop. These were:
- Stop to Think – My first image came back with a stop sign, but also with text on it. I removed the text with the select tool. It was the cleanest, most intuitive result I’ve had so far.
- Check Data – The illustration returned was surprisingly on point. It had a character analyzing numbers with a magnifying glass. A bit more wording than I wanted, but it was easy to clean up.
- Preaching Your Truth – This was more interpretive. The images delivered were bold but still fit my requested style. They felt like a cohesive part of the same collection. I probably will edit further.
Here are the images:
Testing Comics: Biases in a Frame
I decided to brave into the unknown once again. I know AI is not deemed to be the funniest. But I gave it a go and asked for small comics to illustrate two tricky topics:
- Confirmation Bias
- Overconfidence Bias
Now, this was less impressive. The so-called “joke” was not really one and there was no subtil element on it. Granted, the design was still aligned with the original request. I am guessing to try this we may have to provide the text and see if it works, or else bring the image into Canva and edit out the text with our own satirical views.
All in all, I was still really pleased with the outcome. And more importantly, with the fact that I made huge progress in getting consistency, as well as images that I could work with, rather than speaking to a wall (which was what I felt before with an image editor). That was a win.
Lessons Learned
Here’s what stood out to me in this experiment:
✅ Clarifying the prompt upfront matters – Style definition upfront saves you time later.
✅ Select tool is a game-changer – You can now edit instead of starting over.
✅ Stay realistic with expectations – AI is great at imagery, still developing with humor. You may have to still work on that one.
✅ Alignment over perfection – Having a coherent look across concepts was a huge step forward. It does not mean images were perfect, but they fit the style and concept I was after.
What You Can Try at Home
If you’re curious about image generation but frustrated with the results, here’s what to do:
1️⃣ Define your visual style before you start—think colour, tone, texture, and mood
2️⃣ Use tools like DALL·E’s select/edit function to fix details instead of giving up. You can try removing elements, changing colours, etc
3️⃣ Test small variations of one concept to explore how consistent the results are
4️⃣ Don’t aim for perfect—aim for aligned. Especially for internal use or learning support
5️⃣ Play with comic or storytelling formats—but know it may still need your human touch
Final Thoughts
This time, AI didn’t just generate pretty pictures or simple logos. I am using to help me communicate concepts. It brought visual unity to a presentation and it responded to my prompts. Thanks to these small tools, it showed me that with the right knowledge, even visual storytelling is something I can explore—without needing to be a designer.
Thanks for joining me on this episode of Try AI for Growth. Don’t forget to subscribe, leave a review, or share how you’re using AI to build, create, or simplify your work and life. Until next time—keep experimenting, and keep having fun!