There was once a time, not too distant in the past, where building a visual meant heading to Google, sifting through search results to find something close enough, and then dragging it into PowerPoint to tweak colors, resize elements, and hope for the best. Depending on the complexity of what you were trying to illustrate, this process could eat up anywhere from thirty minutes to several hours of your day.
That time is over.
---
The New Way: Talk to Your AI
Large language models capable of image generation (think ChatGPT, think Gemini, and specifically Gemini's Nano Banana model, a personal favorite) have completely changed the game. You no longer need to hunt for the perfect stock image or wrestle with design software. You simply describe what you want.
And here's where it gets interesting: one of the most powerful ways to do this isn't by typing. It's by speaking. (Learn more about how I think about voice in the age of AI.)
All that messiness in your head, the half-formed ideas, the "I know what I mean but can't quite say it" moments... it all gets downloaded into a transcription system. The AI then takes that raw, unfiltered thinking and reasons over it, making it neat, tidy, and actionable.
---
The JSON Trick That Changes Everything
Here's the real magic of this workflow:
Once you've talked through what you want your image to look like, you ask the AI to generate a JSON image prompt depicting what you described.
"Now generate a JSON image prompt to depict the workflow I just described."
The AI writes a structured image prompt for you. You feed it back into the model, click Create Image, and out comes a beautiful, accurate visual.
::: callout {icon="💡" color="blue_bg"}
What is JSON? It stands for JavaScript Object Notation, a coding concept, but don't let that intimidate you. Even if you're not a programmer, all you need to know is that JSON is an organized, structured way of representing information. Think of it like structured data in an Excel spreadsheet. It gives the AI a methodical framework for generating your image rather than interpreting a loose, freeform description.
:::
---
General vs. Specific Prompts
You can be as general or as specific as you need to be, and both have their place:
- Less specific = more generative. Each time you use the prompt, the output may look different. Great for exploration and creativity.
- More specific = more consistent and controlled. Better when you need precision.
After the first image generates, you can also iterate. Simply tell the model what you'd like changed, and it refines accordingly.
---
A Real-World Example
A high-level technologist at my company recently asked whether we had a particular business process documented. Honestly? Probably not on paper. But I knew it. It lived in my head.
So I talked through it. I got the messiness out. I asked the AI to generate a JSON image prompt from my description. A few minor iterations later, I had a clean, professional workflow diagram.
What would have taken at least an hour in Lucid or another diagramming tool took me no more than fifteen minutes.
---
Why This Matters
This workflow unlocks possibilities that simply weren't accessible before. Not just for developers or designers, but for anyone who has ideas in their head and needs to get them out visually. Workflow diagrams, infographics, process maps, concept illustrations... all of it becomes dramatically more accessible when you can just talk about what you want.
The barrier to visualization is gone. All you need is your voice.