If you grew up in the 90s, you're probably familiar with the Japanese anime phenomenon Dragon Ball Z; it was everywhere. Gohan and Goku - two of the protagonist - made us all feel like heroes. I'd head out to our compound for horseplay with my brother, but not before practicing my Saiyan transformation: feet firmly planted, fists clenched tightly, head down with eyes focused, back slightly hunched, chest tight, letting out the loudest roar I could muster (IYKYK).
My older brother had a more artistic way of immortalizing Goku. He was a talented pencil artist who could create realistic drawings of a Saiyan. Like many younger brothers, I looked up to my elder brother and wanted to be like him but had zero artistic ability - That gene skipped right over me. I couldn't even draw a perfect circle to save my life.
So when ChatGPT recently released its advanced image generation model, while everyone else was busy creating Studio Ghibli art, - a unique visual style developed by the renowned namesake Japanese animation studio - my first instinct was to finally create that Saiyan Gohan drawing I'd always wish I could. I was completely blown away. For a moment, the young, inartistic kid in me believed I was Da Vinci.
Vote and then read to the end to find out which of these drawings was created by ChatGPT
I wasn’t alone, the ChatGPT image generation model release created such a viral moment that OpenAI CEO Sam Altman implored folks to “chill on generating images because his servers were melting”
Beyond Studio Ghibli art and my nostalgic attempt to relive my childhood, the real question is: can AI help small business owners create professional ad graphics and logos that can be used commercially? The short answer is not quite yet except you know exactly what you want and have some technical background of design elements.
AI Claim vs Reality
Most small businesses today use social media as their primary distribution channel and spend hundreds, sometimes thousands of dollars monthly with agencies for logos and graphic designs. So the promise of a tool that can create professionally designed logos and ad assets for a fraction of the cost is certainly exciting.
Two weeks ago I wrote about smart guesses and how LLMs generate text. AI image generation works differently than text models, but the core concept is the same; AI is still making smart guesses. When you type "create an image of a kid playing in the mud and eating ice cream," the AI doesn't actually understand what kids, ice cream, or mud are. Instead, it's been trained on billions of images paired with text descriptions, learning what these things typically look like.
Another way to think of this is; imagine you're a designer and I showed you hundreds of thousands of event planning agency logos while describing each one out loud. Eventually, you would start recognizing patterns. So, when a client asks you to design a logo for their new event planning business, you'd have context to create the best approximation of what it should look like. That's exactly what AI does, but at a massive scale.
But can it create a visual image that actually works for your specific brand? That's what I set out to test this week.
My goal was to recreate the logo of Rima, which was designed by a human using three AI tools - ChatGPT, Canva AI and Leonardo AI.
Original Rima Logo
1st shot
For my first test, I gave the three AI tools above a prompt that represented my best description of the logo, as I would explain it to a human designer (see the prompt below). The output was poor overall, with Leonardo AI performing the worst. This wasn't surprising given how image models work. The AI tries to map the content of my prompt to the closest approximation it can find based on patterns from all the images and descriptions it was trained on, then creates what it thinks is the best fit based on its training data.
Prompt: Create a simple and minimalist logo design that’s elegantly beautiful and captures the futuristic nature of AI technology. The logo should be large and circular. The circle should have a smooth gradient that transitions from white to light gray towards the edges to a mix of purple and lavender closer to the center. Again I’d love the overall look and feel to be clean modern and have no other additional design elements beyond what i have described.
2nd shot
To get closer to what I wanted, I decided to give more explicit context by sharing the exact image of my logo and asking ChatGPT to modify its previous output to match it exactly (see the follow-up prompt below). The other AI tools - Canva and Leonardo AI - don't support image to image generation, so I could only test this approach with ChatGPT. This improved performance significantly and created a close approximation of my original logo, though the color shade and gradient density were still slightly off from the original.
Follow up prompt: Extract all the visual elements from the attached picture and use it to update the last image you just generated to look exactly like the logo image i attached
3rd shot
For my third shot, I tried something different. I uploaded my original logo to Claude and asked it to create the exact prompt that could replicate the image on other platforms like ChatGPT, Canva AI, and Leonardo. Then I used that exact prompt across all three tools. It was interesting to see how specific and technical Claude's prompt instructions were. I was also surprised by the huge variation in output across the three tools. ChatGPT's output was very impressive, though I noticed that using the same prompt three different times gave me three different results, similar to each other but with slight deviations in color and shadow. Leonardo and Canva AI performed poorly with one-shot image generation prompts across board.
Claude prompt: Create a minimalist SVG-style image of a large, centered circular shape on a light background. The circle should have a soft radial gradient with these properties: Center color: #9B59B6 (a medium lavender-purple) Outer color: fades smoothly to white or fully transparent at the edges Gradient type: radial, smooth and subtle Background: pure white or very light gray (#F5F5F5) Design style: modern, clean, and ethereal — no text, borders, icons, or embellishments Effect: the circle should appear like a softly glowing, diffused light source Format the image as a square or circle-centered composition suitable for logo use.
The Verdict
So is the claim that AI can design a professional logo for you that can be used commercially accurate? Not quite yet. It depends heavily on the tool you use and your ability to be very specific about what you want.
Today, I believe ChatGPT's image generation works best as a brainstorming partner for experienced designers who can write detailed, technical prompts and iterate quickly on ideas. For small business owners with limited design experience, AI can spark some ideas, but you'll still need to work with an experienced designer to create something truly professional and commercial-ready.
ChatGPT was clearly the best performer among the tools I tested, producing significantly better results than Canva AI or Leonardo. However, even with the best prompts, the outputs from chatGPT still had inconsistencies in color, shadow, and overall polish that would matter for professional use.
So, for small business owners, my recommendation will be to use AI for initial concept exploration, but partner with a designer for final logo. The technology is advancing rapidly, so this gap will likely close soon but we're not quite there yet.
Pro Tip
Use AI to enhance your prompting skills. If you find a picture, design, or color scheme online that you love, upload it to your favorite LLM chatbot and ask it to create a prompt that other AI tools can use to recreate the design. Then use that prompt in ChatGPT or other state of the art image generation AI tools to recreate it, and use that initial design as a canvas for different iterations with follow-up prompts.
The answer to the poll question is A. Did you get it right? 😊