7 Real Tests of Google's Nano Banana Pro (With Master Prompt)
I put Nano Banana Pro through 7 professional tasks: infographics, text rendering, style matching, and more. Before/after results + a master prompt you can copy.
You know that moment when you’re trying to create a visual with AI and the text comes out looking like alphabet soup?
Distorted letters. Words that make no sense. Headlines where half the letters are backwards or melted into each other.
I’ve seen this a hundred times. And if you’ve tried creating visuals with text in AI, you’ve seen it too.
You spend 20 minutes getting the layout and colors right, then another 20 in Canva fixing every tiny text issue because the model couldn’t render letters properly.
At some point you wonder why you’re even using AI if you’re spending more time fixing things than you would’ve spent just designing the whole visual yourself.
This has been the biggest frustration with AI image generators from the start. Beautiful visuals, mostly unusable text. Even though Nano Banana (Gemini 2.5 Flash Image) made progress, it still didn’t get it right every time.
Until about a week ago, when Gemini 3 launched.
If you haven’t followed the updates, Gemini 3 is Google’s new model family built to handle reasoning, planning, and more complex tasks in general. If you want to see everything else that came with the release, you can check the launch video or Google’s announcement article.
But for today, we’re zeroing in on Nano Banana Pro and testing together how well it handles tasks that older models kept failing at. Nano Banana Pro is part of a much bigger picture.
Nano Banana Pro is part of a much bigger picture. Google now has over 36 AI tools and products across research, creative work, automation, and more. I mapped all of the Google tools ecosystem here.
Nano Banana Pro: Perfect text rendering meets reasoning
When I first covered Nano Banana (Gemini 2.5 Flash Image) for AI Supremacy (you can read the deep-dive here), text was still the weak spot.
You could get a good layout or a solid concept, but the letters were always the part you had to fight with. Sometimes the text came out fine, sometimes it didn’t, and you never knew which version you were going to get.
Nano Banana Pro (Gemini 3 Pro Image) finally fixes that.
It’s the first time I can say, without hesitation, that you can create any visual you want without worrying about the text falling apart.
And because it’s built on Gemini 3 Pro, it doesn’t just generate visuals with perfectly rendered text. It also brings reasoning and real-world knowledge into the mix.
Which means it can pull real information when you need it, follow your brand style from a couple of rough references, keep diagrams accurate, and handle complex layouts without breaking anything.
You also get the studio-level controls you’d expect from a proper design tool, so you can tweak lighting, angles, consistency, and the overall look without the usual friction.
And we’re about to test all of it.
Test 1: Can it recreate professional visuals with selective modifications?
The first thing I tried was something past models really struggled with: Can it look at an existing visual, understand the structure, and recreate it with only the changes you specify, without making it look AI-generated?
It all started with one of my readers who had tried this with other AI models before. He wanted to recreate IBM-style visuals but with different text and colors, and every attempt came out distorted in a way that made them unusable.
Here’s the original image he wanted to recreate:
And here’s what ChatGPT produced just two months ago:
Now with Nano Banana Pro, this became actually doable:
It kept the structure, changed the text and colors as requested, and maintained the professional look. The only thing it modified was adding arrows instead of circles between elements - but the overall recreation is accurate and doesn’t have that “AI-generated” look.
Here’s the prompt I used: https://gemini.google.com/share/8e75e4eb862a
Test 2: Can it remove text while preserving structure?
Another challenge he faced was taking an existing diagram, removing all the text, improving the resolution, and keeping the structure exactly the same so he could add his own content later.
This is trickier than it sounds. Consulting-style diagrams like McKinsey’s have complex layouts. You need clean extraction at good resolution so you can use them as templates in your own presentations.
Here’s the original diagram:
This was the failed attempt from two months ago:
And here’s what Nano Banana Pro produced now:
So much better. It did require some iteration from the original prompt, but it’s clearly a massive improvement. The structure is intact, the resolution is clean, and the text is removed. Ready to be a reusable template.
Here’s the prompt I used: https://gemini.google.com/share/5d54dda0692f
Test 3: Can it actually render text correctly?
Next, I tried a simple request: an IBM-style infographic on ‘The 6 Pillars of Enterprise AI Adoption’. Nothing fancy, no instructions. I just wanted to see if the text would come out right and if it could pull off that IBM look on its own.
Result:
Perfect text. Not a single broken letter. Clean, readable, on-style, and it even added the IBM logo.
Here’s the prompt I used: https://gemini.google.com/share/d2a5936f86f7
Test 4: Can it match a reference style with just a logo?
For this one, I gave it a reference infographic as an example of the style I wanted, plus my logo, and asked it to create an infographic with ‘6 Steps to Get Started With AI as a Consultant’ using the exact text I provided.
Result:
It didn’t keep the exact layout of the other infographic, but it understood the style and made it look closer to it, using only my logo for the branding.
And for anyone who doesn’t have a full visual system in place yet, this is huge. You can basically start defining your brand style from scratch, even if you only have a logo and a handful of scattered posts like I do.
Here’s the prompt I used: https://gemini.google.com/share/5bfcc0ccc7a3
Test 5: Can it create infographics from a lot of text without guidance?
This is where I wanted to test something harder: Can it read a lot of text, figure out what matters, and create an infographic without me telling it what to include?
A while ago, Melanie Goodman asked in a comment on one of my posts about creating infographics with AI. So this time, I took one of her articles and two random images from her LinkedIn profile for color and style references.
Result:
It figured out the key points from the article (which is a whole lot of text) without any guidance, organized them logically, respected Melanie’s visual style and colors, and rendered all the text perfectly.
Here’s the prompt I used: https://gemini.google.com/share/d4e88428b193
Test 6: Can it translate a process into a complex visual?
For this one, I wanted to push it into something completely different: an isometric 3D illustration that visualizes my four-phase expertise borrowing framework I shared in this article.
This isn’t a simple diagram. It requires understanding the logic behind each phase, turning them into scenes, and then connecting those scenes into one coherent isometric world.
Result:
It nailed it. And I only gave it the four phases and a short description of each one so it could translate the ideas into visual elements. It understood the structure, kept the scenes consistent, and turned the whole thing into a clean, readable 3D illustration that actually represents the workflow.
Here’s the prompt I used: https://gemini.google.com/share/2c17af650819
Test 7: Can it generate alternatives with the same structure but different aesthetics?
I thought this would be challenging. I used the exact image from the previous test as a reference, even though I didn’t want the model to copy its look. I asked it to keep the same four-phase processbut reinterpret it in a completely different aesthetic as a storyboard.
Result:
It handled the shift better than I expected. It kept the underlying structure, ignored the original style, and turned the whole thing into a storyboard that actually captures how humans and AI work together when you’re stepping into fields where you have no expertise.
Here’s the prompt I used: https://gemini.google.com/share/bb713255d295
What this upgrade means for how you work
If you create visuals with text - newsletters, LinkedIn posts, presentations, reports, ads, marketing materials - this changes your workflow.
Not in a “everything is different now” way. In a practical “I don’t need to do this manually anymore” way.
Things you can stop doing:
Spending hours in design tools. Generate options in minutes, pick the best one, make minor tweaks if needed.
Hiring designers for simple visuals. Save that budget for complex work that actually needs a human designer’s judgment.
Wrestling with AI outputs. When the text actually renders correctly, there’s way less to fix.
Things you can start doing:
Prototype marketing materials that match your brand style consistently. Ad concepts, poster designs, social media graphics - create dozens of options in the time it used to take to make one.
Create diagrams for work presentations. Flow charts, process diagrams, organizational charts, data visualizations - all the graphics you need for decks and reports, generated in minutes instead of hours in PowerPoint.
Recreate any visual you want. That diagram from a McKinsey report? That infographic style you saw on LinkedIn? That design aesthetic from a brand you admire? Any visual you like can be recreated with your own content and branding.
Ready to test it on your own?
If you want to play with it on your own and recreate any of these examples, you can find it here:
Gemini app or gemini.google.com: Select “🍌Create images” with the “Thinking” model
Google AI Studio: For developers
AI Mode in Google Search: For Pro and Ultra subscribers (US only for now)
Google Workspace: Rolling out in Slides and Vids
Also available in NotebookLM, Google Ads, Vertex AI, and Flow
Free users get limited attempts before it switches back to regular Nano Banana. If you have Google AI Plus, Pro, or Ultra, you get higher quotas.
How to prompt Nano Banana Pro
Getting access is one thing. Getting the results you want is another. So let’s talk about how to actually ask for the visual you want.
To help with that, I put together a master prompt you can use to create perfect prompts for Nano Banana Pro. It’s based on Google’s image generation best practices and my own testing.
How it works: The prompt first deconstructs your request to understand what type of visual you want to create, then applies the specific prompting rules that match that type. This way you get the best possible output without needing to know all the technical prompting details yourself.
How to use it: Drop it into any conversation with your preferred AI model (ChatGPT, Gemini, Claude), or create a CustomGPT (here’s how) or a Gem inside Gemini (here’s how) so you can reuse it anytime without repeating the prompt.
Copy my prompt: