I tried using Nano Banana Pro to create complex infographics

Did it actually work?

Disclosure: This article was written by me, a human, with an assist from Sam Illingworth, also not a bot, who provided a genuinely interesting research paper to serve as a guinea pig for my infographic adventures. Claude helped me regenerate style guides and edit my very messy first draft. Nano Banana Pro generated all the images.

Confession time. I’m a pretty good writer, and I’m quickly getting better at coding. But I’m a terrible graphic designer.

Until recently, my approach to branding my newsletter was simply to do nothing. I would randomly generate AI images at the last minute with no style guide, no reference images, and no serious thought. The results were...fine? They filled the thumbnail slot, but they didn’t create any kind of visual coherence or brand recognition.

I also carried the same approach over to LinkedIn, where I have occasionally spent hours generating mediocre carousels and info-slides. And most of these efforts, perhaps not surprisingly, melted away into the void. (My posts typically get about two hundred impressions, a save or two, and the occasional polite comment from someone I already know.)

The only thing I’ve gotten better at designing is user interfaces, and that’s because they’re made of code. It also helps that Tailwind has preprogrammed styles for everything now.

Why I decided to care about visuals

There’s something about consistent visuals that signals competence. When I see a newsletter with a recognizable style—the same illustrated avatar appearing in every thumbnail, a consistent color palette, typography that feels intentional—I immediately trust it more. It suggests the person behind it cares about details and has their act together.

I wanted that for myself. I’m planning some outreach to people I don’t know, both through referrals and the dreaded cold email, to promote Future Scan, the software application I’m hoping to share soon with researchers at corporate labs and universities.

So I decided to pay more attention to cultivating a coherent visual brand across my Substack newsletter, LinkedIn profile, and the app itself.

A new tool: Nano Banana Pro

Nano Banana Pro is Google DeepMind’s latest image generation and editing model, built on Gemini 3 Pro. You access it by opening the Gemini app or visiting the home screen, selecting the banana icon from the tools menu, and choosing “Thinking” from the model menu. (The standard “Fast” option gives you the original Nano Banana, which is powered by Gemini 2.5 Flash.)

What makes Nano Banana Pro interesting for my purposes is its ability to generate images with legible, well-placed text, which is ideal for infographics, diagrams, and posters. It can also connect to Google Search’s knowledge base to, at least in theory, create “context-rich” visuals based on real-world information.

According to Google, the model supports multiple aspect ratios, can handle up to 14 reference images for style consistency, and outputs up to 4K resolution. It’s also being integrated into Adobe Firefly, Photoshop, and various enterprise tools.

But, I wondered, would it actually work for someone like me, a solo developer who needs “good enough” graphics without hiring a designer? That’s what I set out to test.

Starting with a basic brand

I wanted a consistent color palette and a consistent avatar to use in all my Substack articles, similar to how writers like Jenny Ouyang, Karo (Product with Attitude), and Daria Cupareanu handle their visuals. They have immediately recognizable illustrated avatars that appear across all their thumbnails, creating instant brand recognition. You see the avatar, and you know whose newsletter you’re looking at.

I picked the teal I already used as my accent color in Future Scan and had Claude build a visual style guide based on the “vibe” of my app. The guide specified things like clean and minimal aesthetic, generous white space, teal (#0d9488) as a sparingly used accent on a grayscale foundation, Inter typeface, simple geometric illustrations, and a data visualization aesthetic that feels scientific without being cold.

My new avatar, courtesy of Nano Banana. I created five different versions of the character to cover different expressions and poses.

I then referenced this style guide and prompted Nano Banana to create a simple, illustrated avatar I could use as a reference image for multiple Substack thumbnails. After a few iterations (adjusting the style, asking for simpler linework, refining the color palette), I got something that worked.

Then I fed Nano Banana several reference images of the avatar in different situations plus thumbnail dimensions, and asked it to create “narrative thumbnails” for each of my old Substack articles. While the exercise required some back-and-forth, it worked surprisingly well—and it was fast. What would have taken me a day or more of fumbling in Canva took maybe two hours of iterative prompting.

One of many thumbnails with my consistent avatar and, of course, lots of teal

My Substack newsletters all now have consistent images. This went so well, I wondered what else Nano Banana could do.

An inexplicable LinkedIn mini hit

I’ve been struggling to get views on LinkedIn. Most of my posts get around 200 impressions and zero saves, the engagement equivalent of shouting into the void. But I have access to lots of research papers through Future Scan, so I decided to try something different: creating an infographic based on a research paper about how well people who use AI can detect AI-generated text when scanning copy.

A very imperfect yet strangely effective social image

The resulting infographic wasn’t that great, objectively speaking. The design was iffy, with lots of colors that didn’t quite harmonize. Nothing about it was cohesive or on-brand. But something about it caught attention: it got 4,000 views and 5 saves. For me, that’s roughly 20x my normal engagement.

That unexpected success made me curious. Was this a fluke because the topic (AI detection!) just happened to resonate? Or could Nano Banana Pro actually create usable infographics for social media? It was time for a proper test with controlled variables.

The hypothesis: Nano Banana Pro can create usable infographics for social media.

I decided to test Nano Banana systematically, trying a variety of prompting strategies from simple to complex, with different amounts of background information and iteration. Sam Illingworth, who writes the Slow AI newsletter and is a professor and science communication researcher, volunteered one of his papers as the test subject: “Rhyme and Reason—Using Poetry to Talk to Underserved Audiences About Environmental Change,” published in Climate Risk Management.

The paper explores how poetry workshops can help underserved communities express their thoughts about environmental change. Through an interpretive phenomenological analysis of the poems participants created, the researchers found that poetry helped level hierarchies between “experts” and “non-experts,” allowing community members to express their fears, resignation, and hopes about climate change in ways that traditional science communication methods often fail to capture.

It’s a genuinely interesting piece of qualitative research, and importantly, it has enough substance to make a meaningful infographic. The question was whether Nano Banana Pro could translate that substance into a visual format.

Each test was conducted in its own separate chat, so tests weren’t affected by previous tests’ histories.

Test 1: The ultimate lazy approach

Method: One-line prompt + PDF attachment

I started by uploading the PDF of Sam’s paper to Gemini and asking Nano Banana to “please create an infographic based on this paper.”

This immediately caused its Thinking Mode to crash; I tried six times just to be sure. Most likely, the complete lack of direction caused it to explore too many possibilities at once, thereby torching its entire context window.

Pro Tip: Be sure to choose the banana tool BEFORE you send your prompt, otherwise you won’t get an image, you’ll just get text. This tripped me up multiple times.

Test 2: Two-step process

Method: Generate copy first, then request infographic

I gave Gemini the paper and asked it to write copy for a narrative infographic that pulls out the key points.

After I got what I thought was acceptable copy—and understood a lot more about the paper’s methodology and findings in the process—I prompted Nano Banana to create an infographic and copyedit the text to fit as needed. I didn’t specify an aspect ratio; I wanted to see what it would do by default.

The prompt: Use this copy to build a vertical scrolling infographic that communicates the key points concisely. Edit the copy as needed to suit the design. Include a hook, main insights, and a takeaway.

The result: It produced a very colorful and busy cartoon with all the key information included. The text was tiny and hard to read, but...it was on the edge of being “good enough” for social. Honestly, I probably would have considered posting it on LinkedIn.

It’s vibrant and tells a story…but it’s really busy and hard to read

Bonus experiment: I asked it to recreate the graphic as a 9:16 (portrait) infographic, and it did...okay. The layout was crowded, and “Lived experience” became “Liver experience.” This is a classic AI text rendering mishap and why you always proofread.

Same concept, different aspect ratio. Plus “Liver Experience”

Aspect Ratio Note: When asking for an infographic, be sure to specify aspect ratio or you’ll get a landscape rectangle by default. Nano Banana Pro supports 1:1 (square, ideal for social feeds), 9:16 (portrait/vertical, perfect for TikTok and Reels), 16:9 (landscape/wide), 4:5 (taller Instagram format), 3:2 and 2:3 (classic photo ratios), 21:9 (ultra-wide/cinematic), and 9:21 (extra tall for banners and scrolling infographics).

Test 3: Three-step process with style guide

Method: Copy + style guide + detailed prompt with aspect ratio

Next, I simplified the copy and added more guidance to the prompt, such as: include enough white space at the top and bottom, use the copy exactly as written (don’t improvise), and make it a tall 9:21 aspect ratio for scrolling.

I also added my detailed style guide, the clean, minimal one I use for Future Scan, with its teal accents, gray foundation, and emphasis on simplicity.

The prompt: Create a vertical scrolling infographic (9:21 aspect ratio) using the copy below. Follow the attached style guide exactly. Include generous white space at top and bottom—at least 15% of the canvas height on each end. Use the copy exactly as written; do not add or modify text. Keep the layout clean and uncluttered.

The result: Now we were getting somewhere. The output was getting cleaner, more professional, and easier to read.

But there was a problem. The shaded boxes were not a single uniform color and appeared to have odd ripples. And no amount of prompting could make them go away.

The persistently blotchy background. Had I found the fatal flaw?

Test 4: Same approach, opposite style guide

Method: Testing whether style affects output quality

Curious to see if I would get the same quality results with a dramatically different aesthetic, I asked Claude to “study this style guide and build a style guide that is the original’s aesthetic opposite.”

The result was a warm, photographic style: Nunito and Playfair Display fonts instead of Inter, coral and terracotta instead of cool teal, rich saturated colors instead of minimal grayscale, and an emphasis on lifestyle photography and organic, rounded shapes rather than geometric precision.

The prompt: Create a vertical scrolling infographic (9:21 aspect ratio) using the copy below. Follow the attached style guide exactly: warm colors, rounded typography, rich photography aesthetic. Include generous white space at top and bottom. Use the copy exactly as written.

The result: Quality was similar to Test 3. The infographic looked well-constructed and true to its radically different style. But I still got weird, uneven color blocks, a persistent issue across all my tests that seemed unrelated to the specific style guide. The model seems to struggle with maintaining consistent fill colors across different shapes.

The fill color behind the copy is subtly blotchy.

Test 5: Stripped-down monochrome

Method: Short copy + directive prompt + monochrome style + no solid colored shapes

Next, I went for a totally stripped-down appearance: very simple and basic, with outlined shapes only and no solid color fills that AI could mess up. The idea was to eliminate the color consistency problem entirely. Any color could be added later in Adobe Express or Canva as needed, as simple overlays on a clean black-and-white base.

The prompt: Create a vertical scrolling infographic (9:21 aspect ratio) using the copy below. Use a monochrome color scheme: black, white, and grays only. Use outlined shapes only; do not use any solid color fills or shaded backgrounds. Keep the design minimal and clean with generous white space.

The result: This was successful in solving the color block problem—no weird fills, no inconsistent shading. It was basically a usable infographic and the closest yet to something ready to post without modification.

But there were text inconsistencies: sometimes a line return appeared after “Insight” blocks, sometimes not. Text embedded in images is a pain to edit once generated, and Nano Banana could not get this perfectly consistent across all sections.

Copy issues introduced by the model include periods outside of quotations. This can be fixed by prompting. The inconsistent line breaks, however, were resistant to specific instruction.

Also, there still wasn’t quite enough white space, despite my explicit instructions. The model seems to want to fill available space.

My approach to prompting = Less is more: LLMs are non-deterministic; a long prompt gives them more different things to fixate on, leading to less consistent results. Context window is also a concern for very verbose prompts, especially if you’re providing extra context like a style guide. In my experience, it helps to keep your instructions tight and prioritized.

Test 6: Pulling it all together

Method: Short copy + brand-specific style guide + emphasis on white space

For the final test, I had Claude create one more style guide, this time based specifically on colors and imagery from Sam’s Slow AI newsletter.

The guide emphasized: calm and minimal aesthetic (”visuals should feel like a deep breath”), generous white space (minimum 15-20% of canvas height on each end), outlined shapes only with thin clean strokes, soft neutral colors (warm white, light gray, cream) with one muted accent, and a hand-drawn feel. The overall directive: “Someone viewing the graphic should feel calmer, not more stimulated. Go slow.”

The prompt: Create a vertical scrolling infographic (9:21 aspect ratio) using the shortened copy below. Follow the attached Slow AI style guide exactly. Prioritize generous white space—minimum 15-20% of canvas height at top and bottom. Use outlined shapes only with thin strokes. Keep it calm and minimal.

The result: At first glance, this looked pretty good. It had a reasonable amount of white space and the blotching on the background was relatively subtle. But the copy lapsed into nonsense: Bristor, Fecilitators, neuse resiguation, etc.

Nice design, but the background is subtly blotchy, and the copy lapses into AI word salad.

Should you leave the Gemini icon? Actually, I think yes—in the interests of transparency. While images created with Nano Banana Pro do include a SynthID digital watermark (like those generated by some earlier versions), being upfront about AI-generated content seems like the right call, given how strongly some audiences can feel about AI-generated visuals.

So, what’s the verdict?

For my purposes—LinkedIn posts, Substack notes, maybe presentations—I believe Nano Banana Pro is “good enough” for me to quickly create usable images, or images that are close enough that I could fix them in post-processing with minimal effort.

I also think I could have gotten even better results in this test if I had just kept removing copy. Verbosity is the biggest issue. The simpler and shorter the content, the cleaner the output. If you find yourself cramming in lots of text, you’re probably asking for trouble.

And, it might also be helpful to standardize the infographic structure. For a research study, it might be something like:

• What we’re testing

• The test design

• What we expected

• What we discovered

• Final thoughts

However, the outputs are not pixel-perfect and may have defects like inconsistent spacing, occasional text errors, and uneven color fills. Some of this is likely forgivable coming from me, a solopreneur posting to LinkedIn, but may not be as well-received from a big company that can afford human designers.

What you can do with Nano Banana Pro

• Give it infographic copy and get a surprisingly decent draft

• Specify aspect ratio upfront to get roughly the format you need

• Specify preferred graphical treatments and mood through a clear style guide

• Include photos and illustrations as reference images for consistency

• Continually improve by tweaking copy length and prompts

What you can’t do

• Get images at the exact pixel dimensions you need for specific social formats

• Get pixel-perfect results that require no editing

• Get very high-resolution files consistently (despite the 4K promise, results vary)

• Guarantee text will render correctly every time—always proofread

• Create complex, information-dense infographics without at least a few glitches

• Maintain perfect color consistency across solid shapes

Final thoughts

Nano Banana Pro isn’t going to replace graphic designers, and it’s not going to give you polished, brand-perfect assets out of the box. But for creators and solopreneurs who need “good enough” visuals quickly—especially for social media content that has a short shelf life anyway—it’s a genuinely useful tool that can save hours of fumbling in Canva or begging designer friends for favors.

The key lessons from my experiments: keep your copy short (shorter than you think), provide a clear style guide with specific direction, specify your aspect ratio upfront, and embrace the need for some post-processing. Don’t expect perfection; expect a solid draft that gets you 80% of the way there. That remaining 20% is often just a few minutes in Adobe Express or Canva.

And always, ALWAYS proofread carefully.

Thanks again to Sam Illingworth for volunteering his paper as the test subject, and for being a good sport about having his research on poetry and climate change turned into a wide variety of less-than-perfect infographics.

If you’re interested in mindful approaches to AI, check out his Slow AI newsletter.

For further reading

How I Create Consistent Hero Images, And Why I Haven’t Switched to NanoBanana by Jenny Ouyang

Visual ASMR: How to Make Oddly Satisfying Christmas Designs with Nano Banana by Karo (Product with Attitude)

🍌 Pro by Ruben Hassid

Complete Nano Banana Pro Prompting Guide by Anfernee

I Can’t Believe AI Can Do This Now (Hands On with Nano Banana Pro) by Daria Cupareanu