AI image generation: concrete examples with Nano Banana Pro

Since the release of Nano Banana Pro, which runs on the Gemini 3 reasoning model, AI image generation has reached a new milestone.

This milestone makes it possible to use AI for content creation, and I am going to share a series of concrete examples. But first, I wanted to come back to 2 points:

1. Why does Nano Banana succeed where other models fail?
2. How to simplify image generation with a prompt generator?

As a reminder, if you want to use Nano Banana Pro, you can go through Gemini, then select “Create Image” with the “thinking” model.

1/ Why does Nano Banana succeed where other models fail?

The problem with older models came from their inability to respect the structural and logical constraints of our environment, in other words to reason visually.

Here, by using a reasoning model during generation, image generation is constrained to respect the spatial layout and the logical relationship between visual elements.

The model uses different generation logics depending on the prompt and manages to produce structured visual artifacts that are semantically coherent (a little less so in French) and visually legible.

As we will see, the model is also able to handle:

• layout: coherent spatial structures (keeping grids and visual hierarchies), which makes the infographic legible
• typography: clean, legible text, even at small sizes, with formatting preserved.
• data visualization: turning numbers into charts and dashboards, like a spreadsheet or Figma would
• the style universe: aesthetic consistency of elements, style and color palette, able to recognize brand elements (logo) and reuse them.

2/ Your Nano Banana Pro prompt generator

Google specified that the model was much weaker when the request was short and vague, such as “create an infographic”, and that you should instead generate complete, very precise instructions, as if you were briefing an expert design intern on their very first day.

But to generate advanced instructions, you might as well use AI and an assistant that will build a complete brief. (this assistant is based on Google's official recommendations)

I created it on ChatGPT, available to everyone for free; it gives you an optimized prompt in English in both a text format and a JSON format (more effective). Here is the link to the assistant :

For example, here are the 2 results on Gemini with the text prompt and the JSON one:

In general, it is worth structuring your prompt with:

• the type of image to generate (dashboard, web page, infographic, etc.)
• the layout → how the elements should be arranged on the image
• the list of required components
• the list of constraints at the end: “the text must stay sharp even at small sizes”, “use a consistent icon style”

The new use cases

You may have noticed a rise on LinkedIn of infographics, diagrams and dashboards generated by AI. Whereas normally you would easily pay 100 to 200€ for a complete one-pager on a topic.

Here are a few examples I am sharing to show you the range of possibilities:

A diagram of how an LLM predicts the next word:

Turning a fairly complex post on X into a diagram worthy of a teacher's explanation on a blackboard (in English)

Generating a comic strip with consistency across scenes (and without mistakes)

Translating content directly on the image

Turning a blurry photo into an ultra-sharp, realistic one without changing the composition or the objects.

Creating an image with the different elements that make up the shoe on the left. Laying out the elements individually on a neutral surface.

Adding illustrated bulging eyes and sticking hands onto (the object/the animal/your best friend) in an exaggerated comical pose.

Generating a map of Paris with 3D monuments

The possibilities are really broad when it comes to image generation

Slides via Gemini's “build” mode

Here is a short demo of slide generation on Gemini simply by sharing a study. Nano Banana plays a minor role here, but it was a good chance to share this use case.

Generating a photo of yourself

Recently, I shared on LinkedIn how you can generate photos of yourself for free.

Getting a nice, up-to-date photo of yourself is often a hassle: decent outfit, so-so lighting, 12 forced poses that look like nothing… And in the end, an “ok” photo, chosen by default.

Now AI is changing our relationship with images: we choose the style while staying behind our screen.

Here is how to do the same thing:

1. Open Gemini (to use Nano Banana)
2. Import a photo of yourself (a selfie for example)
3. Paste one of the prompts
4. Choose “Tools” and “Create images”, use the “Reasoning” model
5. Adjust and iterate

(Tip: use a photo taken facing the camera, with a simple background)