A poster and a portrait fail differently. A weak illustration still gets read. A wrong word makes the whole image wrong. Google publishes rules for the words themselves, and that is the part most prompt collections skip.
Nano Banana Pro infographic and poster text: the short answer
State the copy, state how it should look, then state the layout. Google’s own advice is one sentence long: “Clearly state what text should appear and how it should look.”
Leave any of those three out and the model invents them, spelling included.
What Google says the model does with text
The model page opens its text section with the use case. The model can “generate clear text for posters and intricate diagrams, then translate your designs for different locales”.
It also answers the typeface question: “Describe what type of font you want, or simulate different types of handwriting.”
The prompting guide’s first tip makes the same claim: sharp, legible text is what makes “impactful posters, intricate diagrams, and even detailed product mockups” possible.
Then the limitations section cuts it back down. Rendering “small text, fine details, and producing accurate spellings may not work perfectly”.
The honest reading: strong on big deliberate type, unreliable on the small print.
Name the text, then describe how it looks
Four things per text element: the exact string, the typeface or mood, the weight or colour, and where it sits. The guide’s own example does all four for a single headline.
Quotation marks around the string matter more than any adjective. They tell the model these characters are content, not description.
Keep the strings short. A twelve-character headline has fewer places to go wrong than a full sentence, and the published poster examples are taglines, not paragraphs.
Give the layout a numbered structure
A wall of text is a layout failure before it becomes a rendering failure. Google’s own diagram prompts fight that with numbers. One asks for a flat lay where “simple, numbered labels—“01”, “02”, “03”, “04” - mark each stage of the process”.
Another builds the flow out of arrows: the visual story “flows from left to right in clear steps”, with hand-drawn arrows guiding the eye from the sun to the house. Borrow both habits. Number the blocks, then say which direction the reader’s eye travels.
When several panels are needed, ask for them one at a time. Two published prompts do that. One demands four separate images and warns that a single composite will not do. Another asks for ten versions “one at a time, each it’s own server call”.
Worked examples: three layouts you can copy
Three prompts built on sentences Google publishes. The notes mark which lines are the model’s own examples and which are the scene you supply.
Example one: a how-to infographic
Starting point: a process you can describe in four steps, with no figure that has to be exact.
Create an infographic that shows how to make elaichi chai. Flat lay photography on a clean light gray background. Numbered labels "01", "02", "03", "04" mark each stage, and simple hand-drawn arrows guide the eye from left to right. All text in a clean sans-serif font, correctly spelt. Format 16:9The first sentence is Google’s, verbatim. The rest copies the numbered-label and arrow habits from its other diagram examples. Check the spelling before the image goes public, because accurate spelling is a documented weak spot.
Example two: a poster that has to work in another language
Starting point: one poster concept, needed in two markets, with the tagline preserved word for word.
Create a poster ad for a sparkling water called 'Aura Fizz'. The setting is a London street at dusk with neon lights. The tagline should be "Taste the Aura". The can design should be minimalist. Make the aspect ratio 16:9.Then, as a second turn on the same image: Localize this concept to a Japan setting, keep the can exactly the same, making sure to translate the words "Taste the Aura" accurately and authentically
Both instructions are Google’s, verbatim. The localization pass is where grammar slips, so have a speaker of the target language read the tagline before it ships.
Example three: a data infographic with a factual constraint
Starting point: a topic where a wrong number is worse than a missing one.
A scientifically accurate cross-section diagram of the topic in the attached reference. Keep every label short and legible, use numbered callouts, and add no figure that is not in the source. Format 16:9The opening phrase is Google’s own example of a factual constraint. The guide pairs it with a warning: your inputs have to be factual too, because the model renders what you give it.
One caveat: a cross-section diagram is mostly fine details, which the model page lists as a weak spot.
When the numbers carry the meaning, check them
Infographics are sold on trust. A chart with a shifted decimal point is worse than no chart at all.
Google is explicit that this is unsolved. The model “may misinterpret information or produce factually incorrect results” when generating infographics, annotating diagrams or representing complex data.
The routine is short. Never let the model source the numbers. Put them in the prompt or the attached reference, then read each one back off the finished image.
Then read the small print aloud. Small text is the documented failure mode, and it is always the unit, the disclaimer or the source line that goes first.
For the rest of the prompting framework, the Nano Banana Pro prompt guide covers it, and the how-to guide shows where to start. The photo editor article handles images you already have, the Nano Banana Pro vs Nano Banana comparison explains the two generations, and builders should continue with the API guide.
Frequently asked questions
Will it spell my headline correctly? Usually, for short strings. Accurate spelling is listed as a weak spot, so read every word once at full size.
Should the prompt be in English when the poster is in Chinese? Not necessarily. English instructions with the target string quoted work, and the model page shows designs translated for different locales.
How many blocks should one infographic hold? Few enough to stay legible. Google’s own diagram examples use four numbered stages, which is a sensible ceiling.
Does a higher resolution fix small text? No. 1K, 2K and 4K are available, and resolution does not repair a letterform the model rendered badly.
Write the words yourself, then let the model draw them.