Nano Banana Prompts

Which Nano Banana you're using

«Nano Banana» started as the nickname of one model and is now Google's name for a family of them. A prompt written for the first one still runs on the newer ones, but the newer models read text, hold characters and combine references differently, so it helps to know which one answered you. As of October 2026, Google's own documentation lists five:

In practice, anything on this page that uses two uploads, or that you plan to refine over several messages, wants Nano Banana 2 or 2.1, or Pro. If the model menu says Flash-Lite, switch before you start.

How to write a Nano Banana prompt

The model works from sentences, so a list of comma-separated keywords leaves it guessing at how the pieces fit together. Google's own starting formula is a verb, a subject, an action and a scene, and the prompts here build on that with four more pieces.

1. Create or edit. «Create an image of» starts from nothing. «Use the attached photo» edits what you uploaded. Blur the two in one sentence and you can get a brand-new picture when you wanted your own one changed.
2. The scene in plain sentences. Who or what, doing what, where. «A ceramic mug on a pale oak table by a window» beats «mug, table, window, cozy».
3. Camera and light, in a photographer's words: close-up or wide, eye level or from above, soft window light from the left.
4. The purpose. Google's prompting guide recommends saying what the image is for, because «a logo for a high-end skincare brand» steers style better than «a logo».
5. What stays fixed, for an edit: the people, the layout, the background, the product's shape.
6. Exact text in quotation marks, with the font style and where it sits.

Put together, a pattern you can type from memory:

Create an image of [subject] [doing what] in [setting].
[Shot type and angle]. [Light]. [Style].
It's for [what you'll use it for]. Aspect ratio [16:9].
The text "[exact words]" appears [where], in [font style].

Which prompt for which job

Every prompt on this page names its uploads in the first line. The middle column tells you what to have ready before you paste.

If you only want to restyle a selfie (film looks, new backgrounds, portraits), the Gemini photo prompts are written for exactly that.

Text inside the picture

Lettering is one of the things Google says got better. It lists better text rendering as a headline feature of Nano Banana 2, and says Redo with Pro adds detail especially for images with text and for infographics. Spelling can still slip, so a few habits help:

• Put the words in quotation marks, spelled exactly, and keep them short. A headline and a date are fine; a paragraph is asking for typos.
• Describe the type instead of naming a font: «a heavy condensed sans serif», «thin handwritten script».
• Say where it sits: top center, on the bottom strip, printed on the bag.
• Write the text first. Google's developer documentation says Gemini does better when you settle the wording in one message and ask for the image with that text in the next.

Read every letter at full size before you use the result. The Action Figure Toy Box prints your name on the packaging, which makes it a quick way to see how your model version handles a single line.

A poster, before and after

Here's a request of the kind people type for a quick flyer, and a rewrite of it. We wrote both to show the pattern at work; neither is model output.

make a poster for my bakery opening saturday

The model has to invent the bakery's name, the date, the style and the layout, and it will spell whatever it invents. You'll spend the next five messages undoing guesses.

Create a poster for the opening of a small neighborhood bakery.
A flat lay of three loaves and a few croissants on floured
parchment, shot from directly above, soft daylight, warm cream
and brown tones, simple and uncluttered. It's for a shop window
and an Instagram post. Aspect ratio 4:5.
Text: "Crumb & Co." at the top in a heavy serif, and
"Open Saturday, 8 am" across the bottom in a small clean
sans serif. No other text.

Every word that appears is one you wrote, the layout has two fixed places for type, and «No other text» stops stray lettering on the bread bags. If the type comes out wrong, that's the moment to try Redo with Pro.

Follow-up lines that repair a result

Nano Banana edits conversationally, so the fastest fix is usually one line in the same chat. Google's prompting guide suggests small changes phrased as «keep everything the same, but change…», and that's the shape of every line below.

Where Nano Banana goes wrong

Most failures come from asking one image to do too many jobs. Others are limits Google documents itself.

• Counts. Ask for «four variations» and you may get a different number; the developer docs say the model won't always follow an exact count of images. Ask for one, then «another version».
• Long edit chains. Each round redraws the image, so small drifts add up. After a handful of edits, start fresh from the original with the full instruction.
• Product and logo details: a label's small print, the number of buttons, the curve of a letter. Compare against your source file before a mockup goes to a client.
• Real places. The Isometric Building Miniature keeps the shape and window pattern it can see, but the side of a building that isn't in your photo is invented.
• Colorizing is a guess. A black-and-white print holds no color information, so the coat in Colourise a Vintage Photo gets a believable color, not its real one. Label it as colorized if it's going into a family history.

Plans, uploads and marks

These come from Google's Gemini help pages and hold as of October 2026.

• Uploads. The app takes up to 10 files in one prompt, subject to availability, which covers every prompt here.
• Free and paid. Nano Banana 2 is on every plan. Redo with Pro needs Google AI Plus, Pro or Ultra. Downloads are 2K with a plan and 1K without one, and if you hit your daily Nano Banana 2 quota, Redo with Pro stops working until it resets.
• Age. You need to be 13 or older to generate images and 18 or older to edit them.
• Marks. Every image created or edited in Gemini carries an invisible SynthID watermark and Content Credentials. The visible watermark is a setting (Settings, then Media Watermark), except in India, South Korea and Vietnam, where only AI Ultra subscribers can turn it off. To check a picture, upload it to Gemini and ask whether Google AI created or edited it; there's a quota of roughly 10 image checks a day.

Prices and the rest of the app are in our Gemini review.