Prompting guide

How to prompt Nano Banana 2

Nano Banana 2 prompt guide: how to prompt Google's image model for exact text, one-sentence edits, and consistent characters, with copyable examples.

Try Nano Banana 2.1 Nano Banana 2 model page

Updated October 26, 2026

Made with Nano Banana 2A white dragon perched on a city tower at dusk, in a telephoto news-photo style

Quick start

Your first great Nano Banana 2 prompt

Everything a beginner needs in one minute: what to know first, the rules that matter most, a template to fill in, and prompts to copy.

What to know first

Nano Banana reads a prompt the way a designer reads a brief: full sentences about what the image is for, what is in it and where, and how it should look. Upload an image and the prompt becomes an edit, so say what changes and what stays. Each generation starts fresh, with no memory of the one before: to refine a result, upload it and say what to keep. Aspect ratio and resolution are settings.

The rules that matter most

Write sentences, and say what it is for
Describe the scene in full sentences, not keywords, and name the purpose: "a logo for a high-end, minimalist skincare brand" gives Nano Banana far more than "a logo".
Lead with the job
Open with a verb that names the operation: Create, Change, Remove, Combine, Transform. The first word tells Nano Banana whether to draw something new or rework what you uploaded.
Quote every word you want to see
Write the final text yourself and put it in double quotes, with its lettering and its place: the headline "SUMMER RUN" in heavy condensed capitals across the top.
Edit one thing, and say what stays
"Change only the blue sofa to a brown leather chesterfield. Keep everything else exactly the same." Set the aspect ratio to match the image you upload, or the result comes back square.
Refine by uploading the last result
Nano Banana does not remember your previous image. Upload the one you like, ask for one change, and say again what stays the same.

Fill in this template

Replace each part in square brackets with your own idea, and leave out a part you do not need.

Template

Create [what the image is for]: [subject] [doing what] in [setting]. [Composition]. [Style]. [Light].

The template, filled in

Create a vertical social post for a neighborhood bakery: a baker in a flour-dusted apron pulls a tray of croissants from a brick oven in a small bakery kitchen at dawn. Medium shot at eye level with a 50mm lens and a shallow depth of field, the oven glowing behind her. Photorealistic. Warm morning light from a window on the left catches the flour in the air.

[what the image is for] The format and purpose: a vertical social post, a product photo for a shop, a book cover, a widescreen backdrop.
[subject] Who or what it is about, with a few visible facts: a baker in a flour-dusted apron.
[doing what] The action or the pose: pulls a tray of croissants from a brick oven.
[setting] Where and when: a small bakery kitchen at dawn.
[Composition] The framing and the camera: a medium shot at eye level, a 50mm lens, a shallow depth of field.
[Style] The medium and the look: photorealistic, a watercolor illustration, a kawaii sticker with bold outlines.
[Light] One light with its direction: warm morning light from a window on the left.

Prompts to start from

A photoreal scene, described like a shot

A photorealistic wide-angle shot of a vibrant coral reef teeming with tropical fish. Crystal-clear turquoise water with sunbeams filtering down from the surface, lighting a sea turtle gliding over the coral. Shot from a low perspective with a wide-angle lens. Widescreen image.

Why it works: It follows the photoreal formula: the kind of shot, the subject, the setting, the light, then the angle and the lens. Set the aspect ratio to 16:9 to match the last line.

A product with its label

A high-resolution, studio-lit product photograph of a matte black coffee bag standing on a polished concrete surface. The front label reads "SINGLE ORIGIN" in a clean white sans-serif, with "Huila, Colombia" in small letters below it. No other text. Three-point softbox lighting with soft, diffused highlights. Slightly elevated 45-degree angle. Square image.

Why it works: Each line of text is quoted with its own size and style, and "No other text" stops stray words from appearing on the bag.

An edit in one sentence

Using the provided image of a living room, change only the blue sofa to a vintage brown leather chesterfield sofa. Keep the rest of the room, including the pillows on the sofa and the lighting, unchanged.

Why it works: Upload the photo and set the aspect ratio to match it. Naming the one change and listing what stays is how Nano Banana edits without a mask.

Your Character in a new scene

Create a photo for a travel blog: Maya (@maya), a woman with a short silver bob and a mustard raincoat, laughs as she steps off a tram on a rainy Lisbon street at dusk. Keep her face and hair exactly as in her image. Medium shot at eye level, 35mm lens, warm shop lights reflecting on the wet cobblestones.

Why it works: The handle brings in her face, and the name with two visible traits lets Nano Banana track her across a series. Replace @maya with your own Character.

Two uploads, two roles

Using the napkin sketch in @image1 as the structure and the fabric in @image2 as the texture, transform this into a high-fidelity 3D render of an armchair. Place it in a sun-drenched, minimalist living room.

Why it works: Upload the sketch first and the fabric swatch second. Each image gets one job, then the prompt says what to make from them.

Advanced

Advanced Nano Banana 2 prompting

Techniques for more control once the basics work.

Nano Banana 2.1, Nano Banana 2, and the settings that matter

Nano Banana 2.1 is the default: it improves on Nano Banana 2 in image quality, text, prompt following, and character consistency. Keep Nano Banana 2 for small drafts or to match work made on it. The same prompt works on both.

Thinking Level decides how much Nano Banana plans before it draws. Minimal is the default and suits most single images. High is slower: the model works out the composition first, which pays off for multi-step scenes, infographics, dense text, and many references.

Web Search grounds the image in current facts: a recent event, today's weather, a product that just launched. Image Search looks up what a real place or object looks like; it does not return photos of people.
With a search on, say in the prompt what to look up and what to do with it: "First search for reference images of the Clos Lucé manor, then draw it as a color chalk sketch. No text."
Name the format in words as well as in the Aspect Ratio setting: "vertical social post", "widescreen backdrop", "square sticker".

Structure: one formula per job

Nano Banana is most precise when the prompt is specific and describes one scene in sentences. Each kind of job has a shape that works, and the most useful is the order for a new image: subject, action, place, composition, then style.

For a complex scene, give the steps in order: "First, create a background of… Then, in the foreground, add… Finally, place…". Building back to front keeps each layer where you put it.

New image: "A striking fashion model in a tailored brown dress poses with a confident stance against a seamless cherry-red backdrop. Medium-full shot, center-framed. Fashion editorial shot on medium-format analog film, pronounced grain."
Photoreal: "A photorealistic [kind of shot] of [subject] in [setting]. [Light]. Shot from [angle] with a [lens]."
With references: what each image is, how they relate, then the new scene.
With search: what to look up, what to work out from it, and how to show it.
Edit: "Using the provided image, change only [element] to [new element]. Keep everything else exactly the same, preserving the original style, lighting, and composition."
For options, run separate generations. Ask for "three variations side by side" only when one comparison sheet is what you want.

A busy scene, built in steps

Create a widescreen poster illustration for a summer night market. First, create a background of a narrow street strung with paper lanterns under a deep blue evening sky. Then, in the middle ground, add rows of food stalls with steam rising from woks and grills. Finally, place a young couple sharing a paper boat of dumplings in the foreground, lit warmly by the nearest stall. Flat illustration with risograph texture in coral, teal, and gold.

Why it works: Each layer is placed in its own step, and the medium and palette come last, once the layout is settled. Set the aspect ratio to 16:9.

References and Characters

Mention saved Characters and References (outfits, locations, poses, objects) by their @handle, and images you upload as @image1, @image2. Nano Banana sees them as one numbered list: your uploads first, then Characters, then References. Before it reads the prompt, Mage swaps each mention for its number, so with one upload, @maya becomes "character from image 2" and a saved location @loft "location from image 3".

So give every image a role in words, and say how they relate: "Create a new image by combining the elements from the provided images. Take the blue armchair from image 1 and place it in the sunlit living room from image 2, matching its light and perspective."

Protect what matters by describing it and saying it stays "completely unchanged". A face you describe is a face Nano Banana knows to keep.
Up to about ten object images keep their detail in one generation. For more, build the scene in steps.
Name the target of an edit by what it visibly is: "the woman on the left", "the tallest glass", never "her" or "it".
To turn stills into video, make the first and last frames here, then animate them on Lemon, which takes both.

A logo on a shirt, face untouched

Take the woman in @image1, with brown hair, blue eyes, and a neutral expression. Add the logo from @image2 onto her black t-shirt. Ensure the woman's face and features remain completely unchanged. The logo should look naturally printed on the fabric, following the folds of the shirt.

Why it works: Upload the photo first and the logo second. Describing the face is what protects it, and the last sentence asks for the physics that makes the print look real.

Keep a character consistent

Nano Banana tracks people by name. Describe each character once with two or three visible traits ("Maya, a woman with a short silver bob and a mustard raincoat"), then use the same name and traits in every prompt of a series. With a saved Character, put the traits right after the handle: Maya (@maya), a woman with a short silver bob and a mustard raincoat.

Faces stay recognizable for up to about four people in one image. For a bigger cast, build the group in steps, a few people per pass, uploading the last result each time.

To build a character from scratch, make one angle per generation. Start from the best front view. For each new angle, upload the front view and the angles made so far, and ask for a single view. Save the cleanest front view as a Character for every scene after that.

For a pose that is hard to describe, add a pose image with one line on its role: "Use only the pose from image 3."
Ask for new angles on a plain background, as in the example below, so nothing competes with the face.
For a story, generate the scenes one at a time with the same Characters and names, rather than asking for every scene in one image.

A turnaround, one angle at a time

A studio portrait of the man in @image1 against a plain white background, in profile looking right. Same face, hair, and clothing as in @image1.

Why it works: Upload the front view as @image1 and any earlier angles after it. One view per generation keeps the face from drifting.

The next page of a picture book

Create the third illustration for a children's picture book: Maya (@maya), a girl with a short silver bob and a mustard raincoat, and her dog Pip, a scruffy white terrier with one brown ear, cross a rope bridge over a misty river. Maya holds a lantern; Pip trots ahead of her. Soft watercolor illustration with ink outlines, in the same mustard, teal, and gray palette as the rest of the book. Wide shot from the riverbank.

Why it works: The Character, the dog, and the palette are described in the same words as on every other page, which is what keeps a series on model. Swap in your own Character.

Text that reads right

Write the copy first, then render it. Settle every word before you generate, put it in the prompt in double quotes, and never let Nano Banana invent the copy for a final image.

Give each line its own text and style: the headline "SUMMER RUN" in heavy, blocky condensed capitals; below it, "Saturday 6 June, 8 am" in a light sans-serif. Font names work as cues for the look: Century Gothic, Impact.

Stop stray words with "No other text" or "no placeholder text, all text in English".
For dense text, menus, and infographics, set Thinking Level to High and choose a higher resolution.
To localize, write the prompt in English, name the target language and region, and quote the exact translated copy. Nano Banana is strongest in English, Arabic, German, Spanish, French, Hindi, Indonesian, Italian, Japanese, Korean, Portuguese, Russian, Ukrainian, Vietnamese, and Chinese.
Leave the @ out of words on the image: Mage reads @ followed by letters as a Character or Reference handle. Write a social handle as a web address instead.
Fix a misspelling with an edit: upload the result, quote the corrected line, and ask for the same font, size, and color.

A magazine cover with exact copy

A photo of a glossy magazine cover. The minimal blue cover has the large bold word "NORTHBOUND" in a serif font that fills the width. No other text except the issue number "No. 12" and the date "Feb 2026" in the top corner, and a small barcode. In front of the title, a woman in a sleek, minimal dress playfully holds up a folded paper map. The magazine sits on a shelf against an orange plastered wall in a designer store.

Why it works: Every word on the cover is quoted and accounted for, and "No other text except…" closes the list. Use a 3:4 aspect ratio and Thinking Level High.

The same poster, in Japanese

Using the provided poster, translate all of the text into Japanese for a Tokyo audience: the headline becomes "夏のランニング" and the date line becomes "6月6日(土) 午前8時". Keep the layout, the colors, the weight of the lettering, and the runner illustration exactly the same.

Why it works: Upload the finished poster. The prompt stays in English, names the language and the audience, and gives the exact translated words, so nothing is left to guesswork.

Edits and the edit loop

Nano Banana takes no mask: name the region in words and list what stays. "Change only the blue sofa" protects more than "change the sofa", and naming the pillows, the rug, and the light keeps them from shifting.

Refine in a loop. Each generation starts fresh, so upload the result you like, ask for one change, and restate what stays: "Keep everything the same, but make her expression more serious." One change per round is easier to judge, and easier to undo, than five at once.

Set the aspect ratio to the uploaded image's own shape for every edit. It starts at square.
Re-stage a photo by changing only the conditions: "Make it a winter evening with fresh snow on the street. Keep the buildings, the cars, and the camera angle exactly as they are."
When a round goes wrong, go back to the last good image rather than stacking fixes on a bad one.
Turn a sketch into a finished piece: say what must survive from the sketch and what to add.

Round two of an edit loop

Keep everything the same, but make her expression more serious. Her face, her pose, the green coat, the street, and the evening light stay exactly as they are.

Why it works: Upload the last result. One change, then the list of what stays, restated every round because Nano Banana has no memory of the round before.

From sketch to product photo

Turn this rough pencil sketch of a sports car into a polished studio photo. Keep the body shape and wheel design, add chrome trim and a candy-red finish. Seamless gray backdrop, soft overhead light, a three-quarter front view.

Why it works: Upload the sketch. Saying what to keep makes the drawing the structure rather than a loose suggestion.

Camera, light, and style words

Nano Banana knows the language of photographers and designers. Name the shot, the lens, the light, and the materials instead of stacking adjectives: "navy blue tweed", "ornate elven plate armor etched with silver leaf patterns", "a minimalist ceramic mug in matte black".

Color grading reads too: "cinematic color grading with muted teal tones", "as if on 1980s color film, slightly grainy".
Golden-hour backlighting with long shadows, or soft diffused light from a named direction, sets a mood better than a dozen mood words.

Write	What it does
`wide-angle shot`	A broad view that shows scale and setting.
`macro shot`	A small subject in extreme detail.
`low-angle perspective`	The camera looks up, so the subject looms.
`45-degree elevated shot`	The classic product angle, slightly from above.
`top-down flat lay`	Objects arranged on a surface, seen from straight above.
`perfectly isometric`	A clean 3D view with no perspective, for diagrams and miniature worlds.
`shallow depth of field (f/1.8)`	A sharp subject against a soft, blurred background.
`cheap disposable camera`	Raw, nostalgic snapshots with flash.
`medium-format analog film`	Rich tones and pronounced grain, like a fashion editorial.
`chiaroscuro`	Harsh, high-contrast light against deep shadow.
`three-point softbox setup`	Even studio light with soft highlights and no harsh shadows.
`kawaii sticker`	Cute characters with bold, clean outlines.
`risograph`	Grainy, slightly misaligned print in a few bright inks.

Common mistakes

Avoid this, do this instead

Avoid: Comma-separated keywords.

logo, skincare, minimalist, elegant, gold

Instead: Sentences that describe the scene and say what it is for.

A logo for a high-end, minimalist skincare brand called "LUMA": a thin gold line drawing of a single leaf above the name, set in a light, widely spaced serif on cream.

Avoid: A list of things to leave out: "no cars, no people".
```
A street at dawn, no cars, no people.
```
Instead: The scene you want, described positively. A short "No other text" is fine.
```
An empty, deserted street at dawn, with no signs of traffic.
```
Avoid: An edit at the default square aspect ratio.
Instead: The aspect ratio of the image you upload.

Avoid: "Now make it warmer" with no image uploaded.

Now make it warmer.

Instead: The last result uploaded, then the one change and what stays.

Keep everything the same, but make the light warmer, like late afternoon sun. The people, the room, and the framing stay exactly as they are.

Avoid: Six people with exact faces in one image.
Instead: Up to about four, or a bigger group built in steps.
Avoid: Asking Nano Banana to write the copy.
Instead: The final words, quoted, line by line.
Avoid: A social handle with its @ on the image.
Instead: The name without the @, or a web address.

FAQ

Nano Banana 2 prompting questions

How long should a Nano Banana prompt be?

Long enough to place everything that matters. A single subject needs two or three sentences; a poster, an infographic, or a scene with several people reads best as a step-by-step paragraph. Specific beats long: every sentence should add something you can see.

Can Nano Banana 2 put text in images?

Yes, and it is one of its strengths. Write the exact words in double quotes with their lettering and place, add "No other text", and set Thinking Level to High for dense text. Nano Banana 2.1 renders text better than Nano Banana 2.

How do I keep a character consistent in Nano Banana?

Save the person as a Character and mention the same @handle in every prompt, followed by the same name and two or three visible traits. Keep to about four people per image, and make new angles one view at a time from the front view.

Does Nano Banana support negative prompts?

No. Describe the scene you want instead: "an empty, deserted street" rather than "no cars". Short lines like "No other text" do work.

Why did my Nano Banana edit come back square?

The Aspect Ratio setting applies to edits too, and it starts at square. Set it to the shape of the image you uploaded.

Does Nano Banana remember my last image?

No. Each generation starts fresh. To refine a result, upload it and say what changes and what stays.

What does Thinking Level do?

On High, Nano Banana plans the composition before it draws. It is slower, and it helps with multi-step scenes, infographics, dense text, and many references. Minimal suits most single images.

Should I use Nano Banana 2.1 or Nano Banana 2?

Nano Banana 2.1, the default, for nearly everything. Nano Banana 2 for small drafts or to match work made on it. The same prompt works on both.

Prompting guide

More prompting guides

How to prompt Anima
How to prompt Anima on Mage: Danbooru tag order, rating tags, strong weights, escaped parentheses, and Anima prompt examples for characters and scenes.
How to prompt Chroma
Chroma prompt guide: how to prompt Chroma with one caption-style paragraph, use its negative prompt, tune steps and guidance, and copy Chroma prompt examples.
How to prompt Flux 2
Flux 2 prompt guide: how to prompt Flux 2 Dev with subject-first word order, JSON prompts, exact hex colors, and multi-image edits, with examples.
How to prompt GPT Image 2
GPT Image 2 prompt guide: how to prompt GPT Image 2.5 Flare and Sunburst for exact text, precise edits, and consistent characters, with examples.
How to prompt Grok Image
Grok Imagine prompt guide: how to prompt Grok Image with xAI's subject-first order, a check-then-edit loop, and copyable photo prompt examples.
How to prompt Guava
How to prompt Guava, Mage's photorealism model: a quick start, a fill-in template, and expert tips for portraits, skin, light, lenses, edits, and text.

Make your first image with Nano Banana 2.1

New accounts get 300 free Gems, enough to try Nano Banana 2.1. No install, no credit card.

Try Nano Banana 2.1