GenMagic logoGenMagic
Sign in

Nano Banana Prompt Guide: Google's Own Prompt Structure, Exact Text, Edits and Reference Images, With Examples You Can Run (2026)

Live figures last changed · 6 min read · The GenMagic team

Google's own advice for Nano Banana is to describe the scene as a sentence, not a list of keywords, in this order: subject, action, location, composition, style, and to start with a strong verb that says what you want done. Put any words that must appear in quotes and name the font; for an edit, say exactly what must stay the same. In our tests, Nano Banana 2.1 set every word of two text-heavy ads right and kept a product's label as photographed.

This guide covers Google's prompt structure, text in the image, editing a photo, reference images, style words, shapes and sizes, what to avoid, and examples you can run.

A note on who wrote this: GenMagic runs every Nano Banana model. The structure and quoted advice come from Google's ultimate Nano Banana prompting guide (Google Cloud blog, March 6, 2026, written for Nano Banana 2 and Pro; Nano Banana 2.1 is their successor), and the results from our own tests, linked where used. Which Nano Banana model to call and what each costs: Nano Banana API.

Table of contents

  1. Google's prompt structure
  2. Text in the image
  3. Editing a photo
  4. Reference images
  5. Style words
  6. Shapes and sizes
  7. What to avoid
  8. Examples you can run
  9. Questions

Google's prompt structure

For a picture from text, Google's guide gives the order [Subject] + [Action] + [Location/context] + [Composition] + [Style], and three habits:

  • Lead with a strong verb that names the job: create, paint, draw, edit, remove, place.
  • Write narratively: "A simple list of keywords won't cut it; you need to describe the scene narratively."
  • Be specific about subject, lighting and composition, and iterate with follow-up requests in the same conversation.

A template to copy: " [Verb] [subject with its details] [doing what], [where], [shot and framing], [style, light and color]."

Text in the image

From Google's guide, and confirmed in our tests:

  • Put the exact words in quotes: "Happy Birthday", "URBAN EXPLORER".
  • Name the font or its style: "bold, white, sans-serif", "flowing, elegant brush script".
  • Say where each line goes: headline at the top, a badge bottom right, a footer.
  • Other languages: write the prompt in one language and name the language the text should be in.
  • Text first: Google says the models "work best if you first converse with it to generate the text concepts, and then ask for an image with that text".

In our ad test, Nano Banana 2.1 set a headline, three benefit lines and a footer word for word; in our image generator test, it spelled a three-line shop sign right and added a small crown nobody asked for. Every model that set every word right: images with exact text.

Editing a photo

  • Name the change with a verb: "Remove the man from the photo", "Replace the sky with a sunset", "Place the bottle on a marble shelf".
  • Be explicit about what to keep exactly the same, in Google's words: the label, the face, the furniture, the framing.
  • Edit step by step: one change per request, each on the last result.

In our virtual staging test, Nano Banana 2.1 changed the empty parts of a room least of every model while adding furniture, and in our headshot test it kept both people recognizable.

Reference images

  • Up to 14 reference images in one prompt, by Google's guide.
  • Say what each one is for, the way Google's example does: "Using the attached napkin sketch as the structure and the attached fabric sample as the texture, transform this into a high-fidelity 3D armchair render."
  • For a character across a series, attach the same character sheet every time and repeat the character's fixed traits in words: character sheets.

Style words

Google's guide asks for concrete creative direction:

  • Light: "three-point softbox setup", "golden hour backlighting creating long shadows", "chiaroscuro lighting with harsh, high contrast".
  • Camera and lens: "low-angle shot with a shallow depth of field (f/1.8)", "wide-angle lens", "macro lens".
  • Color and film: "as if on 1980s color film, slightly grainy", "cinematic color grading with muted teal tones".
  • Materials: "navy blue tweed" rather than "suit jacket"; the more exact the texture, the better.

Shapes and sizes

  • Aspect ratios on both Nano Banana 2 and Pro: 1:1, 3:2, 2:3, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9 and 21:9; Nano Banana 2 adds 1:4, 4:1, 1:8 and 8:1 for banners, by Google's guide.
  • Resolution: 1K, 2K or 4K.
  • On GenMagic a Nano Banana 2.1 image is listed at $0.058 today; larger images and long prompts cost a little more, because the model bills by the tokens an image uses.

What to avoid

  • Negative phrasing. Google's guide: "Describe what you want, not what you don't want (e.g. 'empty street' instead of 'no cars')."
  • Keyword lists instead of a described scene.
  • Asking for something that is not there to keep: in our headshot test, another model added glasses after a prompt asked it to keep them. Name only what the person actually has.
  • Real people and brands you have no rights to: not yours to use, and Google's safety filters turn some such requests down. A refused image costs nothing on GenMagic.

Examples you can run

Each link opens the GenMagic studio on Nano Banana 2.1 with the prompt ready; you see the price before you generate.

A product ad with exact text: "Create a square social media ad: a glass bottle of cold-pressed orange juice on a sunlit marble kitchen counter, orange slices beside it, soft morning light from the left, shallow depth of field. At the top, the headline "Fresh by 8 a.m." in a bold, white, rounded sans-serif font; at the bottom right, a small round badge reading "NEW". Clean, bright commercial photography." Open in the studio

A stylized portrait: "Paint a portrait of an elderly fisherman with a grey beard and a navy blue wool cap, looking out to sea from the deck of a small wooden boat at golden hour, medium close-up from a low angle, in the style of a loose oil painting with visible brushstrokes, warm colors and long shadows." Open in the studio

A character sheet for a series: "Draw a character sheet of a small red fox in a green knitted scarf and a brown leather satchel: front view, side view and back view side by side on a plain white background, the same proportions and colors in every view, clean flat illustration style for a children's book." Open in the studio

An edit that keeps the product: upload a photo of your product and write "Place this exact bottle, with its label unchanged, on a marble bathroom shelf beside a eucalyptus sprig, in soft morning light."

More prompts by what you make: product photos, posters, YouTube thumbnails, children's books. Nano Banana in a workflow: Nano Banana in n8n.

Prompt guides for other image models: GPT Image, FLUX 3. For video: Seedance, Veo 3.1, Kling 3.0.

Teaching AI video or images on YouTube, in a course or in n8n templates? The creator program pays 15% in cash of what the people you bring pay in their first month.

Questions

Is there an official Nano Banana prompt guide? Yes: Google's "ultimate Nano Banana prompting guide" on the Google Cloud blog, published March 6, 2026, for Nano Banana 2 and Nano Banana Pro. Its structure is subject, action, location, composition and style, written as a described scene.

How do I get Nano Banana to spell text correctly? Put the exact words in quotes, name the font, say where each line goes, and keep the text short. In our ad test, Nano Banana 2.1 set every word right.

How do I keep something the same when editing? Say so explicitly ("keep the bottle and its label exactly as they are") and make one change per request.

How many reference images can Nano Banana use? Up to 14 in one prompt, by Google's guide; say what each image is for.

Should I tell Nano Banana what not to include? Google recommends describing what you want instead: "empty street" rather than "no cars".