Nano Banana models read prompts as plain language, not as a list of tags. The prompts that work best describe a scene the way you would brief a photographer or illustrator: what is in the frame, where it is, how it is lit, and what must not change. This guide gives you a repeatable structure, prompt patterns for common tasks, rules for readable text, how to iterate when a result is close but not right, and prompts you can copy into the image studio.
Model options, resolutions and credit prices in this guide are the ones the NanoBananas studio uses on November 9, 2026.
A five-part prompt structure
Every strong Nano Banana prompt answers the same five questions. You do not need all five every time, but skipping one leaves the model to guess, and guesses are where results drift.
| Part | What to describe | Example |
|---|---|---|
| Subject | Who or what, with the details that matter | a ceramic coffee mug with a matte sage glaze |
| Setting | Where the subject is | on a worn oak café table |
| Composition | Framing, angle and space | close-up, eye level, empty space on the left for text |
| Light | Source, direction and time of day | soft morning window light from the right |
| Style | Medium and finish | natural lifestyle product photography |
Put together, that becomes: A ceramic coffee mug with a matte sage glaze on a worn oak café table, close-up at eye level with empty space on the left, soft morning window light from the right, natural lifestyle product photography. It reads like a sentence, and every phrase gives the model a decision it no longer has to invent.
Write descriptions, not keyword lists
Older image models rewarded comma-separated keywords and long negative prompts. Nano Banana models are built on Gemini and follow natural language, so a list such as mug, cafe, 8k, ultra detailed, masterpiece gives them less to work with than one clear sentence.
| Instead of | Write | Why |
|---|---|---|
| a nice car | a red 1960s convertible with chrome bumpers | Specific details remove guesswork |
| a beach, no people | an empty beach at sunrise | Describe what you want, not what you do not want |
| cat, books, window | a cat sitting on a stack of books beside a window | Relationships place objects relative to each other |
| 8k, masterpiece, ultra detailed | sharp focus, fine texture on the fabric | Quality buzzwords add little; resolution is a setting |
| a city and a forest and a robot | one scene per image | Unrelated ideas end in a compromise of both |
Prompt patterns for common tasks
The five parts stay the same, but which ones matter most changes with the job.
| Task | Lead with | Pattern |
|---|---|---|
| Photo-realistic scene | Light and camera | Subject + setting + framing + named light source + "realistic photography" |
| Illustration | Style and shapes | Subject + setting + medium (watercolor, flat vector, ink) + palette |
| Poster or social graphic | Layout and text | Background + main visual + exact text in quotes + font and placement |
| Infographic or diagram | Structure | Number of sections + label text in quotes + clean flat style |
| Product shot | The reference image | "Keep the exact product" + surface + light + copy space |
| Character sheet | Consistency | One character + several poses or angles + plain background |
For product and portrait work, see the dedicated guides on product photography prompts and portrait and headshot prompts.
Getting readable text in the image
Text rendering is where model choice and prompt wording matter most.
- Quote the exact words: headline reading "SUMMER FESTIVAL". Unquoted text gets paraphrased.
- Say where it goes: across the top third, on the label, bottom right corner.
- Describe the type: bold condensed sans-serif, hand-painted serif, white on dark blue.
- Keep lines short: three to six words per line is the reliable range. Split long copy into a headline and a short subline.
- Use enough resolution: small type needs 2K or 4K.
Text-heavy poster
A minimalist event poster on a warm cream background with a large paper-cut sun over layered teal hills. Headline in bold condensed sans-serif reading "NIGHT MARKET" across the top, smaller line reading "FRIDAY 7 PM" at the bottom. Flat colors, generous margins.
For posters, labels and infographics, choose Nano Banana 2.1. Google designed it to plan layout before rendering, and it renders natively at up to 4K. Nano Banana 2 is fine for short headlines on drafts. If the image needs a lot of copy, generate the visual with space for text and set the text in a design tool.
Prompts for editing an existing photo
In Image to Image mode the model already has a picture, so the prompt is an instruction, not a description. Lead with what must stay the same, then describe the change in one sentence.
Edit template
Keep the person's face, hairstyle and pose exactly as in the photo. Replace the background with a quiet city street at dusk, with warm shop lights softly out of focus. Match the lighting on the person to the new scene.
If an edit changes more than you asked, add the protected element to the keep sentence and run it again. Our photo editing guide covers backgrounds, object removal, relighting and retouching in detail.
Using reference images
The studio accepts up to four reference images per run (JPG, PNG or WebP, up to 10 MB each). Refer to them by role in the prompt: the woman from the first image wearing the jacket from the second image. Naming the role tells the model which features to copy from where.
- Crop each reference so the thing you want copied fills the frame.
- Use one reference per role: one for the face, one for the outfit, one for the style.
- Say which elements to ignore if a reference contains extras: use only the jacket from the second image.
For keeping a character the same across a whole series, see the character consistency guide.
Aspect ratio and resolution are settings
Do not try to control size through the prompt. Aspect ratio and resolution are chosen in the studio settings, and the model composes for the ratio you pick.
| Setting | Options | Tip |
|---|---|---|
| Aspect ratio | 1:1, 4:5, 5:4, 3:4, 4:3, 2:3, 3:2, 9:16, 16:9, 21:9, or Auto | Pick the final ratio before you generate rather than cropping later |
| Resolution | 1K, 2K, 4K (Nano Banana v1 Classic has one standard size) | Iterate at 1K, render the final at 2K or 4K |
| Images per run | 1 to 4 | Several versions help when exploring; each one is charged |
How to iterate when a result is close
The fastest way to improve a result is to change one part of the prompt at a time, so you can see what each change does. Use the five-part structure to find which part to fix.
| What is wrong | Part to change | Example fix |
|---|---|---|
| Wrong object or missing detail | Subject | Add the specific material, color or count |
| Wrong place or clutter | Setting | Name one location and remove extra props |
| Cropped, crowded or off-center | Composition | State framing and where empty space goes |
| Flat, harsh or mismatched look | Light | Name one light source and its direction |
| Looks like the wrong medium | Style | Name the medium and finish explicitly |
When you find wording you like, save it. Reusing a tested style sentence across many images is how you get a consistent look. Browse 400 example prompts for starting points.
Matching the prompt to the model
| Model | Prompt strengths | Credits (1K / 2K / 4K) |
|---|---|---|
| Nano Banana 2 | Fast drafts, simple scenes, many variations | 3 / 4 / 8 |
| Nano Banana 2.1 | Text, layouts, multi-character scenes | 6 / 8 / 16 |
| Nano Banana Pro | Cinematic portraits, texture and mood | 8 / 12 / 20 |
| Nano Banana v1 Classic | The original look at a fixed cost | 2 (one standard size) |
A practical workflow is to iterate on wording with Nano Banana 2 at 1K, then run the final prompt on 2.1 or Pro at the resolution you need. Ten drafts at 1K on Nano Banana 2 cost 30 credits, less than two 4K renders on Pro.
Prompts to copy
Editorial portrait
A natural editorial portrait of a woman in her thirties with short curly hair and a cobalt wool coat, standing on a quiet tree-lined street in autumn. Waist-up framing at eye level, soft overcast light, shallow depth of field, realistic photography.
Event poster
A minimalist event poster with a large paper-cut sun over layered teal hills. Headline in bold condensed sans-serif reading "NIGHT MARKET" across the top, smaller line reading "FRIDAY 7 PM" at the bottom. Flat colors, generous margins.
Storybook illustration
A small fox resting beside a glowing lantern in a woodland clearing at dusk, fireflies in the air, deep blue trees and a warm amber glow. Gentle storybook illustration with clear shapes and soft texture.
Simple infographic
A clean flat infographic on a white background showing three steps in a row, each with a simple icon in a teal circle: a seed labeled "PLANT", a watering can labeled "WATER", a flower labeled "GROW". Thin gray arrows between steps, bold sans-serif labels.
Character sheet
A character sheet of a cheerful young inventor with round glasses, a brown leather apron and a red scarf, shown in three poses: front view, side view and waving. Plain light gray background, consistent proportions and colors, clean cartoon illustration.
Frequently asked questions
How long should a Nano Banana prompt be?
One to three sentences is usually enough. Length matters less than covering subject, setting, composition, light and style. Very long prompts with conflicting details tend to blur the result.
Do negative prompts work with Nano Banana?
There is no separate negative prompt field. Describe what you want instead; for example "an empty street" rather than "no people". For edits, list what to keep.
Can I write prompts in languages other than English?
Yes, the models understand many languages. For text that must appear in the image, write it exactly as you want it rendered, inside quotation marks.
Why does the same prompt give different results each time?
Each generation samples a new image. Keep the prompt fixed and run a few versions, or add a reference image when you need a consistent subject.
Should I write the aspect ratio or resolution in the prompt?
No. Choose them in the studio settings. The model composes for the ratio you select.
How do I get the same style across many images?
Write one style sentence (light, palette, medium) and reuse it unchanged, changing only the subject. A reference image of the style helps too.
Which model follows prompts best?
Nano Banana 2.1 is the strongest at following detailed layouts and text. Nano Banana Pro is best for mood and texture. Nano Banana 2 is the cheapest place to test wording.