Nano Banana Prompts

Practical examples & proven techniques for better results.

Michael Portwell
September 14, 2026

A Nano Banana prompt works when it names three things: the subject, the change, and what must stay the same. Keep it short. If two instructions fight, the model drops one. This page is the prompting guide: five rules, a copy-paste format, then examples for photo editing, backgrounds, and product shots.

Five rules that decide whether a run holds together:

  1. One change per run. Don’t restyle, re-light, and swap the background in one sentence.
  2. Name what stays. Keep the same face, outfit, and pose.
  3. Be concrete. pure white #FFFFFF beats nice studio background.
  4. Put the subject first. Start with who or what is in the frame, then the edit.
  5. If the result drifts, shorten. Cut adjectives, rerun, change one variable.

How to prompt Nano Banana

Run a prompt in three steps. Upload up to nine images in one pass and choose the generation model before you type, because the Base engine wants a short prompt while Pro digests layered ones. Paste one instruction, generate, then change a single variable per rerun instead of rewriting the whole line.

That discipline is the entire trick. A prompt that names one change and guards what stays gives you a repeatable result. A prompt that asks for four changes at once gives you whichever one the model picked at random.

Nano Banana prompt format

The best prompts follow one structure. It helps the model separate the subject from the style from the visual intent.

  • Subject: the main object or character
  • Style: realism, anime, fantasy, product photo
  • Visual details: illumination, colors, angle, mood
  • Extra attributes (optional): background or small details
  • Intent (optional): high detail, clean background, studio look

Quick template:

[subject] + [style] + [visual details] + [extra attributes] + [intent]

Example:

a young woman, anime style, soft glow, warm colors, close‑up, clean background

Nano Banana Prompts Examples

These examples show how simple prompts drive the Base model and how structured ones unlock Pro. The table below is the quick-reference version; the longer sets after it show the range.

Use case Prompt Keep
White background Keep the same subject, edges, and lighting. Replace the background with pure white #FFFFFF. No props, no text. subject, edges
Product shot Keep the same product and label. Place it on a seamless light gray sweep, soft contact shadow only. product, label
Photo retouch Keep the same face and identity. Soften skin shine on the forehead. Do not change hair or makeup. face, hair
Lighting Keep the same scene and pose. Change lighting to soft window light from the left. pose, scene
Face swap Keep the identity from image 1. Place that face on the body in image 2. Do not change the outfit. identity
Architecture Keep the same building volume. Show a dusk exterior, warm interior lights, photoreal. volume

Run any row in the Nano Banana image editor.

Portraits – Base model (simple examples)

  • a young woman, soft warm glow, warm colors, close‑up
  • a young woman, gentle illumination, warm colors, close‑up
  • a smiling girl, bright colors, soft shadows

These stay under six words and pair one subject with one lighting idea, which is all the Base model needs to stay consistent. Reach for them when you want clean, warm close-ups without chasing fine detail.

A close‑up portrait using gentle illumination and warm hues

Portraits – Pro version (advanced examples)

  • a young woman in a red jacket, realism, shallow depth, neon reflections, night city street
  • a man with glasses, detailed skin texture, studio illumination, cinematic mood
  • a woman with curly hair, warm tones, subtle rim glow, indoor café scene

Every line adds a third layer, a location, a material, or a lighting nuance such as “neon reflections” and “cinematic mood”. That extra context is why Pro handles them, it reasons through several relationships at once and returns studio-grade depth.

Anime – Simple instructions for the base nanobanana model

  1. anime girl, bright colors, soft glow, close‑up
  2. anime boy with blue hair, warm tones, simple background
  3. chibi character, pastel colors, cute expression

Short, subject-first phrasing plays to the Base model’s pattern matching, so one subject plus one color or glow cue is enough. Use these for quick style tests and mood checks.

Demonstration of a brightly colored, softly lit close‑up anime girl

Anime – Advanced prompts for the Pro version

  1. anime warrior, dramatic lighting, detailed armor, glowing sword, battlefield at sunset
  2. anime girl in a school uniform, soft glow, cherry blossoms, windy day
  3. anime mage, blue magic effects, night forest, floating particles

These add action and environment, armor, a glowing sword, falling petals, a battlefield. The Pro engine reasons across the whole scene, so the detail stays cohesive instead of turning into a collage of half-finished ideas.

Realism – Basic commands for the fast model

  • a cup of coffee, diffused ambience, wooden table
  • a red apple, clean background, natural illumination
  • a small plant, bright colors, simple shadows

One object, one surface, one light source. Nothing competes for attention, so the Base model reproduces these reliably, which makes them ideal for simple product-style shots.

Prompt for Nano Banana example: generating a coffee cup

Realism – Detailed inputs for the Pro version

  • a realistic cat, soft fur texture, warm ambience, shallow depth, cozy living room
  • a modern living room, natural illumination, wide shot, minimal design, detailed materials
  • a mountain landscape, sunrise glow, high detail, morning mist

Material words such as “fur texture” and “detailed materials” are where Pro earns its keep, it resolves surface quality the Base model flattens. Expect richer output and slightly longer renders.

Fantasy – Simple instructions for the base nanobanana model

  1. a small dragon, bright colors, simple background
  2. a magic sword, glowing aura, dark background
  3. a fairy, diffused illumination, pastel colors

Single subjects with one glow or color cue stay inside the Base model’s comfort zone. These work well for fast character ideas and mood boards.

A glowing magic sword generated against a dark background

Fantasy – Advanced descriptions for the Pro version (complex)

  1. a fantasy castle, dramatic sky, detailed stone texture, flying creatures
  2. a fire mage casting a spell, glowing particles, dark forest, cinematic contrast
  3. a knight in armor, foggy battlefield, cold ambience, broken weapons on the ground

Scene-driven prompts with several layered elements, sky, stone texture, flying creatures, play to Pro’s reasoning. It keeps the composition coherent where the fast model tends to drift.

Products – Simple examples for the base nanobanana model

  • a perfume bottle, studio illumination, clean background
  • a smartwatch, soft shadows, white background
  • a sneaker, bright ambience, simple setup

Clean background plus one studio cue keeps the object sharp and the frame tidy. These suit e-commerce thumbnails and catalog shots that need to look neutral.

A studio‑lit perfume bottle generated on a clean background

Products – Nanobanana Pro Product Examples

  • a luxury watch, macro shot, reflective metal, detailed textures
  • a skincare bottle, studio illumination, water splash effect, glossy surface
  • a laptop on a desk, natural illumination, minimal interior, soft reflections

Reflective surfaces and macro framing are the real test. Pro resolves the glossy detail and keeps reflections logical, where the Base model tends to blur them into glare.

Objects – Nanobanana Object Prompts

  1. a wooden chair, soft gentle glow, clean background
  2. a ceramic mug, warm tones, simple shadows
  3. a wooden chair, subtle ambience, clean background

One object, one ambient cue. The near-duplicate chair prompts are deliberate: comparing them shows how a single changed word shifts the output, which is exactly how you learn the model’s vocabulary.

Example of wooden chair generation with soft illumination

Objects – Nano Banana Pro Object Prompts

  1. a glass vase with flowers, soft reflections, detailed petals, window light
  2. a metal key, macro shot, textured surface, dramatic shadows
  3. a book on a table, warm glow, shallow depth, cozy interior

Texture and reflection words such as “detailed petals” and “textured surface” push Pro to render physical detail. Choose Pro when the material itself is the point of the image.

Scenes – Simple Nano Banana Scene Prompts

  • a beach at sunset, warm colors, simple shapes
  • a forest path, soft illumination, green tones
  • a small street, bright colors, clean lines

Broad scenes, simple shapes, and one light tone let the Base model hit the mood fast. These work well for layout sketches and concept previews.

Demonstration of a warm, simple‑shaped beach scene at sunset

Scenes – Advanced Nanobanana Pro Scene Prompts

  • a cyberpunk city at night, neon lights, wet pavement reflections, wide shot
  • a medieval village, warm light, detailed textures, people walking
  • a modern kitchen, natural light, clean design, wide angle, reflective surfaces

Environments with reflections and wide framing need Pro’s scene understanding. Wet pavement and neon stay coherent and aligned instead of merging into one blur.

Nano Banana prompts for photo editing

Three of the table rows above carry most photo-editing jobs. For a clean e-commerce cutout, paste: Keep the same subject, edges, and lighting. Replace the background with pure white #FFFFFF. No props, no text. For a store-ready product frame, try: Keep the same product and label. Place it on a seamless light gray sweep, soft contact shadow only. For a portrait cleanup, run: Keep the same face and identity. Soften skin shine on the forehead. Do not change hair or makeup.

Each one guards what must stay and changes exactly one thing, which is why they hold up on the Base model and get even cleaner on Pro.

How Google Nanobanana Works

The generator runs two minds. The Base model uses pattern matching, so it is fast and handles short prompts well, but it flattens fine detail and merges elements when given too many instructions. Pro uses a reasoning engine, reads scene relationships, resolves textures and reflections, follows layered prompts, and can pull live web data for fresher references, at the cost of slightly longer renders. Pick the model first, then size the prompt to its depth.

Core Principles of Writing Prompts

Everything above rests on one habit: give the model a single clear idea per pass. Short sentences, a defined subject, one style cue, and simple visual terms keep the output predictable. When two directions fight, the model drops one silently, which is why every fix here trims toward one intent rather than adding words.

Advanced Techniques

These techniques give you more control over Nano Banana and Nanobanana Pro.
They help you shape style, lighting, and composition with precision.

Multi‑Prompt (split ideas clearly)

Multi‑prompt means breaking your idea into short parts.
It helps the nanobanana model understand structure.

Format:

subject | style | lighting | background | mood

Example:

young woman | anime style | warm soft glow| clean background | calm mood

Use it when the model mixes details or adds unwanted elements.

Example of multi‑prompt AI generation

Negative Prompts (remove unwanted details)

Negative prompts tell the model what to avoid.
They clean the image and reduce artifacts.

Use the format:

negative: [things you want to remove]

This format works for both Nanobanana and Nano Banana Pro.

Useful negatives:

blurry details, extra fingers, distorted face, harsh shadows, unwanted objects

Nanobanana reacts softly.
Nano Banana Pro follows negatives very accurately.

Short example of a prompt:

a young woman, soft warm illumination, close‑up
negative: blurry, distorted face, hat
Example of AI negative prompt generation

Style Mixing (combine two styles)

You can blend styles to create unique looks.

Format:

subject, style A + style B, lighting, details

Nano Banana Examples (text instructions):

  1. realistic portrait + anime colors
  2. fantasy armor + modern fashion photo
  3. product photo + cinematic mood

Pro handles mixed styles better.

Lighting Control (shape the mood)

Lighting changes the entire atmosphere.
It is one of the strongest tools in ai image prompts.

Common lighting types:
soft diffusion, hard contrast, rim highlight, backlit effect, neon accents, warm tones, cool tones

Example:

realistic portrait, soft warm light, rim light on hair

Nano Banana prefers simple lighting.
Pro can handle multiple lighting instructions at once.

Composition Control (guide the frame)

Composition defines the camera angle and layout.

Useful terms:
close‑up, medium shot, wide shot, top‑down, centered, rule of thirds, dynamic angle

Example:

fantasy knight, cinematic style, wide shot, rule of thirds

Pro understands composition instructions more reliably.

Example of Composition Control to guide the frame

Face Swap (keep identity across two images)

For identity-sensitive edits, one prompt is not enough. Use: Keep the identity from image 1. Place that face on the body in image 2. Do not change the outfit. Describe the scene, not the portrait, and never list facial features, because the text can override the uploaded reference. Full examples: Nano Banana face swap prompt guide.

Multi‑Image Editing

Multi‑photo workflows are one of the strongest features of nano banana image editing. Nano Banana and Nano Banana Pro can combine elements from several uploaded pictures, letting you merge faces, outfits, light or styles into one final result.

How multi‑photo editing works

The nanobanana model reads each uploaded source separately.
You can tell it which part comes from which file.
Keep the prompt simple and focus on the final result.

Using face from one image and outfit from another

This is the most common multi‑asset workflow.

Upload:

  • Photo 1 for the face you want
  • Photo 2 for the outfit or body you want

Then tell the model which part to use from each source.

Example:

use face from source 1
use outfit from source 2
studio portrait, soft warm light, medium shot

Using one image as a lighting reference

Upload:

  1. File 1 for main subject
  2. File 2 for lighting reference

Example:

use face from photo 1
match lighting from photo 2
cinematic portrait, warm tones

This transfers the lighting mood from file 2 to the subject from file 1.

Using one image as a style reference

Upload:

  • Source 1 => main subject
  • Source 2 => style reference

Example:

use face from photo 1
apply style from photo 2
clean background, soft light

This is useful for matching branding, moodboards, or visual themes.

Combining multiple references

You can mix more than two photos, but keep the nanobanana prompts simple.

Example:

use face from image 1
use outfit from image 2
match lighting from image 3
studio portrait, medium shot

Nanoanana Pro handles multi‑reference prompts more reliably.

Multi‑reference combination demo

Troubleshooting

If nano banana not working or giving unexpected results, the issue usually comes from the prompt, the images, or conflicting instructions.

Sometimes the model also fails because the server receives too many requests at once, and the operation is interrupted. In this case, wait a few seconds and try again.

Common causes & quick fixes

Cause What happens Quick fix
Prompt too long Model gets confused and ignores parts of the request Keep the prompt short and focused
Conflicting instructions Mixed styles or angles reduce consistency Use one style, one lighting, one camera angle
Low-quality reference images Portrait swap or multi-photo editing becomes inaccurate Upload clear, front-facing images
Wrong picture order Model uses the wrong image for the portrait/outfit Specify clearly: “use portrait from photo1”
Describing the portrait Text overrides the uploaded face Do not describe facial features
Negative prompts breaking the result Portrait becomes distorted or inconsistent Avoid heavy negatives during portrait swap
Server overload Operation stops or fails mid-generation Wait a few seconds and retry

FAQ

Nanobanana AI platform is a fast, lightweight tool designed for clean portraits, simple scenes, and quick generations. It works best with short prompts.
Nanobanana Pro is the advanced version of nanobanana. It handles identity, lighting, and multi-image editing more accurately than the base model.
A nanobanana AI prompt is a short, direct instruction describing the scene, style, and lighting. The model performs best when the prompt is minimal.
Base prefers short, subject-first prompts. Pro handles layered instructions with lighting, materials, and composition. Choose the model first, then match the prompt length to it.
Give it structure: subject, style, lighting, then one concrete material or scene detail. Pro resolves relationships between them, so clear, specific detail turns into richer output.
Yes. Use the format “negative: [things to remove]”. The Base model reacts softly to negatives, while Pro follows them closely. Keep them short so they do not override the main prompt.
Yes. Describe the scene, not the portrait, and add “use face from uploaded photo”. Keep the style simple and upload one clear, front-facing reference. Full examples: Nano Banana face swap prompt guide.
Because the model prioritizes the text. If you describe facial features, it may override the uploaded face. Keep the portrait description out of the prompt.
Usually the prompt carries too many or conflicting instructions: it is too long, styles conflict, or negatives break the portrait. Trim to one subject, one lighting cue, and one camera angle, then regenerate. Shorter prompts fix most issues.
Usually because the model doesn’t know which file is which. Always specify clearly: use portrait from image 1, use outfit from image 2.
Sometimes the server receives too many requests at once. If this happens, wait a few seconds and try again.
This guide is designed to be your starting point. We will expand the banana guide and FAQ as new questions appear in the comments.
Nanobanana blocks any prompts involving: Violence or graphic harm, Mature content, Minors in unsafe or inappropriate contexts, Political propaganda or manipulated political imagery, Dangerous or harmful activities, Realistic document forgeries, Harassment, hate, or defamatory content, Copyrighted characters recreated in a realistic or deceptive way. If a prompt includes restricted content, the model will refuse or return a neutral result.