Nano Banana AI Guide

Nano Banana Prompt Guide

A practical guide to writing Nano Banana prompts: constraints, structure, and iteration rules to reduce wasted attempts and improve consistency.

Powered by Gemini 2.5 Flash
Try 1 free generation in the generator — sign-up optional
Quick Answer

A good Nano Banana prompt is specific, constraint-driven, and iterative. The goal is to reduce wasted attempts by changing one variable at a time. Most beginners fail because they write vague prompts or try to change everything at once.

Understanding prompt skill levels

Prompt writing is a learnable skill. Here's how to progress from beginner to advanced:

1

Beginner

Focus on the 3 core elements: subject, action, and one style keyword.

Beginner prompt result: golden retriever in park with warm sunlight

A golden retriever, sitting in a park, warm sunlight

2

Intermediate

Add constraints and negative instructions to control unwanted outputs.

Intermediate prompt result: golden retriever with shallow depth of field

A golden retriever sitting in a park, warm afternoon sunlight, shallow depth of field, no people, no text

3

Advanced

Combine techniques: weighted terms, camera angles, and precise lighting.

Advanced prompt result: golden retriever with 85mm lens bokeh and golden hour lighting

Golden retriever, sitting alert, autumn park, golden hour backlight, 85mm lens, bokeh, eye-level shot, no humans, no watermark

The 5-step prompt structure that works

This structure has been tested across thousands of generations. Follow it to reduce trial-and-error and get consistent results.

1

Define the subject

What is it? Who is it? Be specific in one sentence. The more concrete your subject, the better the output.

XBad: "a person"
YGood: "a 30-year-old professional woman with short dark hair"
2

Define the objective

What do you want to achieve? This determines how the AI interprets your request.

  • Generate new: Create from scratch based on description
  • Edit existing: Modify uploaded image while preserving core elements
  • Create variation: Generate alternatives of existing image
3

Add constraints (most important)

Constraints prevent the AI from making unwanted changes. This is where most prompts fail.

keep subject unchanged
preserve composition and pose
no extra people
no text or watermarks
natural skin texture
maintain original colors

Pro tip: When editing portraits, always include "keep the same person" and "preserve facial features" to prevent identity changes.

4

Add style and lighting

Style and lighting dramatically affect the mood and quality of your output.

Style keywords:

photorealisticcinematic3D renderoil paintingwatercoloranimeminimalistvintage film

Lighting keywords:

soft studio lightgolden hourdramatic side lightneon glowovercast diffusedbacklit silhouette
5

Iterate with discipline

Don't rewrite everything when results aren't perfect. Scientific iteration is key.

  • Change only ONE variable per attempt
  • Keep what works, remove what doesn't
  • Save successful prompts as templates

Complete prompt examples

Here are tested prompts that follow the 5-step structure. Copy and customize them for your needs.

Professional Headshot Enhancement

Perfect for LinkedIn profiles, company websites, and professional bios. Enhances existing photo while maintaining identity.

Professional headshot, soft studio lighting, neutral gray background, keep the same person, preserve facial features, natural skin texture with subtle retouch, sharp eyes, confident expression, head and shoulders framing, no heavy filters, no text
Key constraints:keep same personnatural skinno text

Product on White Background

Ideal for Amazon listings, Shopify stores, and product catalogs. Creates clean, professional product images.

[Product name] on pure white background, studio product photography, soft even lighting from multiple angles, sharp focus on product details, no shadows, no reflections, e-commerce ready, centered composition, high detail
Key constraints:pure white bgno shadowscentered

3D Figurine Style Portrait

Great for avatars, profile pictures, and creative personal branding. Transforms photos into charming 3D characters.

3D rendered figurine of [subject], Pixar-style character, smooth plastic texture, pastel color palette, simple geometric background, cute proportions with slightly oversized head, friendly expression, soft ambient occlusion lighting, no text, no watermark
Key constraints:Pixar stylepastel colorsno watermark

Cinematic Photo Enhancement

Transforms casual photos into movie-like stills. Perfect for social media and creative portfolios.

Cinematic color grading, 35mm film look, subtle grain, teal and orange color palette, shallow depth of field, keep subject unchanged, preserve original composition, enhanced contrast, atmospheric lighting, no cropping
Key constraints:keep subjectpreserve compositionno crop

Lifestyle Product Scene

Creates aspirational product imagery for social media marketing and brand storytelling.

[Product] in lifestyle setting, natural daylight through window, minimalist Scandinavian interior, wooden table surface, subtle props that complement product, shallow depth of field on product, warm color temperature, no text overlay
Key constraints:natural lightminimalistproduct focus

Using negative prompts effectively

Negative prompts tell the AI what to avoid. They're essential for preventing common problems.

Quality Issues

no blurryno low qualityno pixelatedno noisyno overexposedno underexposed

Unwanted Elements

no textno watermarkno logono signatureno borderno frame

Portrait Issues

no extra fingersno distorted faceno unnatural poseno plastic skinno dead eyes

Composition Issues

no croppedno cut offno bad framingno tiltedno cluttered background

Common prompt mistakes to avoid

Too many instructions at once

The AI gets confused with 10+ different directions. Start simple, add complexity gradually.

Conflicting styles

"Photorealistic anime style" or "minimalist with lots of detail" creates confusion. Pick one direction.

Missing identity constraints

When editing portraits, always include "keep the same person" or the AI may change facial features.

Requesting text overlays

AI image generators struggle with text. Add text in post-processing with design tools instead.

Vague descriptions

"Make it better" or "more professional" doesn't give the AI enough to work with. Be specific about what "better" means.

Rewriting the entire prompt after one failure

Change one variable at a time. If lighting failed, fix only the lighting keywords.

Advanced prompt techniques

Prompt Templates

Create reusable templates with placeholders. This ensures consistency across multiple generations.

[SUBJECT] in [SETTING], [STYLE] photography, [LIGHTING], [CONSTRAINTS]

A/B Testing Your Prompts

When optimizing prompts, test systematically:

  1. Keep a base prompt that works reasonably well
  2. Create a variation changing only one element
  3. Generate 2-3 images with each version
  4. Compare results and keep the winner
  5. Repeat with next element

Camera and Lens Terms

Photography terms can dramatically change output:

85mm lens — Portrait compression, flattering
35mm lens — Environmental, wider context
macro lens — Extreme close-up, details
wide angle — Expansive, dramatic perspective
f/1.4 aperture — Extreme bokeh, dreamy
f/11 aperture — Sharp throughout, landscapes

Frequently Asked Questions

Quick answers to common questions about writing effective Nano Banana prompts

A good prompt has three qualities: specificity (clearly describe your subject), constraints (tell the AI what NOT to do), and iteration-friendly structure (easy to modify one element at a time). The best prompts are detailed enough to guide the AI but not so complex that they create conflicts.

Most effective prompts are 30-80 words. Under 20 words is usually too vague and gives unpredictable results. Over 100 words often creates conflicts between instructions. Start with a medium-length prompt and adjust based on results.

Essential constraints include: 'keep subject unchanged' (for edits), 'no text', 'no watermark', 'preserve composition'. For portraits, add 'keep the same person' and 'natural skin texture'. The specific constraints depend on your use case, but always include at least 2-3.

Change only ONE variable at a time. If the lighting is wrong, adjust only lighting keywords. If the composition failed, adjust only composition terms. Keep everything else stable. This systematic approach helps you identify what works and what doesn't.

Generally, no. Conflicting styles like 'photorealistic anime' or 'minimalist with lots of detail' confuse the AI. Pick one primary style direction. If you want to blend styles, be very specific about which elements should have which style.

This happens when you don't include identity preservation constraints. Always add 'keep the same person', 'preserve facial features', and 'maintain identity' when editing portraits. These phrases tell the AI to prioritize keeping the original face.

Negative prompts tell the AI what to avoid: 'no blurry', 'no text', 'no extra fingers'. Use them when you consistently get unwanted elements in outputs. They're especially useful for preventing common AI artifacts like distorted hands or unwanted text.

Create a template prompt and save it. Use the same base structure, lighting, and style keywords. Only change the subject-specific details. This ensures visual consistency across a series of images for your project.

AI models sometimes add text unprompted, especially with product or poster-style compositions. Always include 'no text', 'no words', 'no letters' in your constraints. If text still appears, add 'clean image without typography'.

Photography terms significantly impact output. '85mm lens' creates portrait-style compression. '35mm' gives environmental context. 'macro' produces extreme close-ups. 'f/1.4' creates dreamy bokeh. Use these terms to control composition and depth of field.

Check our Best Prompts page for curated copy-paste templates. We have specialized collections for headshots, product photos, 3D figurines, and more. Each template follows the 5-step structure and includes tested constraints.

API access is not offered at this time. Nano Banana is available through the web interface only. You get 1 free generation without sign-up, plus 3 more after signing up (no credit card required).