szweibel/nano-banana-pro

Nano Banana Pro (Gemini 3 Pro Image) - API Usage & Prompting Guide

Build powerful image generation and editing applications using Google's Nano Banana Pro (Gemini 3 Pro Image), the state-of-the-art AI image model released November 2025.

Overview

Nano Banana Pro (officially gemini-3-pro-image-preview) is Google's most advanced image generation model, offering:

• 4K Resolution Output - Print-quality images up to 4K
• Advanced "Thinking" - Built-in reasoning for complex compositions
• Search Grounding - Real-time data integration via Google Search
• Multi-Image Support - Up to 14 reference images in a single generation
• Accurate Text Rendering - High-fidelity text in multiple languages
• Multi-Turn Conversations - Iterative refinement through chat

Model Comparison

| Feature | Gemini 3 Pro Image (Pro) | Gemini 2.5 Flash Image (Flash) | |---------|-------------------------|--------------------------------| | Model ID | gemini-3-pro-image-preview | gemini-2.5-flash-image | | Max Resolution | 4K | 1024px | | Search Grounding | ✅ Yes | ❌ No | | Thinking Process | ✅ Yes | ❌ No | | Reference Images | Up to 14 | Up to 14 | | Speed | Slower, higher quality | Faster, cost-effective | | Cost (2K) | ~$0.134 | Lower | | Cost (4K) | ~$0.24 | N/A |

Setup & Installation

1. Get API Key

1. Go to Google AI Studio
2. Sign in with your Google account
3. Click "Get API key" in the left sidebar
4. Enable billing (Nano Banana Pro requires paid access)
5. Copy your API key

2. Install SDK

Python:

pip install google-genai

JavaScript/Node.js:

npm install @google/genai

3. Set Environment Variable

export GOOGLE_API_KEY="your-api-key-here"

Or create a .env file:

GOOGLE_API_KEY=your-api-key-here

API Usage Examples

Basic Text-to-Image (Python)

from google import genai
from google.genai import types

# Initialize client
client = genai.Client(api_key="YOUR_API_KEY")

# Generate image
response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents=["A futuristic cityscape at sunset with flying cars, photorealistic, 4K quality"],
    config=types.GenerateContentConfig(
        response_modalities=['TEXT', 'IMAGE']
    )
)

# Save the generated image (access via candidates)
for candidate in response.candidates:
    for part in candidate.content.parts:
        if part.inline_data:
            with open("cityscape.png", "wb") as f:
                f.write(part.inline_data.data)
            print("Image saved!")

Basic Text-to-Image (JavaScript)

const { GoogleGenAI } = require("@google/genai");

const ai = new GoogleGenAI({ apiKey: process.env.GOOGLE_API_KEY });

async function generateImage() {
    const response = await ai.models.generateContent({
        model: "gemini-3-pro-image-preview",
        contents: ["A futuristic cityscape at sunset with flying cars, photorealistic, 4K quality"],
    });

    // Process response
    for (const part of response.parts) {
        if (part.inlineData) {
            // Save or process image data
            const imageBuffer = Buffer.from(part.inlineData.data, 'base64');
            require('fs').writeFileSync('cityscape.png', imageBuffer);
            console.log("Image saved!");
        }
    }
}

generateImage();

High-Resolution Output (2K/4K)

from google import genai
from google.genai import types

client = genai.Client()

response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents=["Professional product photography of a luxury watch on marble surface"],
    config=types.GenerateContentConfig(
        response_modalities=['TEXT', 'IMAGE'],
        image_config=types.ImageConfig(
            aspect_ratio="16:9",
            image_size="4K"  # Options: "1K", "2K", "4K"
        )
    )
)

for candidate in response.candidates:
    for part in candidate.content.parts:
        if part.inline_data:
            with open("watch_4k.png", "wb") as f:
                f.write(part.inline_data.data)

Image Editing

from google import genai
from google.genai import types
from PIL import Image

client = genai.Client()

# Load your existing image
original_image = Image.open("photo.jpg")

# Edit with text prompt
response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents=[
        "Add a rainbow in the sky and make the colors more vibrant",
        original_image
    ],
    config=types.GenerateContentConfig(
        response_modalities=['TEXT', 'IMAGE']
    )
)

for candidate in response.candidates:
    for part in candidate.content.parts:
        if part.inline_data:
            with open("edited_photo.png", "wb") as f:
                f.write(part.inline_data.data)

Multi-Turn Conversation (Iterative Refinement)

from google import genai
from google.genai import types

client = genai.Client()

# Start a chat session for iterative refinement
chat = client.chats.create(
    model="gemini-3-pro-image-preview",
    config=types.GenerateContentConfig(
        response_modalities=['TEXT', 'IMAGE'],
    )
)

# First generation
response1 = chat.send_message("Create a logo for a tech startup called 'NeuralNet'")
print(f"Response 1: {response1.text}")

# Refine the image
response2 = chat.send_message("Make the colors more vibrant and add a neural network pattern in the background")
print(f"Response 2: {response2.text}")

# Further refinement
response3 = chat.send_message("Change the font to something more modern and bold")

# Save final image
for candidate in response3.candidates:
    for part in candidate.content.parts:
        if part.inline_data:
            with open("neuralnet_logo_final.png", "wb") as f:
                f.write(part.inline_data.data)

Search Grounding (Real-Time Data)

from google import genai
from google.genai import types

client = genai.Client()

response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents=["Create an infographic showing the current weather in New York City"],
    config=types.GenerateContentConfig(
        response_modalities=['TEXT', 'IMAGE'],
        tools=[{"google_search": {}}]  # Enable search grounding
    )
)

for candidate in response.candidates:
    for part in candidate.content.parts:
        if part.inline_data:
            with open("nyc_weather.png", "wb") as f:
                f.write(part.inline_data.data)

Multiple Reference Images (Up to 14)

from google import genai
from google.genai import types
from PIL import Image

client = genai.Client()

# Load reference images
ref_image1 = Image.open("style_reference.jpg")
ref_image2 = Image.open("character_reference.jpg")
ref_image3 = Image.open("background_reference.jpg")

response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents=[
        "Create a new scene combining the art style from the first image, "
        "the character from the second image, and the background atmosphere "
        "from the third image",
        ref_image1,
        ref_image2,
        ref_image3
    ],
    config=types.GenerateContentConfig(
        response_modalities=['TEXT', 'IMAGE']
    )
)

for candidate in response.candidates:
    for part in candidate.content.parts:
        if part.inline_data:
            with open("combined_scene.png", "wb") as f:
                f.write(part.inline_data.data)

Batch Image Generation

For large-scale production with up to 24-hour turnaround (50% cost savings):

from google import genai

client = genai.Client()

# Upload batch requests file (JSONL format)
uploaded_file = client.files.upload(path="batch_requests.jsonl")

# Create batch job
file_batch_job = client.batches.create(
    model="gemini-2.5-flash-image",
    src=uploaded_file.name,
)

print(f"Batch job created: {file_batch_job.name}")

# Check status later
job = client.batches.get(name=file_batch_job.name)
print(f"Status: {job.state}")

Prompting Best Practices

Core Principle

"Describe the scene, don't just list keywords."

Use flowing narrative descriptions with context rather than keyword lists.

Template 1: Photorealistic Scenes

[Shot Type] of [Subject] in [Environment], [Lighting], [Mood/Atmosphere].
Shot on [Camera] with [Lens], [Additional Details].

Example:
"Wide-angle shot of a chef preparing pasta in a modern restaurant kitchen,
warm golden hour lighting streaming through industrial windows, energetic
and professional atmosphere. Shot on Sony A7IV with 24mm f/1.4 lens,
shallow depth of field, steam rising from the pot."

Template 2: Stylized Illustrations & Stickers

[Art Style] illustration of [Subject], [Characteristics], [Color Palette],
[Additional Style Notes].

Example:
"Kawaii-style sticker illustration of a smiling coffee cup with big eyes,
chibi proportions, pastel pink and brown color palette, white outline,
glossy finish, cute and friendly expression."

Template 3: Text-Heavy Designs (Logos, Posters)

[Design Type] for [Brand/Concept], [Text Content], [Font Style],
[Color Scheme], [Layout].

Example:
"Modern minimalist logo for 'CloudFlow' SaaS company, sans-serif typeface
with flowing cloud icon integrated into the 'C', blue and white gradient
color scheme, clean and professional."

Note: Keep overlaid text under 25 characters for best accuracy.

Template 4: Product Photography

[Shot Type] product photography of [Product] on [Surface/Background],
[Lighting Setup], [Angle], [Mood].

Example:
"Hero shot product photography of wireless headphones on white marble
surface, studio lighting with soft key light and rim light highlighting
edges, 45-degree angle, premium and modern aesthetic."

Template 5: Minimalist Design

Minimalist [Design Type] featuring [Single Subject], [Composition],
[Color Palette], [Negative Space].

Example:
"Minimalist poster design featuring a single red umbrella, centered
composition with rule of thirds, black and red color palette,
abundant white negative space creating elegant simplicity."

Advanced Features

Character Consistency (Up to 5 Characters)

Maintain the same characters across multiple generations:

# First generation - establish character
response1 = chat.send_message("Create a character: a friendly robot with blue eyes")

# Subsequent generations maintain consistency
response2 = chat.send_message("Show the same robot character waving hello")
response3 = chat.send_message("Show the same robot in a different pose, reading a book")

Multilingual Text Generation

Generate text in multiple languages within images:

response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents=[
        "Create a welcome poster with 'Welcome' written in English, Spanish, "
        "French, and Japanese in a modern typographic layout"
    ],
)

Aspect Ratio Control

config=types.GenerateContentConfig(
    image_config=types.ImageConfig(
        aspect_ratio="16:9"  # Options: "1:1", "16:9", "9:16", "3:4", "4:3"
    )
)

Pro Tips

For Best Results:

1. Be Specific - Include details about lighting, camera angles, materials, mood
2. Use Narrative Flow - Write in sentences, not keyword lists
3. One Change at a Time - When editing, make incremental adjustments
4. Leverage Search - Use search grounding for real-world accuracy (weather, facts, current events)
5. Iterate in Conversation - Use multi-turn chat for progressive refinement
6. Reference Real Cameras/Lenses - "Shot on Canon 5D, 50mm f/1.8" produces more photorealistic results
7. Specify Art Movements - "in the style of Art Nouveau" or "cyberpunk aesthetic"
8. Keep Text Short - For text in images, stay under 25 characters for best accuracy

What to Avoid:

• ❌ Vague requests like "make it better" or "improve this"
• ❌ Keyword lists: "sunset, beach, waves, colors, beautiful"
• ❌ Changing multiple elements simultaneously
• ❌ Overly long text overlays (>25 characters)
• ❌ Copyrighted characters or trademarked content

Error Handling

from google import genai
from google.genai import types

client = genai.Client()

try:
    response = client.models.generate_content(
        model="gemini-3-pro-image-preview",
        contents=["Your prompt here"],
        config=types.GenerateContentConfig(
            response_modalities=['TEXT', 'IMAGE']
        )
    )

    for candidate in response.candidates:
        for part in candidate.content.parts:
            if part.inline_data:
                with open("output.png", "wb") as f:
                    f.write(part.inline_data.data)

except Exception as e:
    print(f"Error: {e}")

Rate Limits & Quotas

• Free Tier: Not available for Nano Banana Pro
• Paid Tier: Standard rate limits apply (check current limits in AI Studio)
• Batch API: Higher limits with 24-hour turnaround, 50% cost savings

Pricing (Approximate)

• 1K Resolution: ~$0.134 per image
• 2K Resolution: ~$0.134 per image
• 4K Resolution: ~$0.24 per image
• Batch API: 50% discount on above prices

Check Google AI Pricing for current rates.

Safety & Compliance

• All images include SynthID watermarks for authentication
• Respect copyright and intellectual property when using reference images
• Follow Google's Gemini API Terms of Service
• Generated images must not violate content policies

Use Cases

Professional Applications:

• 📊 Data Visualization - Infographics with search-grounded real-time data
• 🎨 Marketing Assets - High-resolution product mockups and promotional imagery
• 📝 Document Design - Technical diagrams, flowcharts with legible text
• 🎬 Storyboarding - Character-consistent scene generation
• 🏢 Presentation Graphics - Custom illustrations for slides and reports

Creative Projects:

• 🖼️ Concept Art - Iterative refinement of artistic visions
• 🎮 Game Assets - Character designs with consistency across variations
• 📚 Book Illustrations - Story scenes with maintained character appearances
• 🌐 Web Design - Custom hero images and background assets

Additional Resources

• Official Documentation: ai.google.dev/gemini-api/docs/image-generation
• Google AI Studio: aistudio.google.com
• Prompt Guide: ai.google.dev/gemini-api/docs/prompting-strategies
• API Pricing: ai.google.dev/pricing
• Developer Forum: ai.google.dev/community

Quick Reference Card

# Basic generation
from google import genai
from google.genai import types

client = genai.Client(api_key="YOUR_KEY")

# Simple text-to-image
response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents=["Your detailed prompt here"],
    config=types.GenerateContentConfig(
        response_modalities=['TEXT', 'IMAGE']
    )
)

# With full configuration
response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents=["Your prompt"],
    config=types.GenerateContentConfig(
        response_modalities=['TEXT', 'IMAGE'],
        image_config=types.ImageConfig(
            aspect_ratio="16:9",
            image_size="4K"
        ),
        tools=[{"google_search": {}}]  # Enable search
    )
)

# Save image (access via candidates)
for candidate in response.candidates:
    for part in candidate.content.parts:
        if part.inline_data:
            with open("output.png", "wb") as f:
                f.write(part.inline_data.data)

---

Last Updated: November 2025 Model Version: Gemini 3 Pro Image (gemini-3-pro-image-preview)