Nano Banana Build

Nano Banana Image Generation Skill

Use this skill to generate and edit images using the google-genai Python SDK with Gemini's specialized image models (Nano Banana).

Quick Start Setup

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

client = genai.Client()

Reference Materials

• [Model Capabilities](references/model_capabilities.md): Comparison of Gemini 2.5 vs 3 Pro, resolutions, and token costs.
• [Image Generation](references/image_generation.md): Text-to-Image, Interleaved Text/Image.
• [Image Editing](references/image_editing.md): Subject Customization, Style Transfer, Multi-turn Editing.
• [Thinking Process](references/thinking_process.md): Understanding thoughts and signatures (Gemini 3 Pro).
• [Recipes](references/recipes.md): Extensive collection of examples (Logos, Stickers, Mockups, Comics, etc.).
• [Source Code](references/source_code.md): Deep inspection of SDK internals.

Available Models

• `gemini-2.5-flash-image` (Nano Banana): Fast, high-quality generation and editing. Best for most use cases.
• `gemini-3-pro-image-preview` (Nano Banana Pro): Highest fidelity, supports 2K and 4K resolution, complex prompt adherence, and grounding.

Common Workflows

1. Fast Generation

response = client.models.generate_content(
    model='gemini-2.5-flash-image',
    contents='A cute robot eating a banana',
    config=types.GenerateContentConfig(
        response_modalities=['IMAGE']
    )
)

2. High-Quality Editing

response = client.models.generate_content(
    model='gemini-3-pro-image-preview',
    contents=[
        types.Part.from_uri(file_uri='gs://.../shoe.jpg', mime_type='image/jpeg'),
        "Change the color of the shoe to neon green."
    ],
    config=types.GenerateContentConfig(response_modalities=['IMAGE'])
)

Image Editing

Nano Banana models support advanced image-to-image capabilities.

Subject Customization

Change the style or attributes of a subject while preserving its identity.

with open("dog.jpg", "rb") as f:
    image_bytes = f.read()

response = client.models.generate_content(
    model='gemini-2.5-flash-image',
    contents=[
        types.Part.from_bytes(data=image_bytes, mime_type="image/jpeg"),
        "Create a pencil sketch of this dog wearing a cowboy hat."
    ],
    config=types.GenerateContentConfig(response_modalities=["IMAGE"])
)

Style Transfer

Apply the style of one image to the content of another.

response = client.models.generate_content(
    model='gemini-2.5-flash-image',
    contents=[
        types.Part.from_uri(file_uri="gs://.../style_ref.png", mime_type="image/png"),
        "Using the concepts and colors from this image, generate a kitchen with the same aesthetic."
    ],
    config=types.GenerateContentConfig(response_modalities=["IMAGE"])
)

Multi-turn Editing (Chat)

Iteratively refine an image in a chat session.

chat = client.chats.create(
    model='gemini-2.5-flash-image',
    config=types.GenerateContentConfig(response_modalities=['TEXT', 'IMAGE'])
)

# First turn
response = chat.send_message("Create an image of a perfume bottle.")
image_data = response.candidates[0].content.parts[1].inline_data.data

# Second turn (pass previous image back)
response = chat.send_message([
    types.Part.from_bytes(data=image_data, mime_type="image/png"),
    "Make the bottle purple."
])

Multiple Reference Images

Combine elements from multiple images.

response = client.models.generate_content(
    model='gemini-3-pro-image-preview',
    contents=[
        types.Part.from_uri(file_uri="gs://.../person.jpg", mime_type="image/jpeg"),
        types.Part.from_uri(file_uri="gs://.../background.png", mime_type="image/png"),
        "Generate an image of the person from the first image standing in the background from the second image."
    ]
)

Image Generation

Gemini 2.5 Flash Image and Gemini 3 Pro Image (Nano Banana) support advanced image generation capabilities.

Text-to-Image

Generate images from text prompts.

response = client.models.generate_content(
    model='gemini-2.5-flash-image',
    contents="A futuristic cityscape at sunset",
    config=types.GenerateContentConfig(
        response_modalities=["IMAGE"],
        image_config=types.ImageConfig(
            aspect_ratio="16:9",
            # image_size="1K" # Optional
        ),
        candidate_count=1,
    ),
)

for part in response.candidates[0].content.parts:
    if part.inline_data:
        # Display or save image
        image = part.as_image()
        image.save("generated.png")

Interleaved Text & Image

Generate tutorials or stories with mixed text and image outputs.

response = client.models.generate_content(
    model='gemini-2.5-flash-image',
    contents="Create a 3-step tutorial for making a sandwich. For each step, provide text and an illustration.",
    config=types.GenerateContentConfig(
        response_modalities=["TEXT", "IMAGE"],
    ),
)

Image Sizes (Gemini 3 Pro Image)

Gemini 3 Pro Image supports 1K, 2K, and 4K.

config=types.GenerateContentConfig(
    image_config=types.ImageConfig(
        image_size="2K",
        aspect_ratio="1:1"
    )
)

Model Capabilities & Differences

Choose the model best suited for your specific use case.

Aspect Ratios & Resolutions

Gemini 2.5 Flash Image

Gemini 3 Pro Image Preview

Supports 1K, 2K, and 4K sizes. Use uppercase 'K' (e.g., 2K).

Configuration Example

config = types.GenerateContentConfig(
    image_config=types.ImageConfig(
        aspect_ratio="16:9",
        image_size="2K", # Only for Gemini 3 Pro
    )
)

Nano Banana Recipes

Practical examples for common tasks using Gemini 2.5 Flash Image and Gemini 3 Pro Image.

Blank Canvas (Aspect Ratio Control)

Force a specific aspect ratio by providing a blank image.

from PIL import Image
import io

def create_canvas(width=1280, height=720):
    img = Image.new("RGB", (width, height), "white")
    buf = io.BytesIO()
    img.save(buf, format="PNG")
    return types.Part.from_bytes(data=buf.getvalue(), mime_type="image/png")

response = client.models.generate_content(
    model='gemini-2.5-flash-image',
    contents=[
        create_canvas(1280, 720),
        "A cinematic wide shot of a desert planet."
    ]
)

Virtual Try-On

Place a garment on a model.

response = client.models.generate_content(
    model='gemini-2.5-flash-image',
    contents=[
        types.Part.from_uri(file_uri="gs://.../model.jpg", mime_type="image/jpeg"),
        types.Part.from_uri(file_uri="gs://.../dress.jpg", mime_type="image/jpeg"),
        "Realistically place the dress from the second image onto the person in the first image."
    ]
)

Product Recontextualization

Place a product in a new scene.

response = client.models.generate_content(
    model='gemini-2.5-flash-image',
    contents=[
        types.Part.from_uri(file_uri="gs://.../product.png", mime_type="image/png"),
        "Place this product on a marble countertop in a luxury kitchen."
    ]
)

Character Consistency

Maintain a character across different scenes.

response = client.models.generate_content(
    model='gemini-2.5-flash-image',
    contents=[
        types.Part.from_uri(file_uri="gs://.../character_ref.png", mime_type="image/png"),
        "Generate an image of this character eating an apple in a park."
    ]
)

Stylized Stickers

Create stickers with transparent backgrounds.

response = client.models.generate_content(
    model="gemini-2.5-flash-image",
    contents="A kawaii-style sticker of a happy red panda wearing a tiny bamboo hat. It's munching on a green bamboo leaf. The design features bold, clean outlines, simple cel-shading, and a vibrant color palette. The background must be white.",
)

Logo Design

Create modern, minimalist logos using accurate text rendering (Gemini 3 Pro).

response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents="Create a modern, minimalist logo for a coffee shop called 'The Daily Grind'. The text should be in a clean, bold, sans-serif font. The color scheme is black and white. Put the logo in a circle. Use a coffee bean in a clever way.",
    config=types.GenerateContentConfig(
        image_config=types.ImageConfig(aspect_ratio="1:1")
    )
)

Product Mockups

Generate high-resolution commercial photography.

response = client.models.generate_content(
    model="gemini-2.5-flash-image",
    contents="A high-resolution, studio-lit product photograph of a minimalist ceramic coffee mug in matte black, presented on a polished concrete surface. The lighting is a three-point softbox setup designed to create soft, diffused highlights and eliminate harsh shadows. The camera angle is a slightly elevated 45-degree shot to showcase its clean lines. Ultra-realistic, with sharp focus on the steam rising from the coffee. Square image.",
)

Minimalist Backgrounds

Create negative space designs for text overlays.

response = client.models.generate_content(
    model="gemini-2.5-flash-image",
    contents="A minimalist composition featuring a single, delicate red maple leaf positioned in the bottom-right of the frame. The background is a vast, empty off-white canvas, creating significant negative space for text. Soft, diffused lighting from the top left. Square image.",
)

Sequential Art (Comics)

Create storytelling panels using character references.

response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents=[
        "Make a 3 panel comic in a gritty, noir art style with high-contrast black and white inks. Put the character in a humorous scene.",
        types.Part.from_uri(file_uri="gs://.../character.jpg", mime_type="image/jpeg")
    ],
)

Grounding with Google Search

Generate images based on real-time data.

response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents="Make a simple but stylish graphic of last night's Arsenal game in the Champion's League",
    config=types.GenerateContentConfig(
        tools=[types.Tool(google_search=types.GoogleSearch())],
        image_config=types.ImageConfig(aspect_ratio="16:9")
    )
)

Sketch to Life

Turn a rough drawing into a polished image.

response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents=[
        types.Part.from_uri(file_uri="gs://.../car_sketch.png", mime_type="image/png"),
        "Turn this rough pencil sketch of a futuristic car into a polished photo of the finished concept car in a showroom. Keep the sleek lines and low profile from the sketch but add metallic blue paint and neon rim lighting."
    ],
)

Multi-Image Composition (Advanced)

Combine up to 14 reference images (Gemini 3 Pro).

response = client.models.generate_content(
    model="gemini-3-pro-image-preview",
    contents=[
        "An office group photo of these people, they are making funny faces.",
        types.Part.from_uri(file_uri="gs://.../person1.png", mime_type="image/png"),
        types.Part.from_uri(file_uri="gs://.../person2.png", mime_type="image/png"),
        types.Part.from_uri(file_uri="gs://.../person3.png", mime_type="image/png"),
        # ... up to 14 images total
    ],
    config=types.GenerateContentConfig(
        image_config=types.ImageConfig(aspect_ratio="5:4", image_size="2K")
    )
)

Google GenAI SDK Source Code (Nano Banana)

Use web_fetch to retrieve raw code for deep inspection of image generation parameters.

Base URL: https://raw.githubusercontent.com/googleapis/python-genai/main/google/genai/

Key Modules

Models (Generation)

• File: models.py
• URL: https://raw.githubusercontent.com/googleapis/python-genai/main/google/genai/models.py
• Purpose: generate_images (helper) and generate_content (core logic).

Types (Configuration)

• File: types.py
• URL: https://raw.githubusercontent.com/googleapis/python-genai/main/google/genai/types.py
• Purpose: Definitions for ImageConfig, GenerateImagesConfig, PersonGeneration, SafetyFilterLevel.

Client

• File: client.py
• URL: https://raw.githubusercontent.com/googleapis/python-genai/main/google/genai/client.py

Thinking Process & Thought Signatures

Gemini 3 Pro Image Preview uses a default "Thinking" process to refine composition and logic before generation.

Accessing Thoughts

You can inspect the thoughts that led to the final image.

for part in response.parts:
    if part.thought:
        if part.text:
            print(f"Thought: {part.text}")
        elif part.inline_data:
            # Interim thought image
            image = part.as_image()
            image.show()

Thought Signatures

Crucial for Multi-turn: Thought signatures preserve reasoning context across turns.

• Automatic Handling: If using client.chats or appending the full response object to history, the SDK handles this automatically.
• Manual Handling: If manually constructing JSON payloads, you must pass back the thought_signature field exactly as received.

Structure

• The first non-thought text part has a signature.
• All inline image parts (except thought images) have signatures.
• Thought parts do not have signatures.

Failure to circulate thought signatures may cause the response to fail.