nano-banana-pro

name	nano-banana-pro
description	Generate images using Google Gemini 3 Pro - up to 4K resolution, 14 reference images, advanced reasoning
user_invocable	true
allowed_tools	["Bash"]

Nano Banana Pro

Generate images using Google's Gemini 3 Pro Image model. Professional quality, up to 4K resolution, advanced reasoning.

⚡ Launch the result

The image is the deliverable — show it the instant it's done instead of leaving the user to hunt for a path. After saving ANY generated image, immediately open it (open <abs_path> on macOS → Preview; xdg-open <abs_path> on Linux) and surface the file to the user, before writing any explanation. Lead with the image; keep commentary short. Applies to every image path.

Trigger Conditions

Activate when:

• User says "generate image", "create image", "make me an image" AND wants high quality
• User invokes /nano-banana-pro or /nbp
• User needs higher resolution
• User needs character consistency across multiple images
• Complex prompts requiring advanced reasoning

Model Details

Property	Value
Model ID	gemini-3-pro-image-preview
Speed	Slower (more processing)
Resolution	Up to 4K
Max Reference Images	14 (6 objects, 5 humans max)

Prerequisites

1. API Key: Set GEMINI_API_KEY in .env file
2. Package: pip install google-genai Pillow python-dotenv

Get API key: https://aistudio.google.com/app/apikey

Usage

Basic Generation

from google import genai
from google.genai import types
from dotenv import load_dotenv
import os

load_dotenv()
client = genai.Client(api_key=os.environ.get('GEMINI_API_KEY'))

response = client.models.generate_content(
    model='gemini-3-pro-image-preview',
    contents='YOUR PROMPT HERE',
    config=types.GenerateContentConfig(
        response_modalities=['IMAGE'],
    ),
)

# Save the image, then launch it so the user sees it immediately
import subprocess
for part in response.parts:
    if part.inline_data:
        image = part.as_image()
        image.save('output.png')
        subprocess.run(['open', 'output.png'])  # macOS (use 'xdg-open' on Linux)
        print('Image saved + opened: output.png')

With Aspect Ratio and Image Size

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

SDK note: older google-genai versions don't expose types.ImageConfig. Guard it with if hasattr(types, 'ImageConfig'): and only add image_config when present (otherwise omit it) so the call still runs across SDK versions.

With Reference Images

from PIL import Image

ref_image = Image.open('reference.png')

response = client.models.generate_content(
    model='gemini-3-pro-image-preview',
    contents=['Create a variation of this image in watercolor style', ref_image],
    config=types.GenerateContentConfig(
        response_modalities=['IMAGE'],
    ),
)

Image Size Options (Gemini 3 Pro Only)

image_size	Use Case
1K	Standard web use (default)
2K	High-quality displays
4K	Print, professional assets

Note: Use uppercase K (e.g., '2K'). Lowercase will be rejected.

Aspect Ratios

Ratio	Use Case
1:1	Square (social media, profile pics)
16:9	Landscape (presentations, thumbnails)
9:16	Portrait (stories, mobile)
4:3	Standard photo
3:4	Portrait photo
21:9	Ultrawide/cinematic

Advanced Capabilities

• Character consistency: Same characters across multiple generations
• World knowledge: Semantic understanding for realistic scenes
• Complex prompts: Better understanding of nuanced instructions
• Professional quality: Suitable for marketing, product shots

When to Use Pro vs Flash

Use Case	Model
4K resolution	Pro
Character consistency	Pro
Complex scenes	Pro
Professional assets	Pro
Many reference images	Pro
Quick iterations	Flash
Cost-sensitive	Flash
Batch generation	Flash

Limitations

• Preview status (may change)
• Higher cost ($0.134 vs $0.039)
• Slower generation
• SynthID watermark on all images

Documentation

Full docs: External_Documentation/GOOGLE_GEMINI_IMAGE_GENERATION.md