Nano Banana Pro

Generate images using the Nano Banana Pro model via Google Generative AI API. Supports any image type: illustrations, realistic photos, product mockups, logos, scenes, patterns, UI concepts, and more.

Setup (first time only)

If GOOGLE_AI_API_KEY is not set:

1. Go to https://aistudio.google.com/apikey
2. Click "Create API Key" (free, no credit card needed)
3. Add to shell profile:

   echo 'export GOOGLE_AI_API_KEY="your-key-here"' >> ~/.zshrc
   source ~/.zshrc
   

Interactive Workflow

1. Ask for concept - What image do you want to create?
2. Ask for type - Illustration, photo, mockup, logo, etc.
3. Ask for model - Show options with cost (see Models below)
4. Ask for size - What dimensions/aspect ratio?
5. Ask for style - User describes their aesthetic, colors, or picks a preset
6. Build prompt - Combine type + style + concept + size into a generation prompt
7. Generate - Run scripts/generate.py
8. Preview - Open in default viewer

Models & Cost

Present these options and confirm cost before generating:

Always confirm: "This will cost approximately $X.XX. Proceed?"

Sizes

Ask user what size they need. Add the size instruction to the END of the prompt:

Building the Prompt

If user has brand/style preferences

Ask: "Do you have specific colors, aesthetic, or brand style you'd like?"

Use their answer to build the prompt with this structure:

[IMAGE TYPE], [AESTHETIC/STYLE], [COLOR PALETTE].

[CONCEPT DESCRIPTION]

[STYLE CONSTRAINTS]. [ASPECT RATIO]

Image Type Presets

Illustrations

Watercolor - Soft, whimsical, hand-drawn

Whimsical hand-drawn watercolor illustration, soft [USER_COLOR] palette with warm accents.

Simple composition with ONE/TWO main elements:
[CONCEPT]

Style: Simple and clean like Notion illustrations. NOT complex. NOT busy.
NO text. Cream background. Square 1:1.

Monochromatic - Single color, clean lines

MONOCHROMATIC [USER_COLOR] illustration - use ONLY shades of [USER_HEX] throughout. Ultra high quality.

[CONCEPT]

Style: Clean lines, modern like Notion/Slack. NO other colors.

Flat/Editorial - Detailed, informative

Detailed flat illustration in [USER_COLOR] palette, editorial quality.

[CONCEPT]

Style: Clean linework. Modern editorial feel. High detail.
Wide format 16:9.

Minimal line art - Simple outlines, one accent

Minimal line art, thin black outlines with [USER_COLOR] as single accent color.

[CONCEPT]

Style: Minimal, lots of whitespace. NO fills except accent. White background.

Photos / Realistic

Product photo - Clean, studio-style

Professional product photography, studio lighting, clean [white/colored] background.

[PRODUCT DESCRIPTION]

Ultra high quality, sharp focus, soft shadows. [ASPECT RATIO]

Lifestyle photo - Natural, contextual

Lifestyle photography, natural lighting, [warm/cool] tones.

[SCENE DESCRIPTION]

Authentic feel, shallow depth of field. [ASPECT RATIO]

Scene / Environment - Places, spaces

[Photorealistic/Cinematic] scene, [LIGHTING DESCRIPTION], [MOOD].

[SCENE DESCRIPTION]

High detail, professional photography quality. [ASPECT RATIO]

Design & Branding

Logo concept - Clean, symbolic

Minimalist logo design on white background. Clean vector style.

[BRAND/CONCEPT DESCRIPTION]

Simple, memorable, works at small sizes. NO text. Square 1:1.

UI mockup - Interface concepts

Clean UI design mockup, modern interface, [light/dark] theme.

[SCREEN/COMPONENT DESCRIPTION]

Style: Modern SaaS aesthetic, clean typography, subtle shadows. [ASPECT RATIO]

Pattern / Texture - Repeating visuals

Seamless [PATTERN TYPE] pattern, [USER_COLOR] palette.

[PATTERN DESCRIPTION]

Tileable, consistent spacing, high quality. Square 1:1.

Generation

Single image

python3 scripts/generate.py "<full_prompt>" -o ~/Desktop/<filename>.png -m <model> --open

• -m selects the model (default: nano-banana-pro)
• --open opens the image in the default viewer after generation
• Use descriptive kebab-case filenames (e.g., laptop-notifications.png)
• If user doesn't like the result, tweak the prompt and regenerate

Batch generation (same style)

When user needs multiple images in a consistent style (e.g., blog series, icon set, product shots):

1. Define the style prompt once (colors, aesthetic, constraints)
2. Ask user for a list of concepts/subjects
3. Generate each image reusing the same style prefix, only changing the concept
4. Save all to the same folder with numbered or descriptive names

Example batch flow:

# Same style, different concepts
python3 scripts/generate.py "<STYLE PREFIX> A laptop with sparkles. <STYLE SUFFIX>" -o ~/Desktop/batch/01-laptop.png -m nano-banana-pro
python3 scripts/generate.py "<STYLE PREFIX> A rocket launching. <STYLE SUFFIX>" -o ~/Desktop/batch/02-rocket.png -m nano-banana-pro
python3 scripts/generate.py "<STYLE PREFIX> A lightbulb moment. <STYLE SUFFIX>" -o ~/Desktop/batch/03-lightbulb.png -m nano-banana-pro

Confirm total cost before starting: "This batch of N images will cost approximately $X.XX total. Proceed?"

Tips for Better Results

• Keep concepts focused: describe the main subject clearly
• Specify what you DON'T want (no text, no watermarks, no blurry elements)
• Include aspect ratio (1:1 for social, 16:9 for headers, 9:16 for stories)
• Add "Ultra high quality" or "High detail" for sharper output
• For illustrations: reference styles (Notion, Slack, Linear, Dribbble)
• For photos: describe lighting, angle, and mood
• For logos/design: emphasize simplicity and scalability
• If result isn't right, iterate: adjust the prompt and regenerate