Gemini Nano Banana Prompt: Examples, Tips, and API Tutorial

A strong Gemini Nano Banana prompt describes the image you want in plain language: the subject, scene, composition, visual style, lighting, and details that matter most. For edits, provide the image and say what to change—and what to preserve.

“ Nano Banana” is Google’s name for Gemini’s native image-generation capabilities, not a single model. Google lists Nano Banana 2.1 (gemini-nano-banana-2.1) as a high-efficiency model for image generation and conversational editing, and recommends it for new projects. Nano Banana Pro (gemini-3-pro-image) is positioned for complex visual tasks. (aistudio.google.com)

A reusable Gemini image prompt template

Use this as a writing guide, not as a required JSON format or special syntax:

Create [type of image] of [subject] in [setting]. Show [action or pose]. Compose it as [framing, angle, and placement]. Use [style or medium], [lighting], and [color palette]. Include [important details or exact text]. Keep [features to preserve] consistent. Avoid [specific unwanted result].

You don’t have to fill every slot. Start with the details that would most change the image. If the first result misses something, ask for that correction directly rather than rewriting the whole prompt.

Copy-and-adapt Nano Banana prompt examples

These are original starting points, not reported test results. Adapt the bracketed details to your project.

1. Product image from text

Create a premium studio product photograph of a matte-black insulated water bottle standing upright on a pale sandstone block. Use a three-quarter view, soft side lighting, gentle shadows, and a warm neutral background. Keep the bottle centered with clear space above it for a headline. Show subtle condensation and realistic material texture. No extra objects, logos, or text.

Change: the product, background material, color palette, or camera angle. If you need a real brand mark, provide the logo as an image reference rather than relying on a text description alone.

2. Edit an existing image

Provide the image, then prompt:

Edit the attached café photo: replace the bright midday light with soft late-afternoon window light. Keep the table, cups, people, camera angle, and overall composition unchanged. Make the lighting change look natural; do not add objects or alter anyone’s face.

Change: name the edit first, then list the details that should remain unchanged. That makes the request more specific than “make it better.”

3. Poster with exact wording

Design a clean event poster for a neighborhood plant swap. Use a simple illustrated monstera leaf, a cream background, and dark-green sans-serif typography. Include this exact headline: “PLANT SWAP”. Below it, add: “Saturday, 10 AM–2 PM” and “Bring a plant, take a plant.” Keep all three lines readable, correctly spelled, and separate from the illustration.

For a design that depends on exact copy, specify the wording and its placement. Review the generated text before publishing; image generation can still make mistakes.

How to improve a prompt after the first image

Treat prompting as a short revision loop:

1. Pick the biggest miss. Is the subject wrong, the framing off, or a key detail missing?
2. Name the correction precisely. For example: “Move the camera to eye level and leave more space above the subject.”
3. Say what must stay fixed. For example: “Keep the same outfit, background, and color palette.”
4. Generate again and inspect the change. If another detail is still wrong, revise that next.

For edits, make one clear change per follow-up when you need predictable results. For new images, add concrete visual direction—such as “close-up, eye-level view” or “soft window light from the left”—instead of stacking vague quality words.

Use a prompt in Gemini or through the API

In Gemini Apps

Use Gemini’s image-generation interface to describe the image you want. For an edit, provide the image and describe what to change and what to preserve. This user-facing workflow is separate from making an API request; available controls and access depend on the Gemini interface you use. (aistudio.google.com)

With Google’s Gemini API

The following Python example follows the request-and-image-saving pattern shown in Google AI Studio’s model example. It uses gemini-3.1-flash-image (Nano Banana 2) so the model ID matches that example; check Google’s current model documentation when choosing a model for a new project. (aistudio.google.com)

Install the SDK and Pillow, and set GEMINI_API_KEY in your environment. Keep the key in a server-side environment or secret manager; don’t expose it in browser code.

pip install -U google-genai Pillow
export GEMINI_API_KEY="your_api_key"

import os
from google import genai

client = genai.Client(api_key=os.environ["GEMINI_API_KEY"])

prompt = (
    "Create a clean studio photograph of a ceramic pour-over coffee dripper "
    "on a pale stone counter, soft morning window light, three-quarter view, "
    "warm neutral palette, no text or extra objects."
)

response = client.models.generate_content(
    model="gemini-3.1-flash-image",
    contents=[prompt],
)

saved_image = False
for part in response.parts:
    if part.text is not None:
        print(part.text)
    elif part.inline_data is not None:
        image = part.as_image()
        image.save("generated_image.png")
        saved_image = True

if not saved_image:
    raise RuntimeError("The response did not include an image.")

Google’s example iterates through response parts, converts an image part with as_image(), and saves it as a PNG. The code above adds a no-image check. It has not been independently run here. For production, also handle API errors, secure credentials, and monitor usage. (aistudio.google.com)

Which Nano Banana model should you prompt?

• Nano Banana 2.1 (gemini-nano-banana-2.1): Google describes it as a high-efficiency model for image generation and conversational editing, and recommends it for new projects. (aistudio.google.com)
• Nano Banana Pro (gemini-3-pro-image): Google positions Pro for complex visual tasks, including professional-grade image generation and editing. (ai.google.dev)

If a prompt example uses a JSON wrapper or fields such as negative_prompt, don’t assume Gemini requires or accepts those as API parameters. Write visual instructions in the form supported by the interface or API you’re using. A gallery example can inspire a prompt, but it doesn’t establish official syntax or model compatibility.

Quick checklist

Before sending a Gemini Nano Banana prompt, check:

• Have I named the subject and the requested action?
• Have I described the setting and composition?
• Have I called out the lighting, style, or details that matter?
• For an edit, have I said what must not change?
• If the image contains important wording, have I supplied the exact text and checked the result?
• If I’m using the API, have I chosen the documented model ID and kept credentials server-side?

The simplest practical formula is: describe the image, specify the important visual details, then refine the one thing the output missed.

Sources

Nano Banana image generation — Google AI Studio

Nano Banana — Google AI Studio

Gemini 3 Pro image — Google AI for Developers

Get Started Nano Banana — Google Gemini Cookbook