Nano Banana - AI Image Editing Prompt & Guides

Overview & When to Use

API integration unlocks automated, repeatable, and scalable Nano Banana workflows for web apps, creative tools, and production pipelines. Use it when you need consistent edits at scale, user-facing features, or server-side control.

Batch Processing

Automate thousands of edits for catalogs and archives.

Creative Apps

Embed Nano Banana into editors with user prompts.

Consistency

Enforce preservation rules and audit trails.

Choose Your Integration Path

Dev Friendly

Google AI Studio

Best for prototyping, internal tools, and small-to-medium workloads

• Quick setup, UI playground, generous free tier
• API keys via AI Studio; simple client SDKs
• Ideal for prompt development and testing

Production Ready

Vertex AI

Enterprise-grade platform with governance and scale

• IAM, VPC-SC, CMEK, audit logs
• Quotas, monitoring, regional control
• Suitable for regulated environments

Authentication & Setup

AI Studio (API Key)

Create an API key in AI Studio
Store it in an environment variable (never commit keys)
Use client SDKs to call Gemini Native Image

Vertex AI (Service Account)

Create a service account with minimal roles
Authenticate via ADC or JSON key (server-side only)
Call the Vertex AI Images API with regional endpoints

Security Tip

Use per-environment keys, rotate regularly, and proxy calls from your server to avoid exposing credentials in the browser.

Request Format & Media Inputs

Send your instruction and media. Keep prompts explicit about what to change and what to preserve.

{
  "model": "gemini-2.5-flash",
  "input": [
    "Change the background to a sunset beach while keeping the person identical",
    { "inlineData": { "data": "<base64 image>", "mimeType": "image/jpeg" } }
  ]
}

Supported Inputs

JPEG/PNG images; URLs or inline base64; multiple references allowed.

Response

Image bytes or URL depending on platform and SDK.

Quickstart Code (Node & Python)

Node.js (AI Studio SDK)

Example calling Gemini 2.5 Flash with an image and instruction

import { GoogleGenerativeAI } from '@google/generative-ai'

const genAI = new GoogleGenerativeAI(process.env.GOOGLE_API_KEY!)
const model = genAI.getGenerativeModel({ model: 'gemini-2.5-flash' })

export async function editImage(imageBase64: string, prompt: string) {
  const result = await model.generateContent([
    prompt,
    { inlineData: { data: imageBase64, mimeType: 'image/jpeg' } },
  ])
  return result.response
}

Python (Vertex AI)

Server-side example using service account auth

from google.cloud import aiplatform
from vertexai.preview.vision_models import ImageGenerationModel

def edit_image(image_bytes: bytes, prompt: str):
    aiplatform.init(project='YOUR_PROJECT', location='us-central1')
    model = ImageGenerationModel.from_pretrained('imagegeneration@002')
    result = model.edit_image(
        prompt=prompt,
        image=image_bytes,
        mime_type='image/jpeg',
    )
    return result

cURL (REST)

Useful for smoke tests and automation

curl -X POST   -H "Authorization: Bearer $ACCESS_TOKEN"   -H "Content-Type: application/json"   https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent   -d '{
    "contents": [{
      "parts": [
        {"text": "Change background to a sunset beach; keep the person identical"},
        {"inline_data": {"mime_type": "image/jpeg", "data": "<base64>"}}
      ]
    }]
  }'

Errors, Retries & Timeouts

Common Errors

401 Unauthorized: invalid or missing credentials

429 Too Many Requests: back off and retry with jitter

413 Payload Too Large: compress or stream uploads

Retry Strategy

Use exponential backoff (e.g., 200ms → 400ms → 800ms, cap 5 retries) and client-side timeouts (e.g., 30–60s) for long edits.

Security & Production Checklist

Security

• Never expose API keys in the browser
• Rotate credentials and scope least privilege
• Validate file types and enforce size limits
• Log request IDs, not raw media

Operations

• Use regional endpoints close to users
• Cache non-personal results where allowed
• Monitor quotas and error rates
• Add fallbacks for partial outages

Ready for Production

Confirm SLOs, alerts, and data handling policies before launch.