Skip to main content
For Developers: Using Nano Banana Pro via the Gemini API and Vertex AI

For Developers: Using Nano Banana Pro via the Gemini API and Vertex AI

Nano Banana Pro is not just a consumer-facing image generator—it is also a powerful tool for developers building applications that require high-quality, controllable AI image generation. Through Google Cloud’s Gemini API and Vertex AI, developers can integrate Nano Banana Pro directly into products and workflows without hosting or training any models themselves.

This guide explains how developers can access Nano Banana Pro, how the API works, and best practices for building reliable, safe, and scalable applications using Google’s infrastructure.

Getting Started: Developer Access to Nano Banana Pro

Nano Banana Pro is available to developers via Vertex AI under the model name Gemini 3 Pro Image (Preview). This is the same underlying model that powers Nano Banana Pro in Google’s consumer products, exposed through a cloud API for programmatic use.

Prerequisites

To get started, developers need:

  • A Google Cloud account

  • A project with Vertex AI enabled

  • Access to the Generative AI (Gemini) API

  • Appropriate IAM permissions

As of its preview release, the model is broadly accessible through Vertex AI, though preview features may require explicit API enablement depending on the project configuration.

Model Identification and Variants

In the API, Nano Banana Pro appears as a Gemini 3 Pro Image model, often referenced with identifiers similar to:

  • gemini-3-pro-image-preview

This model differs from the earlier Nano Banana (Gemini 2.5 Flash Image) in that it uses a “Thinking” architecture, prioritising reasoning quality, text accuracy, and complex instruction-following over raw speed.

API Usage Basics

Input Modalities

The Gemini API supports multi-modal inputs, allowing developers to combine:

  • Text prompts (instructions, descriptions, constraints)

  • Image inputs (for editing, reference, or style guidance)

Developers can supply up to 14 reference images in a single request. These can be used for:

  • Style transfer

  • Product or character consistency

  • Scene composition

  • Image editing (inpainting and outpainting)

Images are typically provided as base64-encoded data or via supported URIs, depending on the SDK or REST interface used.

Prompt and Parameter Configuration

Key parameters developers can control include:

  • Aspect ratio (e.g. 1:1, 16:9, 4:5)

  • Output resolution (commonly 1K, 2K, or up to 4K)

  • Response modalities, allowing the API to return:

    • Image data only

    • Image data plus accompanying text (such as captions or explanations)

These options make it possible to tailor outputs precisely for different application contexts.

Output Format and Watermarking

The API returns generated images as binary data or base64-encoded strings, typically in PNG or JPEG format.

All images generated by Nano Banana Pro automatically include an invisible SynthID watermark. This requires no additional configuration by developers and ensures traceability of AI-generated content without altering image appearance.

Developers can store, display, or further process the images as needed within their applications.

Best Practices for Integration

Latency and Performance

Generating high-resolution images may take several seconds per request. For production systems with predictable or high traffic, Google offers:

  • Pay-as-you-go usage for low to moderate volumes

  • Provisioned Throughput for enterprises needing guaranteed performance and consistent latency

Choosing the right model depends on whether the application prioritises speed or output quality.

Working with Multiple Reference Images

When using multiple image inputs:

  • Be explicit in the prompt about how each image should be used

  • For character consistency, include multiple images of the same subject

  • For product or branding use cases, provide logos or product images as references

Clear prompt instructions help the model correctly interpret how to combine inputs.

Prompt Engineering in API Contexts

Unlike chat interfaces, API calls must include all relevant context upfront. Developers should:

  • Inject structured data directly into prompts (e.g. product names, prices, feature lists)

  • Programmatically generate prompts using templates and variables

  • Be explicit about text placement when generating posters, infographics, or ads

This enables large-scale automated image generation without manual intervention.

Iterative Generation Loops

Developers can replicate conversational refinement by:

  1. Generating an initial image

  2. Collecting user feedback

  3. Sending a follow-up API request with refined instructions and the previous image as input

This pattern works well for design tools, creative platforms, and user-driven customization workflows.

Safety, Moderation, and Responsible Use

The Gemini API includes built-in safety systems that may flag or block disallowed content. Developers should:

  • Handle safety responses gracefully in their applications

  • Provide clear user feedback when prompts are rejected

  • Consider adding client-side moderation for user-generated prompts

For enterprise deployments, Vertex AI offers enhanced safety controls to help ensure compliance with organisational and regulatory requirements.

Quotas, Pricing, and Cost Management

Image generation via the Gemini API incurs usage-based costs. While exact pricing depends on region and configuration, developers can generally choose between:

  • On-demand, pay-as-you-go billing

  • Committed or provisioned capacity for high-volume workloads

Monitoring quotas and usage is essential, particularly during preview phases where limits may apply.

Example Applications Developers Can Build

Nano Banana Pro enables a wide range of developer use cases, including:

  • Design platforms for posters, social media graphics, or branded visuals

  • E-commerce tools that generate product imagery in multiple scenes

  • Game development utilities for concept art and asset generation

  • CMS plugins that auto-generate blog illustrations or headers

  • Creative editing apps that turn sketches or photos into polished visuals

The ability to combine text and image inputs unlocks advanced workflows not possible with text-only or image-only models.

Enterprise Deployment Considerations

For businesses deploying Nano Banana Pro in production:

  • Vertex AI provides enterprise-grade reliability and monitoring

  • Customer data is not used for model training by default

  • Google has announced plans for copyright indemnification when the model reaches general availability

These assurances make Nano Banana Pro suitable for professional and commercial environments where data governance and legal clarity matter.

Conclusion

By exposing Nano Banana Pro through the Gemini API and Vertex AI, Google has made one of its most advanced image generation models directly accessible to developers. With minimal setup, teams can integrate high-quality, controllable AI image generation into applications ranging from creative tools to enterprise platforms.

As the model moves toward full general availability, developers can expect continued improvements, stronger guarantees, and deeper integration across Google’s developer ecosystem—making Nano Banana Pro a compelling foundation for the next generation of AI-powered visual applications.