Skip to main content
Gemini 3 Pro Image, also known as Nano Banana Pro, is Google’s professional image generation and editing model for complex layouts, text-sensitive graphics, high-fidelity product concepts, and high-resolution deliverables. Use gemini-3-pro-image as the API model ID. The model supports 1K, 2K, and 4K output through the same Gemini request structure used by the rest of the Nano Banana family. Pro is a workload choice, not a universal quality upgrade: teams should compare it with Gemini 3.1 Flash Image on the assets that matter to their acceptance process.

​
Gemini 3 Pro Image capabilities and best-fit work

Google positions the upstream model for professional design, high-fidelity mockups, text rendering, and factual visualizations. Refer to the Gemini 3 Pro Image model page for upstream capabilities and lifecycle information.

​
Selection criteria

​
Compare total production cost

Start with gemini-3.1-flash-image for broad production traffic. Move representative, difficult assets to Pro and compare accepted-output rate, retry count, manual correction time, and final delivery value. The premium makes sense only when it reduces a more expensive part of the workflow.

​
Match the required canvas

Pro supports ten standard aspect ratios and 1K, 2K, or 4K output. It does not expose Flash Image’s 1:4, 1:8, 4:1, or 8:1 ratios. A project that needs an extreme banner or vertical canvas may require Flash even when the visual task is complex.

​
Use the right model for iteration

Drafting every variation at 4K increases transfer and storage cost without improving the decision process. Iterate at 1K or 2K, then request 4K only for an approved composition.

​
High-resolution response handling

Image bytes are returned as Base64 inside the JSON response, so a 4K request can produce a substantially larger payload than a 1K request. Production clients should:
  • allow sufficient response size and timeout at the HTTP client and reverse proxy;
  • avoid writing inlineData.data to ordinary logs or error trackers;
  • decode and move completed images to object storage promptly;
  • use asynchronous tasks when a connection cannot remain open safely;
  • retain responseId, modelVersion, task state, and the application’s storage URL for diagnostics.
Do not hard-code JPEG or PNG. Iterate through candidates[].content.parts[] and use the returned inlineData.mimeType.

​
Supported API pattern

The parameters page contains exact fields and enums. Broader upstream features such as multi-image composition or search grounding are not implied by the model name.

​
Pro versus Flash Image

Use Flash Image for everyday generation, extended aspect ratios, and mixed-resolution traffic. Evaluate Pro when complex typography, professional layouts, or high-value product assets repeatedly fail the Flash acceptance bar. Use the current pricing table together with correction time and output acceptance—not request price in isolation. For load planning, read traffic capacity and availability. Gateway RPM and the availability objective are platform-level boundaries; they are not a Pro image completion-rate or latency SLA.

​
Start with a controlled request

Synchronous generation calls POST /v1beta/models/gemini-3-pro-image:generateContent. Use /v1/images/tasks with action=generate|edit for long-running or queue-based work, then retrieve the task until completed or failed.

Configure Pro parameters

Set the output tier, aspect ratio, response modalities, and edit input.

Run Pro examples

Generate 4K output, submit tasks, edit a source image, and save the result.
Last modified on September 3, 2026