Google: Nano Banana Pro (Gemini 3 Pro Image)

Google: Nano Banana Pro (Gemini 3 Pro Image) is an image generation and editing model from Google, built on Gemini 3 Pro. It is the successor to the original Nano Banana, and Google describes it as its most advanced image-generation and editing model, with improved multimodal reasoning and real-world grounding compared with the earlier version. It sits in the image category while also handling text, so it is a multimodal model rather than a pure text or pure image system.

The model accepts image and text inputs and can produce both image and text outputs. It has a context window of 131,072 tokens and a maximum output of 32,768 tokens. Image generation is configurable: resolution can be set to 1K, 2K or 4K, and aspect ratio can be set to 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9 or 21:9. The n parameter accepts a value from 1 to 1, so each request returns a single image, and input_references accepts from 0 to 14 images, which is what allows reference-driven editing and composition.

Those capabilities point to a few practical patterns. You can generate an image from a text prompt at a chosen resolution and aspect ratio, which suits anything from square social assets to wide 21:9 banners or tall 9:16 formats. Because the model takes image input alongside text, you can also use it for editing and transformation work, passing up to 14 reference images to guide style, subject or composition. The text output side means a single call can return commentary or structured content together with the generated image, and the reasoning and structured output parameters support workflows where the model needs to plan before producing the result.

Through api14 you call the model with the id "google/gemini-3-pro-image" using the OpenAI-compatible API, so existing OpenAI SDKs and clients work with a change of base URL and model id. The supported request parameters are include_reasoning, max_tokens, reasoning, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools and top_p. Image-specific options such as resolution, aspect_ratio, n and input_references are set alongside those parameters. The seed parameter is useful when you want reproducible generations, and tools and tool_choice let you combine image generation with function calling in an agent loop.

Two constraints are worth noting before you commit. Each request returns only one image, since n is fixed between 1 and 1, so batches require multiple calls. Input references are capped at 14 images per request. If your task is text-only reasoning or chat without any image output, a dedicated text model will usually be the simpler and more appropriate choice, and this model is best reserved for work where image generation or image editing is actually part of the output.