Google: Nano Banana 2 (Gemini 3.1 Flash Image)

Google: Nano Banana 2 (Gemini 3.1 Flash Image) is an image generation and editing model from Google, identified on api14 as google/gemini-3.1-flash-image. It sits in the image category with text as a secondary category, and it is the successor to Google's earlier Nano Banana image model. The provider release date is 2026-06-18. Google positions it as delivering Pro-level visual quality at Flash speed, which tells you the intent: high output quality without the slower turnaround usually associated with larger image models.

The model accepts image and text inputs and can return image and text outputs, so it handles both generation from a prompt and editing of images you supply. Its context window is 131,072 tokens, and the maximum output is 32,768 tokens. Image generation is configurable: resolution can be set to 512, 1K, 2K or 4K, and aspect ratio supports 1:1, 1:4, 1:8, 2:3, 3:2, 3:4, 4:1, 4:3, 4:5, 5:4, 8:1, 9:16, 16:9 and 21:9. The n parameter accepts a minimum of 1 and a maximum of 1, so each request produces a single image. Input references accept from 0 to 14 images, which is what makes multi-image editing and composition practical.

Those capabilities point to a fairly wide set of uses. Text to image generation covers illustration, concept art, product mockups and marketing assets, with the aspect ratio list handling everything from square social posts to wide banners and tall mobile formats. Because the model takes image input alongside text, it also fits editing workflows: retouching, restyling, background changes, and combining elements from several source images, since up to 14 references can be passed in. The 4K resolution option suits work that needs to hold up when printed or displayed large, while 512 keeps drafts cheap to iterate on. The text output side means the model can return explanations or descriptions alongside the image it produces.

You call it through api14's OpenAI-compatible API using the model id google/gemini-3.1-flash-image, so existing OpenAI client libraries work with only the base URL and model name changed. The supported request parameters are include_reasoning, max_tokens, reasoning, reasoning_effort, response_format, seed, structured_outputs, temperature and top_p. Seed is useful when you want reproducible image output across runs, and temperature and top_p control how much variation you get between generations. The reasoning related parameters let you trade latency for more deliberate handling of complex prompts.

A few constraints are worth knowing before you commit. The n parameter is fixed at 1, so you cannot request multiple images in a single call; if you need variations, issue separate requests, optionally with different seeds. Input references cap at 14, so very large reference sets need to be narrowed down first. If your task is purely text generation with no image component, a dedicated text model will usually be a better fit, and if you need heavier reasoning over long documents, the image focus here is not the point. For image generation and editing within these bounds, it covers the common cases directly.