Nano Banana family

Google’s Gemini image models, five in total. We recommend calling them through the Gemini native API, which gives precise control over aspect ratio and resolution; the OpenAI-compatible API also works, but supports only some parameters.

Common tasks

Set aspect ratio and resolution Landscape, portrait and 4K images, and the values each model supports.Editing & multi-image Upload images and edit them with a text instruction.Troubleshooting errors 400 invalid argument, wrong aspect ratio, no image returned, 410.

Choosing a model

Key pointNano Banana 2 is recommended for most cases; choose Pro for 4K output or complex compositions, and 2 Lite for generating small images in bulk.

Nano Banana 2.1Latest

Three tiers: 1K, 2K, 4K
Aspect ratios from 1:8 to 8:1; 2K and 4K measured at about 2 minutes

gemini-nano-banana-2.1

Nano Banana 2Everyday pick

Four resolutions from 512 to 4K
Widest choice of aspect ratios, including 1:8 and 8:1

gemini-3.1-flash-image

Nano Banana ProQuality first

Three tiers: 1K, 2K, 4K
Reasons before generating, so it takes longer

gemini-3-pro-image

Nano Banana 2 LiteSpeed first

1K only
Measured at about 5 seconds per image; outputs JPEG

gemini-3.1-flash-lite-image

Nano BananaPrevious gen

Fixed at about 1K; resolution cannot be set
Fine for existing projects

gemini-2.5-flash-image

Model IDs (copy them exactly)

Nano Banana 2.1google/gemini-nano-banana-2.1Latest

Nano Banana 2google/gemini-3.1-flash-imageEveryday pick

Nano Banana Progoogle/gemini-3-pro-image4K, complex compositions

Nano Banana 2 Litegoogle/gemini-3.1-flash-lite-image1K only

Nano Bananagoogle/gemini-2.5-flash-imagePrevious gen

For live prices, see the model pages: Nano Banana 2 , Nano Banana Pro . The actual charge follows usageMetadata in the response.

Choosing an endpoint

Key pointUse the Gemini native API when you need to control aspect ratio and resolution.

	Gemini native API (recommended)	OpenAI-compatible API
Endpoint	`POST /gemini/v1beta/models/<model ID>:generateContent`	`POST /v1/images/generations`, `/v1/images/edits`
Aspect ratio	`imageConfig.aspectRatio`	Top-level `aspect_ratio`
Resolution	`imageConfig.imageSize`: 512, 1K, 2K, 4K	Not supported; sending `image_size` returns 400
`size: "WIDTHxHEIGHT"`	Not applicable	Not supported; set the ratio with `aspect_ratio`
`n` (multiple images)	Not applicable	Not supported; one image per request
Editing	Put `inlineData` in `parts`	Upload files to `/v1/images/edits`
Reference images `input_images`	Not applicable	Not supported; use `/v1/images/edits` for editing
Returned in	`candidates[0].content.parts[].inlineData.data`	`data[0].b64_json`

Latency and timeouts

Key pointSet the client timeout to 300 seconds or more.

The API is synchronous. We measured about 5 to 20 seconds at 1K and about 30 seconds at 4K; Nano Banana 2.1 at 2K and 4K takes about 2 minutes. It takes longer at peak times. Pro reasons before generating the image (the response includes thoughtsTokenCount), so it is slower than Nano Banana 2 at the same resolution. The common 60-second default timeout does not cover 4K or peak-time requests; if the client disconnects, the request is still billed.

Response structure

200 OK (native API, data truncated)


{
  "candidates": [{
    "content": {
      "parts": [{ "inlineData": { "mimeType": "image/png", "data": "iVBORw0KGgo…" } }],
      "role": "model"
    },
    "finishReason": "STOP"
  }],
  "modelVersion": "google/gemini-3-pro-image",
  "usageMetadata": {
    "promptTokenCount": 7,
    "candidatesTokenCount": 2000,
    "candidatesTokensDetails": [{ "modality": "IMAGE", "tokenCount": 2000 }],
    "thoughtsTokenCount": 211,
    "totalTokenCount": 2218
  }
}

The image is in the parts item that contains inlineData, not necessarily the first item; text may come before it. Iterate over parts to find it.
The IMAGE entry in candidatesTokensDetails gives the number of output image tokens. We measured 1,120 at 1K and 2,000 for Pro at 4K.
thoughtsTokenCount appears only in Pro responses and counts the reasoning tokens used before generating the image.

Official docs

The official docs describe behavior when calling Google directly; when calling through OfoxAI, the measured results on this page take precedence.

FAQ

How do I generate a 4K image with Nano Banana Pro?

Use the Gemini native API and pass imageSize: "4K" in generationConfig.imageConfig, with aspectRatio to set the ratio. In our tests a 16:9 4K output was 5504x3072, about 16 MB, and took about 33 seconds. The OpenAI-compatible API does not support imageSize; sending image_size returns 400.

Which aspect ratios does Nano Banana 2 support?

Ten common ones: 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9 and 21:9; plus four extra-wide or extra-tall ratios: 1:4, 4:1, 1:8 and 8:1. Any ratio not on the list returns 400 Request contains an invalid argument.

Does Nano Banana Pro support 1:8 or 4:1?

No. The four ratios 1:4, 4:1, 1:8 and 8:1 are supported only by Nano Banana 2.1, Nano Banana 2 and Nano Banana 2 Lite; use one of these three for extra-tall or extra-wide images.

Should imageSize be 4K or 4k?

We recommend uppercase 1K, 2K and 4K, matching Google. In our tests OfoxAI also accepts lowercase. The 512 tier has no K and is supported only by Nano Banana 2.

How do I set the Nano Banana aspect ratio on the OpenAI-compatible API?

Use the top-level aspect_ratio, e.g. "16:9". That API has no resolution parameter; for 2K or 4K use imageConfig.imageSize on the Gemini native API.

Can Nano Banana generate several images at once?

No, each request generates one image. For more, send several requests concurrently, keeping the rate limit in mind.

What timeout should I set for Nano Banana?

We recommend 300 seconds or more. We measured about 5 to 20 seconds at 1K, about 30 seconds at 4K, and about 2 minutes for Nano Banana 2.1 at 2K and 4K, longer at peak times. Pro reasons before generating, so it is slower than Nano Banana 2.