Gemini Nano Banana 2.1

Gemini Nano Banana 2.1 is Google’s stable, high-efficiency image generation and conversational editing model. Use the model ID gemini-nano-banana-2.1 with the native Gemini GenerateContent endpoint.

Google released gemini-nano-banana-2.1 as a generally available model on October 6, 2026. Google recommends it for new projects that previously used gemini-3.1-flash-image.

Model specifications

Property	Value
Model ID	`gemini-nano-banana-2.1`
Endpoint	`POST /v1beta/models/gemini-nano-banana-2.1:generateContent`
Input	Text, image, video, PDF
Output	Image and text
Input token limit	131,072
Output token limit	32,768
Output resolution	`1K`, `2K`, `4K`; default `1K`
Reference images	Up to 14 total; up to 4 characters and 10 objects for high-fidelity consistency
Thinking levels	`minimal`, `medium`, `high`; default `medium`

Key capabilities

Generate an image from a text prompt.
Edit an existing image through a conversational instruction.
Combine up to 14 reference images in one request.
Preserve character and object appearance across iterative edits.
Generate wide and panoramic layouts, including 1:4, 4:1, 1:8, and 8:1.
Produce text-heavy visuals such as infographics, menus, and diagrams.

All generated images contain Google’s SynthID watermark.

Request examples

Text to image
Image editing
Thinking level

Use responseModalities to request an image, then set the aspect ratio and resolution under responseFormat.image.

cURL

curl "https://www.anyfast.ai/v1beta/models/gemini-nano-banana-2.1:generateContent?key=YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "role": "user",
      "parts": [{
        "text": "Create a clean 16:9 product poster for a translucent blue mechanical keyboard on a dark studio background. Include the headline: TYPE FASTER."
      }]
    }],
    "generationConfig": {
      "responseModalities": ["TEXT", "IMAGE"],
      "responseFormat": {
        "image": {
          "aspectRatio": "16:9",
          "imageSize": "2K"
        }
      }
    }
  }'

Add a Base64-encoded reference image as an inline_data part next to the editing instruction.

cURL

curl "https://www.anyfast.ai/v1beta/models/gemini-nano-banana-2.1:generateContent?key=YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "role": "user",
      "parts": [
        {
          "text": "Keep the product unchanged, replace the background with a warm wooden desk, and add soft morning light from the left."
        },
        {
          "inline_data": {
            "mime_type": "image/png",
            "data": "YOUR_BASE64_IMAGE_DATA"
          }
        }
      ]
    }],
    "generationConfig": {
      "responseModalities": ["IMAGE"],
      "responseFormat": {
        "image": {
          "aspectRatio": "4:3",
          "imageSize": "2K"
        }
      }
    }
  }'

Use thinkingConfig.thinkingLevel to balance latency and image quality. The default is medium.

cURL

curl "https://www.anyfast.ai/v1beta/models/gemini-nano-banana-2.1:generateContent?key=YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{
      "parts": [{
        "text": "Design a technically accurate exploded-view infographic of a mirrorless camera with concise English labels."
      }]
    }],
    "generationConfig": {
      "responseModalities": ["IMAGE"],
      "thinkingConfig": {
        "thinkingLevel": "high"
      },
      "responseFormat": {
        "image": {
          "aspectRatio": "4:3",
          "imageSize": "4K"
        }
      }
    }
  }'

Read the generated image

The response can contain text parts and image parts. Decode inlineData.data from Base64 and save it using the MIME type returned in inlineData.mimeType.

Python

import base64
import requests

response = requests.post(
    "https://www.anyfast.ai/v1beta/models/gemini-nano-banana-2.1:generateContent",
    params={"key": "YOUR_API_KEY"},
    json={
        "contents": [{
            "parts": [{"text": "Create a minimal isometric illustration of a green data center."}]
        }],
        "generationConfig": {
            "responseModalities": ["TEXT", "IMAGE"],
            "responseFormat": {
                "image": {"aspectRatio": "16:9", "imageSize": "1K"}
            },
        },
    },
)
response.raise_for_status()

for part in response.json()["candidates"][0]["content"]["parts"]:
    if "text" in part:
        print(part["text"])
    if "inlineData" in part:
        with open("output.png", "wb") as image_file:
            image_file.write(base64.b64decode(part["inlineData"]["data"]))

Image configuration

Field	Values	Notes
`generationConfig.responseModalities`	`IMAGE`, or `TEXT` and `IMAGE`	Defaults to text and image output.
`generationConfig.responseFormat.image.imageSize`	`1K`, `2K`, `4K`	Default: `1K`. Nano Banana 2.1 does not support `512`.
`generationConfig.responseFormat.image.aspectRatio`	`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`, `21:9`	Without an input image, the default is `1:1`; editing requests normally follow the input image unless overridden.
`generationConfig.thinkingConfig.thinkingLevel`	`minimal`, `medium`, `high`	Default: `medium`. Thinking tokens are billed even when thoughts are not returned.
`generationConfig.thinkingConfig.includeThoughts`	boolean	When `true`, thought parts may be included in the response. Do not treat them as the final image or answer.

Do not send seed, topK, topP, temperature, or logprobs. Google documents these parameters as unsupported for Nano Banana 2.1 and rejects requests that include them.

Input guidance

Put text instructions and reference images in the same parts array.
Use inline_data for Base64 content and file_data for an uploaded file URI.
Supported image MIME types include PNG, JPEG, WebP, HEIC, and HEIF.
The model accepts up to 14 reference images, but character consistency is optimized for up to 4 characters and object fidelity for up to 10 objects.
Video and PDF inputs consume the same 131,072-token context window.

For text inside an image, write the exact required wording in the prompt and describe its placement, type style, and hierarchy.