Gemini Nano Banana 2.1

Gemini Nano Banana 2.1 is an image-generation and conversational-editing model focused on visual quality, text rendering, reference consistency, and wide-format images. Use gemini-nano-banana-2.1; do not substitute a chat-only Gemini model.

Model Information

Field	Description
`model`	gemini-nano-banana-2.1
`input limit`	131,072 tokens
`output limit`	32,768 tokens
`input / output`	Text, images, video, and PDF input; image and text output. No native audio output.
`image output`	1K (default), 2K, or 4K. The 512 tier of Nano Banana 2 does not apply.

These are model specifications. Route availability, billing, file access, and request-size limits depend on the selected channel. Image output uses the native Gemini endpoints below. POST https://api.mixroute.ai/v1/models/gemini-nano-banana-2.1:generateContent POST https://api.mixroute.ai/v1/models/gemini-nano-banana-2.1:streamGenerateContent?alt=sse

Generate an Image

curl --request POST "https://api.mixroute.ai/v1/models/gemini-nano-banana-2.1:generateContent" \
  --header "Authorization: Bearer $MIXROUTE_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
  "contents": [
    {
      "role": "user",
      "parts": [
        {
          "text": "Generate a clean product photograph of a red ceramic mug on a plain white background. No text."
        }
      ]
    }
  ],
  "generationConfig": {
    "responseModalities": [
      "TEXT",
      "IMAGE"
    ],
    "imageConfig": {
      "aspectRatio": "1:1",
      "imageSize": "1K"
    },
    "thinkingConfig": {
      "thinkingLevel": "MINIMAL"
    }
  }
}'

Use the native generationConfig.imageConfig fields for size and aspect ratio on this endpoint. Do not add metadata, asset, an OpenAI size/n field, or a model field inside the request body. The model is selected in the URL.

Request Fields

Field	Description
`contents`	Required user/model turns with parts.
`contents[].parts[].text`	Prompt or editing instruction.
`contents[].parts[].inlineData`	Input image object: mimeType and raw Base64 data; no data-URI prefix.
`contents[].parts[].fileData`	Native fileUri/mimeType references; resources must be accessible to the route account.
`generationConfig.responseModalities`	Use [“TEXT”,“IMAGE”] or [“IMAGE”] for image output.
`generationConfig.imageConfig.aspectRatio`	Native aspect-ratio string, such as 1:1 or 16:9.
`generationConfig.imageConfig.imageSize`	1K, 2K, or 4K, case-sensitive; default 1K.
`generationConfig.thinkingConfig.thinkingLevel`	MINIMAL, MEDIUM (default), or HIGH.
`generationConfig.thinkingConfig.includeThoughts`	Optional thought-summary parts; skip thought=true parts when saving final images.
`generationConfig.maxOutputTokens`	Combined image/text output budget, up to 32,768; too small a budget can prevent a complete image.
`tools`	Native Google Search grounding where enabled by the route. Function calling is not supported by this image model.
`safetySettings`	Native category/threshold settings, subject to provider and account policy.

Limits

Aspect ratios include 1:1, 1:4, 4:1, 1:8, 8:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, and 21:9.
Up to 14 reference images, with model guidance for up to four characters and ten objects. These are input limits, not a guarantee of perfect consistency.
Vertex inline-image inputs are limited to 7 MB per image. PNG, JPEG, WebP, HEIC, and HEIF are supported; other request limits can be stricter.
Omit the unsupported sampling/randomness fields temperature, topP, topK, seed, and logprobs.
Audio generation, arbitrary function calling, and structured JSON output are not supported.

Editing and Multi-turn Use

{
  "contents": [
    {
      "role": "user",
      "parts": [
        {
          "text": "Change the mug color to blue while preserving its shape and the white background."
        },
        {
          "inlineData": {
            "mimeType": "image/png",
            "data": "BASE64_IMAGE_DATA"
          }
        }
      ]
    }
  ],
  "generationConfig": {
    "responseModalities": [
      "TEXT",
      "IMAGE"
    ],
    "imageConfig": {
      "aspectRatio": "1:1",
      "imageSize": "1K"
    },
    "thinkingConfig": {
      "thinkingLevel": "MINIMAL"
    }
  }
}

Replace BASE64_IMAGE_DATA with the source bytes encoded as Base64. For conversational edits, retain the complete prior candidate.content as a model turn, followed by the next user turn. Preserve all thoughtSignature values exactly; do not replace the image history with text alone.

Read the Output

Process candidates[].content.parts[] by type. Save non-thought inlineData parts using their MIME type and decoded Base64 bytes, and read text parts separately. A response without an image requires checking promptFeedback and finishReason; HTTP success alone does not imply image output. Streaming uses streamGenerateContent?alt=sse with the same body, not a JSON stream flag. Parse SSE data events and handle image/text parts in each JSON event. Image generation and editing | Image streaming

Home

Model Introduction

Model Information

Generate an Image

Request Fields

Limits

Editing and Multi-turn Use

Read the Output

​Model Information

​Generate an Image

​Request Fields

​Limits

​Editing and Multi-turn Use

​Read the Output

Model Information

Generate an Image

Request Fields

Limits

Editing and Multi-turn Use

Read the Output