Gemini Chat Format

Call Nano Banana image models via Gemini's native generateContent endpoint: request fields, aspect ratio and size config, chain-of-thought output, examples.

Gemini Content Generation (Gemini Native Format)

This endpoint calls Gemini native multimodal models (such as the Nano Banana series). It supports text generation, image generation, and returning the model's chain-of-thought process.

📌 Basics

Endpoint: POST /v1beta/models/{model}:generateContent/
Path parameters:
- model: model name, e.g. gemini-3-pro-image-preview.
Authentication: Bearer Token
Content type: application/json

📥 Request Parameters (Request Body)

Parameter	Type	Required	Description
contents	`array`	Yes	List of conversation contents.
└─ role	`string`	No	Role (e.g. `user`).
└─ parts	`array`	Yes	List of message parts.
└─ └─ text	`string`	Yes	Prompt or text input.
generationConfig	`object`	Yes	Generation configuration object.
└─ responseModalities	`array`	Yes	Response modalities. Allowed values: `TEXT`, `IMAGE`.
└─ thinkingConfig	`object`	No	Thinking configuration.
└─ └─ includeThoughts	`boolean`	No	Whether to return the model's reasoning (chain of thought) during generation.
└─ imageConfig	`object`	Yes	Image generation configuration.
└─ └─ aspectRatio	`string`	Yes	Aspect ratio, e.g. `16:9`, `1:1`, `4:3`.
└─ └─ imageSize	`string`	Yes	Image quality/size, e.g. `4K`, `1024x1024`.

📤 Response

Status code: 200 OK

Field	Type	Description
candidates	`array`	List of generated candidates.
└─ content	`object`	Contains the generated role and the `parts` (may include text or image data).
└─ finishReason	`string`	Reason generation stopped (e.g. `STOP`).
└─ safetyRatings	`array`	Safety evaluation ratings.
usageMetadata	`object`	Token usage metadata (Prompt, Candidates, Total).

📝 Request Example (Generate an Image)

{
  "contents": [
    {
      "role": "user",
      "parts": [
        {
          "text": "draw a futuristic city at sunset"
        }
      ]
    }
  ],
  "generationConfig": {
    "responseModalities": ["TEXT", "IMAGE"],
    "imageConfig": {
      "aspectRatio": "16:9",
      "imageSize": "4K"
    },
    "thinkingConfig": {
      "includeThoughts": true
    }
  }
}

💡 Developer Tips

Chain-of-thought output: By setting includeThoughts: true, you can see how the model interprets your request and "plans" before drawing — very useful when debugging complex prompts.
Multimodal responses: When responseModalities includes both TEXT and IMAGE, the model first returns descriptive text and then generates the corresponding image.
Model version: Make sure the {model} parameter in the path matches the API permissions you currently have.