Use CometAPI to call gemini-2.5-flash-image-preview for image generation

This document demonstrates how to use CometAPI to generate images with Google’s Gemini series image model gemini-2.5-flash-image-preview, covering three common invocation methods:

OpenAI-compatible Chat interface (text-to-image)

Gemini official generateContent text-to-image interface

Gemini official generateContent image-to-image interface (input and output both Base64)

Notes:

Replace sk-xxxx in the examples with your CometAPI key. For security, do not expose the key in clients or public repositories.

CometAPI Authorization typically uses the key value directly, i.e., Authorization: sk-xxxx.

The returned image is usually provided as Base64-encoded inline_data; you need to decode it on the client and save it as a file.

Basic information:

Base URL: https://api.CometAPI.com

Model name: gemini-2.5-flash-image-preview/gemini-2.5-flash-image

I. OpenAI Chat compatible interface (text-to-image)

Call gemini-2.5-flash-image-preview via the OpenAI-style /v1/chat/completions endpoint. It is recommended to explicitly set model to gemini-2.5-flash-image-preview or gemini-2.5-flash-image.

Example (Windows shell, using ^ for line continuation):

Notes:

stream must be true; the response will be returned as a stream;

The response structure is wrapped by CometAPI for OpenAI compatibility. The response includes a Base64 image; decode and save it on the client as needed.

Prompt tips:

You can specify style, lighting, colors, lens, resolution, etc., in the prompt, for example: “cartoon style, soft color palette, 4K square image”.

II. Gemini official generateContent text-to-image

Use the Gemini official-style generateContent endpoint for text-to-image. Put the text prompt in contents.parts[].text.

Example:

Response highlights:

Image data is usually included in response.candidates[0].content.parts:

Text description:

{ \"text\": \"...\" }

Image data:

{ \"inline_data\": { \"mime_type\": \"image/png\", \"data\": \"<base64>\" } }

Decode the data field (Base64 string) and save it as a file with the corresponding extension.

III. Gemini official generateContent image-to-image (input/output both Base64)

This endpoint supports image-to-image: upload an input image (Base64) and receive a modified new image (also in Base64).

Example:

Notes:

Convert your source image file to a Base64 string and place it in inline_data.data (do not include prefixes like data:image/jpeg;base64,).

The output is also in candidates[0].content.parts and includes:

An optional text part (explanation or hint)

An image part inline_data (where data is the Base64 of the output image)

How to extract and save Base64 images from the response

Using a Gemini-style response as an example, the pseudo-structure is as follows (illustrative only):

{
  "candidates": [
    {
      "content": {
        "parts": [
          { "text": "..." },
          {
            "inline_data": {
              "mime_type": "image/png",
              "data": "<base64-string>"
            }
          }
        ]
      }
    }
  ]
}

Extract inline_data.data and choose the file extension based on its mime_type (e.g., .png / .jpg / .webp).

Decode the Base64 on your client and save it as a binary file.

Python example:

import base64

b64 = "<base64-string>"
with open("output.png", "wb") as f:
    f.write(base64.b64decode(b64))

Node.js example:

FAQs and recommendations

Authorization header format

Use Authorization: sk-xxxx. Refer to the CometAPI documentation as the source of truth.

Streaming responses

Set "stream": true for the OpenAI-compatible interface to return SSE. If you use curl to observe the stream, ensure your client correctly handles data: lines until [DONE].

Prompt optimization

Specifying style keywords (e.g., “cyberpunk, film texture, low contrast, high saturation”), aspect ratio (square/landscape/portrait), subject, background, lighting, and level of detail helps improve results.

Base64 notes

Do not include prefixes like data:image/png;base64, in the data field; keep only the raw Base64 payload.

Troubleshooting

4xx usually indicates request parameter or authentication issues (check the key, model name, JSON format); 5xx indicates server-side issues (try again later or contact support).

The above covers the three invocation methods and key points for using gemini-2.5-flash-image-preview via CometAPI. Choose the interface style that suits your use case, and handle Base64 image persistence and display on the client.
Official API documentation reference link: https://ai.google.dev/gemini-api/docs/image-generation?hl=zh-cn#gemini-image-editing

Guide to calling gemini-2.5-flash-image

Use CometAPI to call gemini-2.5-flash-image-preview for image generation#

I. OpenAI Chat compatible interface (text-to-image)#

II. Gemini official generateContent text-to-image#

III. Gemini official generateContent image-to-image (input/output both Base64)#

How to extract and save Base64 images from the response#

FAQs and recommendations#

Use CometAPI to call gemini-2.5-flash-image-preview for image generation

I. OpenAI Chat compatible interface (text-to-image)

II. Gemini official generateContent text-to-image

III. Gemini official generateContent image-to-image (input/output both Base64)

How to extract and save Base64 images from the response

FAQs and recommendations