GPT-Image-2 Image Generation Guide

Prerequisites

Before you start, prepare the following:

1. A aicentos account
2. A usable API token
3. Confirm that your current token can access gpt-image-2

Get your token here:

• Console: aicentos

Tip

If your console shows model groups or permission settings, make sure your current token includes access to gpt-image-2. If unsure, create a fresh default token and test with that first.

Open-source visual tool

If you do not want to write code first, you can use aicentos's open-source GPT-Image-2 tool to test prompts and generate images directly:

• Live demo: aicentos
• GitHub repository: aicentos

Tip

Use the tool to verify your token, prompt, and image parameters first, then migrate the same parameters into your own code or workflow.

Method 1: Generate Images via the Images API

This is the most direct approach and maps to the OpenAI-compatible image generation endpoint.

• Endpoint: https://www.aicentos.com/v1/images/generations
• Model: gpt-image-2

Python Example

import base64
from openai import OpenAI

client = OpenAI(
    api_key="sk-your-aicentos-token",
    base_url="https://www.aicentos.com/v1"
)

result = client.images.generate(
    model="gpt-image-2",
    prompt="An orange cat wearing an astronaut helmet, sitting on the moon, cinematic lighting, ultra-detailed",
    size="1024x1024"
)

image_base64 = result.data[0].b64_json

with open("gpt-image-2-output.png", "wb") as f:
    f.write(base64.b64decode(image_base64))

Node.js Example

import fs from "node:fs";
import OpenAI from "openai";

const client = new OpenAI({
  apiKey: "sk-your-aicentos-token",
  baseURL: "https://www.aicentos.com/v1",
});

const result = await client.images.generate({
  model: "gpt-image-2",
  prompt: "A futuristic city floating above the clouds, cyberpunk neon style, ultra-detailed",
  size: "1024x1024",
});

const imageBase64 = result.data[0].b64_json;
fs.writeFileSync("gpt-image-2-output.png", Buffer.from(imageBase64, "base64"));

curl Example

curl --request POST \
  --url https://www.aicentos.com/v1/images/generations \
  --header "Authorization: Bearer sk-your-aicentos-token" \
  --header "Content-Type: application/json" \
  --data '{
    "model": "gpt-image-2",
    "prompt": "A minimalist product poster on a white background, with a floating transparent glass keyboard in the center, commercial photography style",
    "size": "1024x1024"
  }'

Note

The main result from the Images API is usually b64_json, so you need to decode the Base64 string into an actual image file on your side.

Method 2: Generate Images via Chat Completions

If your workflow is already built around /v1/chat/completions, you can also call gpt-image-2 there.

• Endpoint: https://www.aicentos.com/v1/chat/completions
• Model: gpt-image-2

Python Example

from openai import OpenAI

client = OpenAI(
    api_key="sk-your-aicentos-token",
    base_url="https://www.aicentos.com/v1"
)

response = client.chat.completions.create(
    model="gpt-image-2",
    messages=[
        {
            "role": "user",
            "content": "Draw a Shiba Inu wearing sunglasses, sitting in a vintage red convertible, road trip style"
        }
    ]
)

print(response)

curl Example

curl --request POST \
  --url https://www.aicentos.com/v1/chat/completions \
  --header "Authorization: Bearer sk-your-aicentos-token" \
  --header "Content-Type: application/json" \
  --data '{
    "model": "gpt-image-2",
    "messages": [
      {
        "role": "user",
        "content": "Draw a Japanese-style cafe at sunset with warm cinematic composition"
      }
    ]
  }'

Tip

If your goal is simply to get stable image output, prefer /v1/images/generations. Use /v1/chat/completions only when you need image generation inside an existing chat-based workflow.

Common Parameters

Commonly used parameters for gpt-image-2 include:

Parameter	Meaning	Example
model	Model ID	gpt-image-2
prompt	Image prompt	a mechanical owl
size	Output size	1024x1024
n	Number of images	1
background	Background mode	transparent / white
quality	Output quality	high / medium

Note

Parameter compatibility can vary across SDKs and proxy layers. If something fails, reduce to the minimum working set first: model + prompt + size.

Using GPT-Image-2 in Cherry Studio

If you prefer generating images directly in Cherry Studio, configure it like this:

1. Add an OpenAI-Compatible Provider

Create a custom provider in Cherry Studio with:

• API Key: your aicentos token
• Base URL: https://www.aicentos.com/v1

2. Add the Model

Use this model name:

gpt-image-2

If Cherry Studio lets you define the model type, choose an image-generation type or an OpenAI Images-compatible type.

3. Start Generating

Create a new session or open the image generation panel, select gpt-image-2, and enter your prompt.

Tip

If Cherry Studio does not show image results, check these first:

• The Base URL must be https://www.aicentos.com/v1
• Your Cherry Studio version must properly support the OpenAI Images API

FAQ

Why did the request succeed but no image file appeared?

Because many responses return b64_json instead of a saved file. You need to decode that Base64 string and write it into a .png file yourself.

Which endpoint should I prefer?

• Image generation only: prefer POST /v1/images/generations
• Need image generation inside an existing chat workflow: consider POST /v1/chat/completions

What should I do if the model does not exist or I have no permission?

Check in this order:

1. Make sure the model name is gpt-image-2
2. Make sure the token comes from the aicentos Console
3. Make sure your token has access to gpt-image-2
4. Make sure the Base URL is https://www.aicentos.com/v1

Why do extra parameters sometimes fail?

Because compatibility differs across clients, SDK versions, and proxy layers. Start with the minimum working set and then add optional parameters like quality, background, and n one by one.