Call GPT Image 2

Every model takes the same request, so switching to GPT Image 2 is a change to the model field. These samples are written for its own contract and send as they are.

import Mynth from "@mynthio/sdk";

const mynth = new Mynth();

const task = await mynth.image.generate({
  model: "openai/gpt-image-2",
  prompt: "A lighthouse at dusk, 35mm film",
  size: "16:9",
});

console.log(task.urls[0]);

Specifications

What GPT Image 2 accepts

Read from the same registry the API validates against. A request outside it is refused when the task is created, before anything is charged.

Modes and inputs

Text → Image: Prompt only
Image → Image: Up to 16 images

Output

Aspect ratios: 1:1
4:3
3:4
16:9
9:16
3:2
2:3
5:4
Leave out size, or send auto, and Mynth picks one for the prompt. Any other ratio snaps to the closest of these. Sizes
4K: No. A 4k size fails the task

Prompt

Negative prompt: Not supported. Dropped before the provider call, no error

The contract as GET /models returns it

{
  "modes": {
    "txt->img": {},
    "img->img": {
      "inputs": {
        "rules": [
          {
            "type": "image",
            "max": 16
          }
        ]
      }
    }
  }
}

Pricing

What it costs

Pay per result, in USD, with no subscription. A failed image or video costs nothing. How pricing works

Rates

Image: $0.12 / image
Input image: $0.02 / input

An input is charged once for every image it goes into: two inputs on four images are eight input charges.

Estimate a request

Images

1

Input images

0

$0.121 × $0.12

Held when the task is created. Only images that succeed are charged.

Guide

Working with GPT Image 2

Notes on getting the most out of GPT Image 2.

GPT Image 2 is OpenAI's image model for production work: layouts with dense, readable text, edits that stay faithful to the source, and scenes that draw on what the model knows about the world rather than on a reference image. OpenAI released it on 21 April 2026, as the snapshot gpt-image-2-2026-04-21.

What it is good at

Typography. Posters, packaging, menus and UI mockups with text that reads, including long lines and small sizes.
Multilingual text. Text rendered in scripts other than Latin.
Edits with references. Up to 16 input images per request, in any role. Describe what should change and what must stay.
World knowledge. Recognisable places, objects and styles from the prompt alone, without an image to copy from.
Flexible framing. Eight aspect ratios from 9:16 to 16:9, each at a native resolution around 2K.

Prompting

Write the prompt the way you would brief a designer: the subject, the setting, the composition, then the finish.

Put any text that must appear in quotes, and say where it goes: a poster with the headline "Night Market" across the top.
Name the medium and the light: 35mm film, overcast daylight, or studio product shot, soft box from the left.
For an edit, say what stays the same before what changes: keep the model and the jacket; replace the street with a beach at dusk.
Ask for one image and refine the prompt before raising count. Every image is charged, and a better prompt costs nothing.

Rights

OpenAI assigns the rights in outputs to you, and commercial use is allowed. Attribution is optional. OpenAI does not train on data sent through its business API unless you opt in.

GPT Image 2.5 Flare
openai/gpt-image-2.5-flare
Text → ImageImage → Image
$0.22 / image
GPT Image 2.5 Sunburst
openai/gpt-image-2.5-sunburst
Text → ImageImage → Image
$0.22 / image

Similar models

Luma UNI-1 Max
luma/uni-1-max
Text → ImageImage → Image
$0.102 / image
Seedream Pro
bytedance/seedream-pro
Text → ImageImage → Image
$0.10 / image
Qwen Image 2.1 Pro
alibaba/qwen-image-2.1-pro
Text → ImageImage → Image
$0.075 / image
Qwen Image 3.0
alibaba/qwen-image-3.0
Text → ImageImage → Image
$0.075 / image
Qwen Image 3.0 Pro
alibaba/qwen-image-3.0-pro
Text → ImageImage → Image
$0.075 / image