GPT Image 2

What it does

GPT Image 2 is OpenAI's image model, available on pipe2 through the Image Generator pipeline. It takes a text prompt and produces a still image, and it also accepts an image you supply and returns a modified version of it. The catalog entry lists multilingual text rendering, 2K native output, and agentic edits as its distinguishing traits. You can render in 1:1, 16:9, 9:16, 4:3, or 3:4.

When to use it

• Posters, covers, and ad creative where the words inside the image have to be legible. The catalog calls out multilingual text rendering, which is what this use case leans on.
• Product shots for a listing or a landing page, generated from a written description rather than a photo shoot.
• Character art and photoreal portraits through Image Generator, where you iterate on a prompt until the look holds.
• Editing an existing image: upload a reference photo and describe the change instead of starting a new render from scratch.
• Producing the same concept in several framings (square for a feed, 16:9 for a header, 9:16 for a story) from the aspect ratios the pipeline exposes.

What to know before you run it

The catalog lists a single wholesale unit for this model: high quality at 1024². That is the unit the Image Generator run is metered against, so it is worth writing the prompt carefully before you submit rather than burning runs on throwaway tests. A reference image is optional, not required: text-to-image runs from a prompt alone, and you only upload an image when you want an edit of something specific. The five aspect ratios are selected on the pipeline form. Current credit cost is shown in the live price table on this page.

Also known as

You may see this model referred to as ChatGPT Images 2.0, or by its versioned API identifier gpt-image-2-2026-04-21. Both refer to the same model listed here.