GPT Image 2 image sizes: what OpenAI’s API docs actually confirm

What can be safely confirmed

Output size: there is a size field, but not a full list here

OpenAI’s Images API documentation describes size as the size of the generated image . The same Images API material also refers to fields such as background, output_format, quality, and token usage information for GPT image models .

The most concrete size value in the supplied sources appears in the Create image edit example, where the response shows output_format as png, quality as low, and size as 1024x1024 .

That does not prove GPT Image 2 only supports 1024x1024. It also does not prove a broader list of supported square, portrait, landscape, or high-resolution sizes. The more accurate wording is: OpenAI’s Images API has a size field; the cited image edit example shows 1024x1024; the supplied sources do not show a complete set of valid size values specifically for GPT Image 2 .

For engineering teams, that distinction matters. An example value in a reference page is not the same thing as a contractual support matrix.

Input images and edits: workflow confirmed, limits not fully specified

OpenAI provides API references for image editing through Create image edit . OpenAI’s cookbook also describes a mask-based editing workflow: a user can provide a mask if they do not want the model to change a specific part of the input image .

The mask caveat is important. The cookbook says the model might still edit some parts of the image inside the mask, although it will try to avoid doing so; if an exact mask is needed, it suggests using an image segmentation model . In other words, masks should be tested carefully if your product depends on preserving a region exactly.

From the supplied sources, these points are supported:

• OpenAI’s Images API includes image editing references .
• Mask-based editing is described in OpenAI’s GPT Image cookbook material .
• A mask should not be treated as a pixel-perfect guarantee that a region will remain untouched .

What the supplied sources do not confirm is just as important: they do not provide a complete GPT Image 2-specific list of accepted input file formats, maximum input file size, maximum input resolution, number of input images per request, or any separate alpha-channel requirement .

Be careful with third-party provider schemas

Some third-party providers also publish GPT Image 2 pages. Runware describes GPT Image 2 as a general-purpose GPT Image family model for text-to-image generation and image editing . Fal.ai has a GPT Image 2.0 page with a playground, API access, and its own schema .

Those pages can be useful if you are calling GPT Image 2 through that provider’s infrastructure. But they should not be mixed into OpenAI’s own API contract without checking. A size enum, file limit, or request schema from a middle-layer provider does not automatically become an official OpenAI API constraint .

Practical implementation checklist

• Do not hard-code a full size list from one example. 1024x1024 appears in the cited image edit example, but that alone is not a complete GPT Image 2 size matrix .
• Validate against OpenAI’s current docs before launch. If you call OpenAI directly, the GPT Image 2 model page and OpenAI Images API reference are the right starting points .
• QA mask behavior with real assets. OpenAI warns that masked regions may still be partially edited, so treat masks as guidance rather than an absolute lock .
• Separate provider configuration. If you use Runware or Fal.ai, follow that provider’s schema for that integration; if you call OpenAI directly, rely on OpenAI’s documentation .

Bottom line

GPT Image 2 appears in OpenAI’s API documentation . OpenAI’s Images API includes a size field for generated images . The cited Create image edit example shows 1024x1024 . Based on the supplied sources, however, it would be premature to publish a complete list of GPT Image 2 image sizes or a definitive set of input-image limits.