GPT Image 2 Image Sizes and Aspect Ratios: What’s Actually Verified

That distinction matters. It does not prove GPT Image 2 has no fixed options. It means the evidence available here does not verify what those options are.

What the sources actually support

Why social aspect-ratio screenshots are not enough

The Threads snippet is useful as a clue because it names concrete aspect-ratio labels: Square 1:1, Landscape 5:4, Story 9:16, and Ultra Wide 21:9.^[10] But a social screenshot or post is not the same thing as an API contract.

For production use, developers need more than a label. They need to know the exact parameter name, the full accepted-value set, whether the values apply to the API or only to a user interface, what pixel dimensions are returned, and whether access differs by account or surface. The provided social snippets do not answer those questions.^[10][11][12]

The Reddit snippet is similarly directional but incomplete: it says GPT Image 2 supports a fixed set of sizes and aspect ratios, but it does not enumerate them.^[11] That makes it a reason to check the API documentation—not a source for a compatibility table.

Do not treat Sora’s 1280x720 as a GPT Image 2 size

The most concrete dimension in the supplied OpenAI excerpts is 1280x720, but it appears in OpenAI’s Sora video-generation guide.^[14] The shown request uses the /v1/videos endpoint, the sora-2-pro model, a size="1280x720" field, and an eight-second generation parameter.^[14]

That supports a claim about a Sora video request. It does not verify that 1280x720 is accepted by GPT Image 2 for image generation.^[14]

How to handle GPT Image 2 sizes in production

Until you have a current official table or verified account-level behavior, avoid hard-coding GPT Image 2 dimensions from social posts, screenshots, or unrelated model docs.

A safer workflow:

1. Start with OpenAI’s GPT Image 2 model page and Images and Vision guide.^[1][13]
2. Confirm the accepted parameter values for the exact model and endpoint you are calling.^[1][13]
3. Test one size or aspect-ratio value at a time rather than assuming that UI labels map directly to API values.
4. Record both the requested value and the returned image dimensions, because prompt wording is not the same as a documented API parameter.
5. Treat claims such as “square works best” or “vertical works best” as unverified unless they come from official guidance or controlled tests; the provided official excerpts do not rank aspect ratios by quality.^[1][13]

What you can safely say right now

You can safely say that the provided source set includes OpenAI documentation pages for GPT Image 2 and Images and Vision.^[1][13] You can also say that the available excerpts do not verify exact GPT Image 2 pixel sizes, supported aspect ratios, or a best-performing aspect ratio.^[1][13]

You should not present the social labels as confirmed API values, and you should not reuse Sora video dimensions as GPT Image 2 image-generation requirements without separate evidence.^[10][14]

Bottom line

Do not publish or build around a definitive GPT Image 2 size table from this evidence alone. The official sources included here point to where the answer should be verified, but the supplied excerpts do not provide the values. Until the accepted sizes and aspect ratios are confirmed in current OpenAI docs or direct testing, exact GPT Image 2 image sizes and “best” aspect ratios remain unverified.^[1][13]