Image aspect ratios per model: GPT 2.5 lists 8, Banana 2.1 lists 14

ChatGPT Image 2.5 lists 8 aspect ratios on Sume, Nano Banana 2.1 lists 14, Ideogram 4.5 lists 15, and Flux 2 Pro lists 13. The count matters because a model only accepts what its catalog descriptor lists, and an unlisted ratio returns 400 unsupported_parameter.

All counts exclude auto, which several models also list for letting the provider choose, and are read from the image catalog in the repository.

The counts

Listed ratios per model, excluding auto.

What a short list does not stop

ChatGPT Image 2.5 has a second route to shapes outside its list. With image_size you can pass custom pixels where both edges are multiples of 16, the longest is at most 3840, the ratio is at most 3:1, and the area is 655,360 to 8,294,400 pixels. That reaches 21:9 as 2352x1008 and 3:1 as 2400x800.

Other models in this catalog are limited to their listed ratios, and for 8:1 or 1:8 only Nano Banana 2.1 lists them.

• Check the aspect_ratio descriptor, not a blog table, before pinning a ratio.
• Use auto on edits to keep the source shape.

How to check a ratio in code

Do not hard-code a table from a blog. Call GET /v1/images/models/{model}/endpoints or the model list, find the aspect_ratio descriptor, and test whether your value is in its enum before you submit. A three-line check at the top of your pipeline turns a 400 at generation time into a clear message at planning time.

Treat auto separately. It is a valid enum value on several models, and on edits it keeps the shape of the reference, but it is not a ratio you can compute a layout from.

• Read the descriptor before each batch.
• Fall back to a custom size where the model allows it.
• Log which ratio each approved image used.

Choosing by shape

For feed work, 4:5 is on GPT Image 2.5, Nano Banana 2.1, Seedream 5.0 Lite and Ideogram 4.5. For wide banners, use a model that lists 21:9 or a custom size. See the Image API docs for the descriptor format and how to read supported_parameters.