3:2 photo ratio image API: GPT Image 2.5 does not list it on Sume

GPT Image 2.5 does not list a 3:2 aspect ratio on Sume, and neither does GPT Image 2. If you want the classic 35 mm photo shape, use one of the 14 rows that do, such as Nano Banana Pro, Nano Banana 2.1, FLUX.2 pro, Seedream or Qwen Image, or render GPT Image 2.5 at 4:3 and crop.

Rows that list 3:2

Rows that do not

ChatGPT Image 2.5, ChatGPT Image 2 and both Imagen 4 rows leave 3:2 out. GPT Image 2.5 lists auto, 1:1, 16:9, 9:16, 4:3, 3:4, 5:4, 9:8 and 4:5. Imagen 4 lists only 1:1, 16:9, 9:16, 4:3 and 3:4. A request for a ratio outside a model's list is not a fallback; check supported_parameters before you submit.

GPT Image 2.5 does accept custom pixels via image_size. Both edges must be a multiple of 16, the longest edge at most 3840, the ratio at most 3:1 and the total between 655,360 and 8,294,400 pixels. A 3:2 shape such as 1536x1024 passes those rules (both edges divisible by 16, 1,572,864 pixels). So pixel sizes are one way to get a 3:2 canvas from that row; the ratio list does not stop you.

Which to use

Match the row to the job; the list below is a starting point, not a ranking.

• Photo look from a reference: Nano Banana Pro or 2.1.
• Cheapest: Qwen Image, FLUX.2 pro, Seedream.
• Text in a landscape: Ideogram 4.5 lists 3:2.
• If you need GPT Image 2.5 specifically: send image_size 1536x1024 instead of aspect_ratio.

When 3:2 matters and when it does not

3:2 is the shape of most camera sensors, so it matters when you mix generated images with real photographs in one gallery: matching ratios avoid uneven grid cells. It matters less on social feeds, where 4:5 and 9:16 dominate, and on web heroes, where 16:9 is common.

If your gallery is real photos at 3:2, generate the missing pieces on a row that lists 3:2 and keep the same model for the whole set, so lighting and grain look related. Sume's sume/auto router does not say which family it used, so pin an id for this.

Check before you spend

The pre-check is cheap and has no per-call fee: list the models and read the ratio descriptor for the id you intend to call. If you send a ratio and an image_size together, the docs note that custom-pixel models ignore the ratio when image_size is set, so send one or the other.

A wrong guess on a ratio fails the call, and failed generations are not billed, but it still burns a retry. For a batch job, validate the whole list of requests against the catalog once at the start.