Instagram 4:5 portrait image API: 14 Sume models list 4:5

Fourteen of the 18 image rows in the Sume catalog list a 4:5 aspect ratio, so most image models can produce an Instagram portrait feed frame directly. The cheapest is Qwen Image at 3 cents per image and the dearest is ChatGPT Image 2 at 27 cents. Only Higgsfield Soul, Grok Imagine and the two Imagen 4 rows do not list 4:5.

All 14 rows, cheapest first

Prices below are catalog list times 1.25, rounded up to the cent, at the default size or quality the catalog uses; read each row's endpoint record for the exact figure on your call.

What 4:5 means on Sume

The docs are specific: 4:5 is Instagram portrait, 1080x1350, not 4:3. Models return their native size for that ratio. Nano Banana Pro, for example, is sent aspect_ratio: "4:5" and returns about 928x1152 at 1K. Exact 1080x1350 comes from a documented post-step through job target_pixels, so ask for that if your pipeline needs the exact pixel size rather than the ratio.

GPT Image 2.5 also accepts custom pixels through image_size: both edges must be multiples of 16, so 1080x1350 is not valid (1080 is not divisible by 16) while 1088x1360 is. That is the same 4:5 shape with 8 extra pixels of width and 10 of height to trim.

Choosing among the 14

Pick by what the frame needs; each bullet names a row from the table above.

• Cheapest drafts: Qwen Image, Seedream 4.0 or FLUX.2 pro at 3 to 4 cents.
• Text in the frame: Ideogram 4.5 (8 cents at medium) or GPT Image 2.5 (7 cents at high).
• Photoreal product shots from a reference: Nano Banana 2.1 at 10 cents or Pro at 19 cents.
• Text-only rows with no edit: Recraft V4 and Qwen Image Max do not accept references.

A request

The same body works across rows because aspect_ratio is a normalized field. Change model and the price changes; the ratio check is done against that model's own list.

{
  "model": "openai/gpt-image-2.5",
  "prompt": "cafe interior, morning light, empty table in the foreground",
  "aspect_ratio": "4:5",
  "quality": "medium",
  "n": 2
}

Reading the edit column

A 'no' in the edit column means the row is text-to-image only: the API refuses input_references for it. If your 4:5 post starts from a product photo, filter the table to 'yes' before you compare prices. That leaves 12 rows, and the cheapest edit-capable one is still Qwen Image at 3 cents.

Image-to-image on GPT Image 2.5 accepts up to 16 references and an optional mask. Ideogram 4.5 takes up to 5 in total. Check the endpoint record for the rest before you build a flow around a reference count.

Carousel cost

A 10-slide carousel at 4:5 on Qwen Image is 30 cents, on Seedream 5.0 Lite 50 cents, and on GPT Image 2.5 at medium about 20 cents. Mixing is fine: render the cover on a stronger row and the body slides on a cheaper one, but keep style references consistent between them.