GPT Image 2 supported aspect ratios on Sume: 5:4, 9:8, 4:5

On Sume, gpt-image-2 takes the shared image ratios and, per the Image 1.0 page, also accepts 5:4, 9:8 and 4:5. For any other shape you can send custom pixels instead. Adobe's September 2026 Firefly notes say GPT Image 2 now supports 10 aspect ratios there; that count is Adobe's and applies to Firefly, not to Sume.

Sume facts below are from Image 1.0 and the Image API, read 2026-09-30.

Which ratios does gpt-image-2 accept?

The Image API page lists the normalized ratios as 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, 4:5, 5:4, 1:2, 2:1, 1:4, 4:1, 1:8, 8:1, 9:21 and 21:9, and also allows "auto". It adds that a model only accepts the values its catalog lists, so read supported_parameters for openai/gpt-image-2 before pinning one. The Image 1.0 page names 5:4, 9:8 and 4:5 for this model specifically.

What are the custom-size limits?

Send image_size as { width, height } or WIDTHxHEIGHT and it wins over aspect_ratio. For GPT the rule is: both edges a multiple of 16, longest edge at most 3840, aspect at most 3:1, and 655,360 to 8,294,400 total pixels.

What size do I send for 9:8?

The payload schema of the hosted MCP server notes that a 720×640 canvas is below the pixel minimum and says to use 864×768 for 9:8. Check it against the rule: 864 and 768 are both multiples of 16, and 864 × 768 is 663,552 pixels, above 655,360. The ratio is 9:8 exactly.

curl -X POST https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: gpt-image-2-9x8-001" \
  -d '{
    "model": "openai/gpt-image-2",
    "prompt": "A product shot on a plain background",
    "image_size": "864x768"
  }'

Should I use a ratio or pixels?

Use aspect_ratio when the shape is in the list; the field is ignored when image_size is set on custom-pixel models, so do not send both and expect the ratio to apply. Use pixels when you need a size the list lacks. For the full catalog across models, see image API aspect ratios and custom sizes.