GPT Image 2.5 custom size: 1920x1080 fails, use 1920x1088

On Sume, openai/gpt-image-2.5 takes custom pixels on image_size only if both edges are multiples of 16, the longest edge is at most 3840, the ratio is at most 3:1, and the area is 655,360 to 8,294,400 pixels. 1080 is not a multiple of 16 (1080 / 16 = 67.5), so 1920x1080 fails the rule and 1920x1088 passes. 1280x720 and 3840x2160 both pass.

Common sizes checked against the rules

The rules come from the Image API docs. I computed each row by hand: edges divisible by 16, longest edge at most 3840, ratio at most 3:1, pixels in range.

Send it

Put custom pixels on image_size, not size; size takes a tier and rejects WxH. Then crop the 8 extra rows if you need an exact 1920x1080 in your own pipeline.

For Instagram portrait, aspect_ratio: "4:5" is the native route and avoids the arithmetic.

curl -s https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"openai/gpt-image-2.5","prompt":"product banner","image_size":{"width":1920,"height":1088},"quality":"medium"}'

Cost note

Credits scale with output size and quality for ChatGPT Image 2.5. The listed catalog rate is high quality at 1024 output, about $0.0527 list; larger sizes cost more. Read the rate card before a batch.