GPT Image 2.5 1536x1024 preset vs custom sizes on Sume

Use 1536x1024 as a preset, or as a custom size: it passes every custom-size rule. OpenAI lists three presets (1024x1024, 1536x1024 and 1024x1536) plus custom sizes, and Sume's ChatGPT Image 2.5 rows accept custom pixels on image_size when both edges are multiples of 16, the longest edge is at most 3840, the ratio is at most 3:1 and the total is 655,360 to 8,294,400 pixels.

The rules

OpenAI's custom rule on its guide matches Sume's: multiples of 16, ratio 1:3 to 3:1, 655,360 to 8,294,400 pixels. Both pages state a 3840 maximum edge.

Worked examples

Worked checks, writer-derived: 1536 x 1024 = 1,572,864 pixels, ratio 1.5, and 1536 / 16 = 96 and 1024 / 16 = 64 are whole numbers, so it passes. 1080x1350 fails because 1080 / 16 = 67.5. The nearest valid pair is 1088x1360 (68 x 16 and 85 x 16), at 1,479,680 pixels. 3840x2160 is 8,294,400 pixels, exactly the upper limit.

A pre-flight check

The function below checks a size against the four rules before you spend a call.

def check_size(w, h):
    px = w * h
    problems = []
    if w % 16 or h % 16:
        problems.append("edges must be multiples of 16")
    if max(w, h) > 3840:
        problems.append("max edge is 3840")
    if max(w, h) / min(w, h) > 3:
        problems.append("ratio over 3:1")
    if not 655_360 <= px <= 8_294_400:
        problems.append("pixels outside 655,360-8,294,400")
    return problems or ["ok"]

for size in [(1536, 1024), (1080, 1350), (1088, 1360), (3840, 2160)]:
    print(size, check_size(*size))

Cost note

Sume says the auto size and presets without a verified pixel mapping reserve the upper bound of output tokens, so the reserve can be higher than a final charge. For a Sume price line, read usage.cost on the response, and for a plan use quality: xhigh at 1024x1024 is $0.09366 and max is $0.21072 at list output rates before input tokens and Sume's margin.