5 product photo variations from one packshot: GPT 2.5 vs Seedream

Five variations of one product packshot cost about $0.104 on GPT Image 2.5 at medium quality with one reference at 1024x1024, and $0.25 on Seedream 4.5 at its flat $0.05 per image. Both take input_references, so the flow is the same: pass your packshot URL, describe the new scene, set n. The Image API docs describe the reference format and limits.

The surprise is the order. The newer model is cheaper here because medium GPT Image 2.5 is billed by tokens, and one reference adds an estimated input-token charge of well under a cent.

What does the request look like?

The packshot must be a public HTTPS URL; Sume rejects localhost, private-network and non-HTTPS URLs before it submits. Four variations in one call, one more in a second:

{
  "model": "openai/gpt-image-2.5",
  "prompt": "Same bottle, unchanged label, on a sunlit bathroom shelf with eucalyptus",
  "input_references": [
    {"type": "image_url", "image_url": {"url": "https://example.com/packshot.png"}}
  ],
  "image_size": {"width": 1024, "height": 1024},
  "quality": "medium",
  "n": 4
}

How do the two models price five variations?

GPT Image 2.5 reserves an estimate that includes the input tokens of your reference, which Sume's catalog describes as an estimate rather than a measured count. Seedream is a flat rate per output image.

Which should I pick?

Run the same packshot and prompt on both for one image each, then compare label fidelity, because a changed label is the failure that costs you. Choose on that result, not on price. Prices here are list x 1.25 in dollars per image before whole-cent rounding; the catalog's billable formula reads "list x 1.25, ceil usd cents", so confirm the first charge in usage.cost and budget from the invoice, not from the sum.