Launch kit from 6 references: 4 angles, a style card, a logo

Answer

ChatGPT Image 2.5 on Sume accepts up to 16 reference images, so a single request can carry four product angles, one style card and one logo, six in all, and still leave ten slots free. Name each reference in the prompt by its position, because the list order is the only handle you have.

Reference URLs must be public HTTPS. Sume rejects localhost, private-network and non-HTTPS URLs before the request is submitted.

Request

{
  "model": "openai/gpt-image-2.5",
  "prompt": "Launch hero: image 1 product centered on a warm stone surface, lighting and palette from image 5, logo from image 6 bottom right",
  "quality": "high",
  "aspect_ratio": "4:5",
  "input_references": [
    {"type": "image_url", "image_url": {"url": "https://example.com/front.jpg"}},
    {"type": "image_url", "image_url": {"url": "https://example.com/style.jpg"}}
  ]
}

Tips

• Use only the references you need. Each one adds input tokens to the estimate, and 2.5 is billed from tokens rather than a flat per-image rate.
• Add mask_url when you want to change one region and keep the rest. It is optional and specific to the 2.5 rows.
• background: transparent gives a cut-out for a launch page. If you leave the quality out, it defaults to high.
• Draft at low first. Sume's estimator puts a low 1024x768 image at about $0.005 against about $0.045 at high.

The 16-reference ceiling is a catalog value, so confirm it with GET /v1/images/models/openai/gpt-image-2.5/endpoints before you rely on it. Docs: Sume Image API.

Before a large run

Prices and descriptors change when the catalog changes, so confirm them before you spend. Call GET /v1/images/models/{id}/endpoints for the row you plan to use and read its pricing line and supported_parameters; both come back in one response.

Then run a pilot of three to five images and read usage.cost on each response. Multiply by your planned count for a forecast you can trust. Completed generations are billed in full and failed or cancelled ones are not, so a pilot that errors costs nothing.

For big batches, use mode: "async" or mode: "webhook" with a public HTTPS webhook_url, so no request waits on the 30-second sync limit. Poll GET /v1/jobs/{id}/status and fetch GET /v1/jobs/{id}/result when the job completes.