One layout reference, twenty posters: GPT Image 2.5 on Sume

OpenAI's Images 2.5 thread lists pre-designed formats for posters and merchandise that make customising faster (OpenAI Developer Community, read 2026-10-04). Those templates live inside ChatGPT. If you need your own template over an API, a single reference image can do the same job.

The recipe

ChatGPT Image 2.5 accepts up to 16 reference images on Sume, and edits with no size keep the source geometry (Sume Image API). Build one approved poster layout, host it at a public HTTPS URL, and send it as the reference with a prompt that changes only the text.

• Image 1 is the layout. Keep its grid, type size and logo position.
• Put the new headline in quotes and say how many times it appears.
• Say nothing about style beyond "same as the layout", so the model has less to change.
• Use aspect_ratio: "auto" so the output matches the layout's shape.

Cost of twenty

The catalog lists ChatGPT Image 2.5 at a high-quality 1024 px rate of about $0.0659 after Sume's margin, but it bills by tokens, so a larger poster at higher quality costs more. At the catalog figure twenty posters are about $1.32. Read usage.cost on each response and total it rather than trusting the estimate.

The loop

Check each result for a misspelt or doubled headline before it ships.

import os
import requests

h = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
LAYOUT = "https://example.com/poster-layout.png"
headlines = ["Autumn Sale", "Winter Market"]
total = 0.0
for text in headlines:
    body = {
        "model": "openai/gpt-image-2.5",
        "prompt": f'Keep the layout of Image 1 exactly. Change the headline to "{text}", shown once.',
        "input_references": [{"type": "image_url", "image_url": {"url": LAYOUT}}],
        "aspect_ratio": "auto",
    }
    r = requests.post("https://api.sume.com/v1/images", headers=h, json=body, timeout=90)
    if r.status_code == 200:
        total += r.json()["usage"]["cost"]
        print(text, r.json()["data"][0]["url"])
    else:
        print(text, r.status_code, "read the job envelope")
print("billed", total)