ChatGPT Images flyer template: a fill-in prompt for the API

A flyer template on Sume is a prompt you fill in, not a request field: the request table lists no template parameter. Write the flyer as labeled slots with the exact words in quotes, send it to openai/gpt-image-2.5 with a portrait image_size such as 1536x2048, and raise quality above the default when the flyer carries a lot of text.

Parameters below come from Sume's Image API docs and the Image 1.0 page, read 2026-09-29.

What goes in the fill-in prompt?

Name every element and give each its literal text. Keep the layout instructions separate from the words to print, so you can swap the words for the next event and leave the layout alone.

Which size and quality suit a flyer?

Flyers are portrait, so use image_size with custom pixels: both edges must be multiples of 16, the longest edge at most 3840, and the ratio at most 3:1. 1536x2048 fits all of that. Quality defaults to high when omitted. Sume's Image 1.0 page tells callers to escalate quality for finals and dense text, so test xhigh on the final pass. Higher quality is one of the settings the docs name as more likely to return a 202 job.

What does the request look like?

Fill the slots, send once, and read the Sume-hosted URL from data[].url. The usage.cost field reports the USD amount billed.

curl -X POST 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": "Portrait event flyer. Headline text: \"Rooftop Jazz Night\". Date line: \"Friday, October 9, 8 PM\". Place line: \"Harbor Street Terrace\". Footer: \"Free entry. Book at example.com\". Warm evening photo, bold serif headline, wide margins.",
    "image_size": "1536x2048",
    "quality": "xhigh",
    "output_format": "png"
  }'

How do I catch misspelled words?

Read the printed text against your slot list every time, word by word. Generated text can still drift, and Sume's docs make no accuracy promise for it. If a line is wrong, change only that slot and resend rather than rewriting the prompt, so the layout you liked stays put.

How do I reuse the layout for the next event?

Keep the layout sentence fixed and store only the slot values. A small JSON file with headline, date, place and cta keys, merged into the prompt by a script, lets you produce the next flyer by editing four strings. The layout text stays identical, so the design stays recognizable from one event to the next.

Save each accepted prompt next to the image it produced. When a client asks for the same flyer with a new date, you resend the saved prompt with one line changed, and check that line first. That is the closest a plain prompt gets to a template, and it needs no product feature.