AI worksheet clip art set: 12 transparent PNGs for $3.37 at xhigh list

A set of 12 transparent PNG clip-art items with three takes each is 36 images. The gpt-image-2.5 xhigh output estimate in Sume's docs is $0.09366 per 1024x1024 image, so 36 images come to $3.37176 at list, before input tokens and before Sume pricing. Pin quality yourself; for flat clip art, the default high is the sensible place to start.

Keep the style identical across the set by generating one anchor item first and passing it as a reference for the other eleven.

Cost of the set (read 2026-10-04)

The only per-image estimates the docs print for gpt-image-2.5 are the 1024x1024 xhigh and max ones. Replace them with the cost_usd pricing line from the model's endpoint record, which already includes Sume's margin.

One request per item

Each item is its own call: the same style line, the anchor as a reference, a different object, and background: "transparent". Send n takes only if you will actually choose between them; a failed generation is not billed, but a completed one always is.

{
  "model": "openai/gpt-image-2.5",
  "prompt": "Same flat classroom clip-art style as the reference. A red apple with a green leaf, bold outline, no shadow, nothing behind it, no text.",
  "input_references": [
    {"type": "image_url", "image_url": {"url": "https://example.com/anchor-pencil.png"}}
  ],
  "background": "transparent",
  "output_format": "png",
  "aspect_ratio": "1:1"
}

Check the arithmetic

The script reproduces the table.

items, takes = 12, 3
for quality, price in [("xhigh", 0.09366), ("max", 0.21072)]:
    print(quality, round(items * price, 5), round(items * takes * price, 5))

Submitting 36 calls

Use mode: "async" or a webhook for a batch, and keep status_url polling as the fallback. If your workspace's concurrency plus queue is full you get queue_full; wait for running jobs to finish. Give each item its own Idempotency-Key so a retry after a timeout does not run, or bill, twice.