40 conference speaker cards from one template: $6.00 or $2.23

The short answer

Forty speaker cards, each an edit of one template image plus a headshot, cost $6.00 on Nano Banana 2.1 at 2K (40 x $0.15) or $2.23 on GPT Image 2.5 medium at 2K (40 x $0.055625 = $2.225). At 1K, Nano Banana 2.1 is $4.00 and the low GPT tier is $0.99.

A card run is a good test of the two launches this week because every output has the same layout, so you can price the whole batch from a single image price.

The arithmetic

Each card is one call with two input references, the template and the headshot. Read the endpoint pricing lines for the model first, because the docs note that GPT Image 2.5 input tokens are part of its underlying rate; the table uses the catalog per-image prices.

The table prices all 40 cards once. A second pass to fix the cards you reject is a separate row, so keep a margin of about 10 to 25 percent in the budget if your template is text-heavy.

How to run it

Send the cards as 40 separate calls rather than one call with n set to 40, because each card has different input references and n repeats one prompt and one set of inputs.

Keep the template and headshot URLs public HTTPS. Sume rejects localhost, private-network and non-HTTPS URLs before the job starts, and a rejected call is not billed.

Retries and cost per card

Some images will be rejected by eye, and rejected images are billed like any other, so budget a margin. The bullets add 20 percent to each Sume total in the table above (the totals multiplied by 1.2) and divide by the number of cards to give a per-card cost.

Treat the per-card figure as the number to compare across models, because it does not change when the count changes. Re-read the public catalog before a large run, since prices can change and the table is a snapshot from 2026-10-08.

• Nano Banana 2.1 at 1K: $4.00 becomes $4.80 with 20 percent extra images, which is $0.12 per card (a 40-card job).
• Nano Banana 2.1 at 2K: $6.00 becomes $7.20 with 20 percent extra images, which is $0.18 per card (a 40-card job).
• GPT Image 2.5 medium (2K): $2.23 becomes $2.68 with 20 percent extra images, which is $0.07 per card (a 40-card job).
• GPT Image 2.5 low (1K): $0.99 becomes $1.19 with 20 percent extra images, which is $0.03 per card (a 40-card job).

What the docs say to check

The endpoint pricing lines are what Sume charges your wallet and already include the Sume margin, so a call costs cost_usd times n. Sume bills image generation all or nothing: a generation that fails or is cancelled is not charged.

Check the status code, not the body shape. POST /v1/images waits up to 30 seconds and returns the images with 200. Slow settings (4K, high quality, a large n) are the most likely to come back as 202 with a job envelope, and you then read the images from the job result endpoint described in Jobs and results.

If the text on the card must be exact, put the speaker name in the prompt and compare a first sample before you spend the other 39 calls.

The numbers above are list prices read on 2026-10-08, not measurements. Sume figures are the catalog prices that already include its margin, Google figures are from its own pricing page, and quality or speed differences are not covered. Re-read both pages before you commit a large budget, because either side can change its rates.