45 listings, 3 takes each: 135 images at 1K on three models
135 images (45 listings x n=3) cost $13.50 on Nano Banana 2.1 1K, $25.31 on Pro 1K and $3.34 on GPT Image 2.5 low.

The short answer
Forty-five listing photos with three takes each is 135 images. At 1K on Sume that is $13.50 on Nano Banana 2.1, $25.31 on Nano Banana Pro and $3.34 on GPT Image 2.5 low. Sume bills cost_usd x n, so asking for n=3 in one call costs the same as three calls.
The gap between the cheapest and the dearest row is large enough that a pick-the-best step should be priced before the model is chosen.
What n does
n repeats the same prompt and inputs and returns several images in one response. The Image API docs say most models take n from 1 to 4 and that the docs' general ceiling is 10, with per-model ceilings lower, so read the n descriptor in the catalog before relying on a number above 4. n=3 is inside every model's range in this post.
Use separate calls when each take needs a different instruction. Use n=3 when you just want variety from one prompt. Either way the bill is the same.
The totals
Each call is 3 x the row; 45 calls make the run. The table shows the arithmetic with the catalog default-options rows as of 2026-10-08.
| Model | Per call (n=3) | 45 calls | Per listing |
|---|---|---|---|
| Nano Banana 2.1 1K | 3 x $0.10 = $0.30 | $13.50 | $0.30 |
| Nano Banana Pro 1K | 3 x $0.1875 = $0.5625 | $25.31 | $0.5625 |
| GPT Image 2.5 low 1K | 3 x $0.02475 = $0.07425 | $3.34 | $0.07425 |
Budgeting a pick step
If you keep one of three takes and run the keeper through a second pass, say Pro at 2K ($0.1875), add 45 x $0.1875 = $8.44. A GPT low draft set plus a Pro 2K polish totals $11.78; three Pro 1K takes with no polish total $25.31.
A failed generation is not billed, so a retry only costs when the retried call completes. Hold a small reserve for slow calls that return a 202 job and finish later; they bill on completion, the same as sync calls.
Reading the totals
Pro at 1K costs 1.8750 times the 2.1 row at the same tier, and 2.1 costs 4.04 times the GPT low row. Those multipliers carry straight through to the 135-image run, so the question is whether the better model needs fewer takes. If Pro gets a usable image in 2 takes where 2.1 needs 3, the Pro run is 45 x 2 x $0.1875 = $16.88 against $13.50, and 2.1 still wins on price.
The tier you pick also matters. The 2.1 row falls to $0.075 at 0.5K, which would put the same 135 images at $10.12. That is fine for thumbnails and contact sheets. It is not the size to hand to a marketplace that wants a large main photo.
Whatever you choose, test with one listing before you submit the other 44, and compare usage.cost with the row.
Sources
Related posts
More in Pricing
- 45,000-character chapter on Sume TTS: split evenly costs 2 cents more
A 45,000-character chapter needs three TTS requests. Splitting 20,000 + 20,000 + 5,000 costs $2.14; an even 15,000 x 3 split costs $2.16.
- A 45,000-character chapter needs three Sume TTS jobs: about $2.14
Sume TTS caps a job at 20,000 characters. A 45,000 character chapter splits 20,000 + 20,000 + 5,000 and costs 45 x $0.0475, about $2.14.
- Drop to 480p: Fabric saves 46.7 percent, H3 lip-sync 37.5 percent
Fabric is $0.10 at 480p against $0.1875 at 720p. H3 Max lip-sync is $0.0625 at 480p against $0.10 at 768p. Savings on 250 clips inside.
- 5 depositions of 3 hours: 15 hours of transcript, priced
Five 3-hour recordings are 15 hours: $8.10 on MAI-Transcribe-2-Streaming's intro price and $9.00 on Sume STT, in 90 jobs. Review needs are noted.
Written by Sume