18 ad images, 6 concepts x 3 ratios: what it costs on Sume rows

An 18-image ad test, six concepts at three ratios each, costs $0.675 on FLUX.2 Pro, $0.7875 on Seedream 5.0 Lite, $1.35 on Nano Banana 2.1 at 0.5K and $1.80 at 1K, at the billed prices in Sume's catalog on 2026-10-10. A spend near a dollar is a fair price for a first creative test, but only if every row in the plan lists all three ratios.

The prices are the endpoint pricing lines described in the Image API docs, already billed with Sume's margin included. The wallet charge for a call is that figure times n.

The test plan

Take six concepts, for example six product angles or six headlines. Render each at 1:1 for a feed square, 4:5 for a portrait feed post and 9:16 for a story. That is 18 images. Because FLUX.2 Pro, Seedream 5.0 Lite and Nano Banana 2.1 all list 1:1, 4:5 and 9:16, the same plan runs on each.

Grok Imagine is left out because it does not list 4:5. Imagen 4 is left out because it takes no references and does not list 4:5 either, which would break a plan that starts from a product photo.

Two-stage spending

A cheaper way to learn which concept works is to draft all 18 on the cheapest row, pick the three best concepts, and re-render those at a higher tier. Drafting 18 on FLUX.2 Pro costs $0.675. Re-rendering nine finals (three concepts, three ratios) at Nano Banana 2.1 2K costs 9 x $0.15 = $1.35. The total is $2.025, against $2.70 for rendering all 18 at 2K.

The saving grows with the batch. At 180 drafts and 90 finals the same plan costs $6.75 plus $13.50, or $20.25, against $27 for 180 finals.

Before you press go

Run the price check first: call the endpoints route for each row, multiply cost_usd by the number of images and compare with your wallet. The errors and credits page explains how an insufficient balance surfaces, so a batch does not fail halfway without a clear cause.

• Confirm each row lists all three ratios in its catalog entry.
• Pin the model id so the batch is repeatable.
• Save the row id and prompt with every output.
• Do not use sume/auto for a comparison, because it hides the family.

Why three ratios matter

A feed square, a portrait feed post and a vertical story are different crops, not resizes. Asking the model for each shape directly keeps the subject framed the way you intend. Cropping a 1:1 into 9:16 loses the sides, and extending it adds invented content.

That is the reason to confirm every row lists all three before you start. FLUX.2 Pro, Seedream 5.0 Lite and Nano Banana 2.1 each list them in the catalog.

Scaling the plan

Double the concepts and the cost doubles. Twelve concepts at three ratios is 36 images, or $1.35 on FLUX.2 Pro and $3.60 on Nano Banana 2.1 at 1K. If you add a fourth ratio such as a 16:9 banner, multiply by four thirds. The arithmetic is simple, and the point of doing it first is that you decide the budget before you press go, not after the wallet drops.

Use the endpoints route to read the live price rather than the numbers in this post, since a price can change after the date shown in the table caption.

What to record

For each of the 18 outputs, save the row id, the ratio, the prompt and the date. After the test you will want to know which concept won and on which row, and a spreadsheet with those four columns is enough. Add a fifth column for the cost, taken from the endpoint price times n, so the totals in your report match the wallet.