500-SKU product photo refresh: cost by Sume image model

Short answer

Refreshing 500 product photos with one edited image per SKU costs $12.50 on Grok Imagine or Qwen Image, $18.75 on Flux 2 Pro, $25 on Seedream 4.5 and $50 on Nano Banana 2 at 1K, using the per-image prices that Sume's image catalog publishes. Those prices are what Sume bills to your wallet, margin included, so the total is simply price times image count.

Every row below is a model that accepts reference images, so you can pass the original packshot as input_references and ask for a new background, a lighter shadow or a cleaner crop. Text-only models such as Imagen 4 and Recraft V4 are left out because they cannot take the photo.

How to read the numbers

The take count matters more than the model choice. Three takes per SKU triples every figure, and a reviewer will usually reject one in three, so budget for the 3-take column if a human picks the winner.

Sume bills a generation only when it completes. A failed or cancelled generation is not charged, so retries on provider errors do not inflate the bill. A request that returns 202 and finishes later is billed once, when it completes.

Run it in batches

• Read the live price first: GET /v1/images/models/{id}/endpoints returns the pricing line, and cost_usd x n is what you pay.
• Send one request per SKU with n set to the number of takes you want. Check the n range descriptor, because several models cap it at 4 and Grok Imagine caps it at 1.
• Set aspect_ratio: "auto" on edits so the output keeps the shape of the source photo. Leaving it out is not the same as auto.
• Use mode: "async" for large batches and poll the job, since a sync call waits only 30 seconds before it returns 202.

Caveats

ChatGPT Image 2.5 is billed from an output-token estimate plus estimated input tokens, so a reference edit there costs more than the catalog display rate and is not a flat figure. For a flat budget, stay on the models in the table.

Prices change when the catalog changes. Treat this table as a 2026-10-04 snapshot and re-read the endpoints call before you commit a large run. The full request shape is in the Sume Image API docs.

A worked example

Say you run 500 SKUs on Seedream 4.5 with two takes per SKU. That is 1,000 images at $0.05, or $50.00. If a reviewer approves 70 percent of first takes and you rerun the rest once, the real bill is closer to 500 + 150 = 650 images, or $32.50. Tracking the approval rate on a 20-SKU pilot gives you a better forecast than any list price.

The same arithmetic on Flux 2 Pro at $0.0375 gives $37.50 for the 1,000-image plan and $24.38 for the 650-image plan. The gap between cheap and mid-priced models is small next to the cost of a human reviewing the output, so pick on edit quality first and price second.

Before a large run

Prices and descriptors change when the catalog changes, so confirm them before you spend. Call GET /v1/images/models/{id}/endpoints for the row you plan to use and read its pricing line and supported_parameters; both come back in one response.

Then run a pilot of three to five images and read usage.cost on each response. Multiply by your planned count for a forecast you can trust. Completed generations are billed in full and failed or cancelled ones are not, so a pilot that errors costs nothing.

For big batches, use mode: "async" or mode: "webhook" with a public HTTPS webhook_url, so no request waits on the 30-second sync limit. Poll GET /v1/jobs/{id}/status and fetch GET /v1/jobs/{id}/result when the job completes.