317 furniture photos: cutouts plus 40 hero upscales cost $15.13

A furniture catalog of 317 SKUs, with a cutout for every photo and an upscale for the 40 hero images, costs $15.13 on Sume at list price. The cutouts are 317 x $0.0225 = $7.1325. The upscales are 40 x $0.20 = $8.00. Together that is $15.1325, which a wallet shows as $15.13.

The useful point is the split. Remove-background is cheap enough that you can run it on every file without a triage step. Image upscale is nearly nine times the price of one cutout ($0.20 against $0.0225), so the real decision is how many images get it.

The line items

Both prices below are the fixed per-image prices Sume publishes in its public catalog. Neither depends on how large the input is. Neither is multiplied by anything else, so the arithmetic is a plain count times a price.

What the 40 hero shots are for

Pick the hero set before you spend on it. A reasonable rule for a furniture shop is one image per collection page, one per category tile, and one per SKU that sits in a paid ad. The other 277 products only need a clean cutout on a plain or generated backdrop, which a listing page shows at a small size.

If you later decide 60 heroes are worth it, the cost moves by $0.20 each, so the total becomes 317 x $0.0225 + 60 x $0.20 = $19.1325. You can price any variant in one line before you submit anything.

The request

Both routes take a public HTTPS image URL. The cutout route is POST /v1/rmbg-1.0/remove and the upscale route is POST /v1/image-upscale-1.0/upscale. Neither accepts a model field, because Sume picks the engine on the server. Send mode: "async" for a batch, keep each job id, and read the PNG from the job result.

Run the cutout first for the products that need transparency, then upscale the cutout output for the hero set. The order matters for alpha, and the post on cutout then upscale or the reverse covers that choice.

curl -X POST https://api.sume.com/v1/rmbg-1.0/remove \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"image_url": "https://example.com/sofa-0042.jpg", "mode": "async"}'

What to budget beyond the list price

A job that fails is not a finished cutout, and the Image API docs say failed or cancelled image generations are not charged. Keep your own ledger of SKU to job id so a retry reuses the same record. For a 317-item run, set aside a few percent for re-cutting photos you reject on review. At 5 percent that is about 16 images and $0.36.

The media tool price sheet for 10, 100 and 1,000 files gives the same prices at other sizes if your catalog is not 317.

A review loop that keeps the cost down

Cutouts fail in predictable ways on furniture: thin chair legs, glass tabletops and fabric that matches the backdrop. Look at a sample of 20 results before you submit the other 297, and note which SKUs need a different source photo instead of a re-run.

A re-run on the same photo usually gives a similar edge, so a second attempt costs $0.0225 and may not help. Re-shooting or re-cropping the source is a better fix for the hard cases.

• Submit 20 cutouts first and open each PNG on a dark and a light backdrop.
• Mark hero SKUs in your sheet before any upscale job is created.
• Store the job id next to each SKU, and never create a second paid job for the same SKU without checking the first.
• Use metadata on each request to carry the SKU, so a result can be matched without a lookup table.