50 Product Photos for a Marketplace: Cutout and Upscale Costs on Sume
Clean 50 SKU photos with RMBG at $0.0225 each, then upscale only the hero shots at $0.20: costs of cutouts, selective upscales and upscaling all 50.
The short answer
Background removal for 50 product photos costs $1.125 at Sume's $0.0225 per image. Upscaling every photo adds $10.00 at $0.20 each, for $11.125. Upscaling only the 10 hero images adds $2.00, for a total of $3.125 (catalog read 2026-10-08).
The two steps
rmbg_create removes the background from one image and image_upscale_create upscales one image; both are paid MCP tools that need an idempotency_key, and both have REST counterparts. Use a script of 50 calls if you do not want to type them: Sume's MCP docs describe script_run for programs that call hosted tools in a loop, with a max_paid_calls limit you can set to 50 to cap the spend.
Do the cutout first, on the original. Upscaling a photo with a bad edge makes a bad edge larger. Then pick the shots that need more pixels: the main image of each listing, anything that will be zoomed, anything for print.
| Plan | RMBG | Upscale | Total |
|---|---|---|---|
| Cutouts only | 50 x $0.0225 = $1.125 | None | $1.125 |
| Cutouts and 10 hero upscales | 50 x $0.0225 = $1.125 | 10 x $0.20 = $2.00 | $3.125 |
| Cutouts and upscale all | 50 x $0.0225 = $1.125 | 50 x $0.20 = $10.00 | $11.125 |
Marketplace rules and quality
Marketplaces set their own image rules, such as a white background or a minimum size, and those rules change. Read the current rules on the marketplace's own seller page for each channel before you build the pipeline; this post does not state any of them. Export cutouts on a flat color you can change per channel.
If the product has transparent or reflective parts, a cutout can lose edges. Check glass, hair-thin lines and shadows on a sample of five before you run all 50. A generated shadow is a separate step: use an image edit with image_urls and a prompt like "keep the product identical; add a soft contact shadow on white" ($0.055625 at medium quality and 2K on gpt-image-2.5).
Order the work so you can stop early. Run a five-photo sample for cutouts only, which is $0.1125, look at the edges, and then submit the other 45. If five of the 50 need a second pass, that is $0.1125 more. Upscale last and only for images that are going to be shown large; a thumbnail does not benefit from a $0.20 upscale.
- Keep the original files; Sume results are new artifacts.
- Use SKU ids in
idempotency_keyso a rerun does not pay twice. - Do not alter what the product is or does; retouching should not misstate color, size or condition.
Sources
Related posts
More in Use cases
- Animate a mascot from one image: H3 start frame or Omni reference
A mascot clip of 10 seconds costs $0.75 as an H3 start frame at 768p, or $1.25 as an Omni reference image at 720p on Sume. The two bodies and the difference.
- Meta GenAI ad tools: the input licence and your uploaded clip
Meta's Ad Creative Generative AI Terms grant Meta a perpetual licence to Inputs. Which clips are Inputs, and how to render a separate copy for those tools.
- Meta's GenAI ad terms join the Self-Serve Ad Terms on Oct 30
Meta says its Ad Creative Generative AI Terms merge into the Self-Serve Ad Terms on October 30, 2026. What it covers, and what to keep.
- Meta image animation or your own image-to-video: which to run
Meta Advantage+ can animate a product image; you can animate it yourself on Sume. Compare control, and note Meta's differing label statements on third-party AI.
Written by Sume