AI product photos: cutout, new scene, upscale, and the cost per SKU

One SKU through cutout, scene generation and upscale costs $0.0225 per image for the cutout, $0.15 for a 2K Nano Banana 2 scene and $0.20 per image for the upscale, before the default agent fee. Each step is its own call to its own endpoint, and each reserves its estimate when you submit.

Prices are from the API pricing rate card and Sume's pricing code; request shapes are from the Image API and API reference docs, read 2026-09-29.

What are the three calls?

How do I chain the cutout into a scene?

The cutout result is a PNG with alpha, mirrored to Sume. Pass its URL as an input_references entry and describe the scene. Keep the product identical in the prompt, and check the label text afterwards.

curl -X POST https://api.sume.com/v1/images \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/nano-banana-2",
    "prompt": "Keep the bottle identical. Place it on a marble counter, soft morning light.",
    "input_references": [
      {"type": "image_url", "image_url": {"url": "CUTOUT_URL"}}
    ],
    "resolution": "2K",
    "aspect_ratio": "4:5"
  }'

How many scenes should I generate per SKU?

Each extra image is another per-image charge, so n images cost n times the scene price. Start with two or three candidates, keep one, and upscale only that one. That keeps a SKU to one cutout, one scene batch and one upscale.

What if a step fails?

A failed or canceled job is not billed. Rerun only the failed step; the steps you already have are unaffected.