AI jewelry product photography from one real photo

AI jewelry product photography works best as an edit: send a real photo of the piece, change only backdrop and light, then check every stone.

5 min readSume
All posts

AI jewelry product photography works best as an edit, not a new picture: photograph the real piece, send that photo to an AI image model as a reference, and ask it to change only the backdrop and the light. Then check the stones, prongs, clasp, and engraving at full size before you publish, because the model draws a new image and small details can come back changed.

The steps below use Sume's Image API and the OpenAPI document behind the Sume API reference, read on 2026-09-28.

Why not generate jewelry from a text prompt?

A prompt alone invents a ring. It may look right at a glance, but the stone count, the setting, and the metal tone are the model's guesses, and a buyer expects the piece in the photo. Starting from your own photo keeps the model working on the piece you sell. Only models that take reference images can do this: a model whose input_references descriptor is {"min": 0, "max": 0} is text-to-image only and rejects references.

How do I turn a phone photo into a studio jewelry shot?

  • Shoot the piece in soft daylight on a plain surface, in sharp focus on the stones. Straight on, plus one angle.
  • Put the photo at a public HTTPS URL. Localhost, private-network, and non-HTTPS URLs are rejected before submission.
  • Send it in input_references to POST /v1/images. On edits, the docs advise aspect_ratio: "auto" to match the reference.
  • Ask for one change and list what stays: the band, the stones, the setting, the engraving, the metal color.
  • Set n for several versions. The docs allow up to 10 per call, with lower per-model ceilings.
curl -X POST "https://api.sume.com/v1/images" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-image-2.5",
    "prompt": "Place this exact ring on a plain light-gray studio surface with soft, even light and gentle reflections. Keep the band, the number and cut of the stones, the prongs, the engraving, and the metal color exactly as they are. Do not add stones or sparkle.",
    "input_references": [
      { "type": "image_url", "image_url": { "url": "https://example.com/ring.jpg" } }
    ],
    "aspect_ratio": "auto",
    "n": 3
  }'

How do I get a white or transparent background?

Cut the piece out with Sume RMBG 1.0: POST /v1/rmbg-1.0/remove with one public HTTPS image_url. The completed job returns a PNG with an alpha channel, and /v1/images results are Sume-hosted URLs you can pass straight in. White background product photo with AI covers flattening a cutout onto white.

What should I check before I publish?

Open each version at full size next to the original photo. The details that matter most in jewelry are the smallest ones:

  • The number, shape, and cut of the stones, and their color.
  • Prongs, bezels, and the links of a chain or bracelet.
  • The clasp, the bail of a pendant, and any hallmark or engraving text.
  • The metal tone: yellow, white, and rose gold should stay what they are.
  • Proportion: a stud must not grow into a statement earring.

How much does it cost, and what are the limits?

Each completed image is billed at its model's pricing line, so cost_usd × n is what you pay; a failed or cancelled generation is not billed. /v1/images result URLs are signed, so download the files you keep.

From Image API, the Sume API reference, and API pricing, read 2026-09-28.
StepEndpointWhat comes backPrice
Restage the photoPOST /v1/imagesImages, or a job to poll after 30 secondsPer image, from GET /v1/images/models/{model_id}/endpoints
Cut out (optional)POST /v1/rmbg-1.0/removeA PNG with alpha$0.0225 per image, plus a 5.5% agent fee by default
Enlarge (optional)POST /v1/image-upscale-1.0/upscaleAn image at upscale_factor 1–4, default 2$0.20 per image, plus a 5.5% agent fee by default

Sources

Related posts

More in Use cases

All Use cases posts

Written by Sume