AI model for jewelry: put your piece on a model

To put jewelry on an AI model, send an image model two references: a clear photo of your piece and a photo of the model (or a written description of one), and say exactly where the piece sits, such as on the ring finger of the left hand or just below the collarbone. Generate several versions and keep only the ones where the piece matches the real one in size, stones, and metal color.

The steps below use Sume's Image API and Format catalog docs, read on 2026-09-28.

How do I put jewelry on an AI model?

• Photograph the piece on its own: sharp, evenly lit, on a plain surface.
• Pick the person. Use a photo of someone who agreed to appear, or describe an invented model in the prompt.
• Host both photos at public HTTPS URLs and send them in input_references to POST /v1/images, on a model that takes references. ChatGPT Image 2.5 takes up to 16.
• Frame the shot around the piece: a hand for a ring, the neck and shoulders for a necklace, a three-quarter profile for earrings.
• Ask for several versions with n and compare each one with the real piece.

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": "Close-up portrait of the woman in the second photo wearing the pendant necklace from the first photo, resting just below her collarbone. Keep the pendant exactly as shown: its shape, stone, chain, and gold tone, at its real size of about 15 mm. Soft window light, plain background.",
    "input_references": [
      { "type": "image_url", "image_url": { "url": "https://example.com/pendant.jpg" } },
      { "type": "image_url", "image_url": { "url": "https://example.com/model.jpg" } }
    ],
    "n": 3
  }'

What should the prompt say?

Each reference is only a type and a URL, so the prompt has to say which photo shows what. Then:

• Placement: the finger, the ear, the wrist, or where the chain falls.
• Scale: the piece's real size, so a pendant doesn't come back twice as large.
• What must stay: stones, setting, clasp, chain, and metal tone.
• Framing and light: close crops keep the piece large in the frame.
• What the model wears: plain clothing keeps attention on the jewelry.

Can Sume's model-and-product Format do jewelry?

It isn't written for it. The catalog Format sume-model-product-portrait describes itself as a “model-and-product portrait image with intimate beauty framing, natural skin, accurate packaging”, for “skincare endorsements, cosmetic portrait campaigns, and model-led product stills”. Its brief is beauty packaging, so for rings and necklaces the direct image request above gives you control over placement. AI model holding your product covers that Format for beauty products.

How many versions can I get per call?

It depends on the model. Read its n range from the catalog before you pin a number.

What are the limits?

The output is a new picture, not a photo of the piece being worn. Size and drape are the model's guesses, so it is no fit or size guide, and small details can change between versions. It is also a finished image, not a live camera try-on that follows a shopper; What is virtual try-on? explains the difference. For motion, animate an approved still: AI video for jewellery shows how.