AI ghost mannequin: remove the mannequin, keep the shape

A ghost mannequin photo, also called an invisible mannequin photo, shows a garment in its worn, 3D shape with no mannequin or model in view. The traditional method is to shoot the garment on a mannequin, shoot the inside of the back neckline separately, and combine the two in an image editor. An AI image model can approximate the look in one step: send the on-mannequin photo, and the inside shot if you have it, as reference images, and ask the model to remove the mannequin, show the empty neck opening, and keep the garment exactly as it is.

On Sume, the AI step is one POST /v1/images request, and an optional cutout is a Sume RMBG 1.0 job. The facts below come from the Image API docs and the OpenAPI document behind the Sume API reference, read on 2026-09-28.

How do I make a ghost mannequin photo with AI?

• Shoot the garment on a mannequin, straight on, evenly lit, against a plain background.
• If you can, also shoot the inside of the back neckline: the part the mannequin's neck hides.
• Put both photos at public HTTPS URLs and send them in input_references to a model that edits from reference images. Localhost, private-network, and non-HTTPS URLs are rejected before submission.
• Set aspect_ratio: "auto" to keep the product photo's shape, and n: 4 for four versions to choose from.
• The call returns 200 with the images within 30 seconds, or 202 with a job to poll when the generation takes longer.

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": "Make a ghost mannequin product photo. The photo of the navy sweater on a white mannequin shows the product; the close-up shows the inside of its back neckline. Remove the mannequin completely, keep the worn 3D shape of the sweater, and show the inside of the back neckline through the empty neck opening. Keep the color, knit texture, seams, and label exactly as they are. Plain white background.",
    "input_references": [
      { "type": "image_url", "image_url": { "url": "https://example.com/sweater-on-mannequin.jpg" } },
      { "type": "image_url", "image_url": { "url": "https://example.com/sweater-inside-neck.jpg" } }
    ],
    "aspect_ratio": "auto",
    "n": 4
  }'

What should a ghost mannequin prompt say?

Five parts:

• The effect: “remove the mannequin completely; the garment keeps its shape as if worn”.
• The opening: “show the inside of the back neckline through the empty neck opening”, or the inside of the waistband for trousers.
• Which photo is which, by what each one shows. A reference carries only its type and URL, with no role field.
• What stays: the color, the fabric texture, the seams, the buttons, the prints, and the labels.
• The background: “plain white background”, or a light gray.

How do I get a transparent PNG of the garment?

Send the edit's result to background removal, POST /v1/rmbg-1.0/remove. It takes one public HTTPS image_url, its schema prefers a Sume media URL, and /v1/images results are Sume-hosted. The finished job returns a PNG with alpha. Remove background API covers the job, and white background product photos covers putting the cutout on pure white, with Amazon's photo guidance.

Will the garment look exactly like mine?

Not guaranteed. The model redraws the whole picture, so seams, prints, labels, buttons, and fabric texture can come back different. The inside of the neckline is redrawn too, and without a photo of it, the model invents it. Compare each version with the real garment before it goes on a product page, and keep your original photos. To show the garment on a person instead, see virtual try-on video API.

What does it cost, and what are the limits?

• The edit: each completed image is billed, so n: 4 is four images, and a failed or cancelled generation is not billed. Each model's price is in the pricing line of GET /v1/images/models/{model_id}/endpoints, and usage.cost in the response is the USD amount billed.
• The cutout: API pricing lists background removal at $0.0225 per image, plus a 5.5% agent fee by default.
• Result URLs from /v1/images are signed, so download the photos you keep.