Declutter a room photo: mask out furniture on ChatGPT Image 2.5

To declutter a room photo, send the original as the single reference and a mask over the furniture as mask_url to ChatGPT Image 2.5. One attempt costs $0.065875 at the catalog's high 1024 rate on Sume, so 20 rooms with two attempts each come to $2.635.

ChatGPT Image 2.5 (the Flare and Sunburst ids) is the family that lists mask_url in the Sume catalog; the other image models do not take a mask. The parameter facts are from the Sume Image API docs (read 2026-10-03).

What you need per room

The table below lists the figures this post uses, as read on 2026-10-03.

Getting a clean result

Draw the mask a little larger than the furniture so shadows and contact marks fall inside it, and name what should replace the item ("bare hardwood floor and plain wall, same light as the rest of the room"). Leave windows, doors and fixed features out of the mask so the model has nothing to change there.

Compare each output against the original for new details the model invented: a plug socket, a skirting board, a window latch. Edits that change the property are a legal problem, not just a quality problem.

• Use aspect_ratio: "auto" on edits; omitting the field is not the same as auto.
• Keep the untouched original and link it from the listing where the rules ask for it.
• Check your local disclosure rules for AI-edited listing photos before publishing.

The request

Swap the two URLs and the prompt per room.

{
  "model": "openai/gpt-image-2.5",
  "prompt": "Remove the sofa and rug inside the mask. Bare hardwood floor, same lighting, no other changes",
  "aspect_ratio": "auto",
  "mask_url": "https://example.com/room-12-mask.png",
  "input_references": [
    {"type": "image_url", "image_url": {"url": "https://example.com/room-12.jpg"}}
  ]
}