Change a garment's colour in a photo with a mask_url edit

Send the original photo in input_references, a mask of the garment as mask_url, and a prompt that names the new colour, to POST /v1/images with openai/gpt-image-2.5. mask_url is documented only for ChatGPT Image 2.5 edits, and it must be a public HTTPS URL.

Details are from Sume's image docs, read 2026-10-01. The docs do not say which mask pixels mark the editable area, so test with one image before running a batch.

What does the request need?

The mask only covers the garment, so the prompt can stay short: name the item and the target colour, and say everything else stays the same.

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": "Change the jacket to deep green. Keep fabric texture, folds, skin and background unchanged.",
    "input_references": [
      { "type": "image_url",
        "image_url": { "url": "https://media.sume.com/artifacts/artf_demo/model.png" } }
    ],
    "mask_url": "https://media.sume.com/artifacts/artf_demo/jacket-mask.png",
    "aspect_ratio": "4:5"
  }'

Which limits apply to the edit?

How do I make several colourways?

Reuse the same photo and mask, and change only the colour in the prompt, one request per colourway. Keep the rest of the prompt identical so the differences you see come from the colour. For a catalogue workflow without masks, see product colour changer colourways.

How should I check the result?

Compare each output with the source at full size: look at seams, buttons, logos and edges where the mask meets skin or background. Runway's own changelog lists GPT Image 2.5 with up to 16 reference images, matching the count Sume documents. A colour edit is not a guarantee of the real product's dye, so do not label it as the actual item colour in a listing.