GPT Image 2.5 mask edit: does the mask need an alpha channel?

For OpenAI's own API, yes: its guide says "The mask image must also contain an alpha channel" and that image and mask must match in format and size. Sume's docs list an optional public HTTPS mask_url for ChatGPT Image 2.5 edits but say nothing about alpha.

Sume's page does not state the alpha rule or any mask format rule, so whether Sume applies OpenAI's rule to mask_url is not documented. Preparing the mask to OpenAI's rule is a cautious choice, not a documented Sume requirement. Both pages were read 2026-09-30.

What does OpenAI require of the mask?

From OpenAI's image generation guide: the image and the mask must match in format and size (under 50MB), and the mask must contain an alpha channel. That rule is for OpenAI's API; this post does not claim Sume enforces it.

What does Sume accept?

On POST /v1/images, ChatGPT Image 2.5 (openai/gpt-image-2.5 and openai/gpt-image-2.5-sunburst) takes up to 16 image references, an optional mask_url, and background: auto|transparent|opaque. Reference and mask URLs must be public HTTPS; localhost, private-network and non-HTTPS URLs are rejected before submission.

What does the request look like?

Reference the source image in input_references and point mask_url at your mask (prepared to OpenAI's rule as a precaution).

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": "Replace the masked area with a ceramic mug, keep everything else",
    "input_references": [
      { "type": "image_url",
        "image_url": { "url": "https://media.sume.com/artifacts/artf_demo/photo.png" } }
    ],
    "mask_url": "https://media.sume.com/artifacts/artf_demo/mask.png"
  }'

What can go wrong?

OpenAI's guide says the image and mask must match in format and size; Sume's docs say URLs must be public HTTPS and are rejected before submission otherwise. Slow edits can degrade to a 202 job envelope; read the result from GET /v1/jobs/{id}/result in that case. Failed generations are not billed.

How do I check this myself?

Test the mask on one low-cost call before a batch. If the edit ignores your mask, open it in an image tool and confirm the alpha channel exists and that its dimensions equal the source image, the two conditions OpenAI names for its own API. The linked docs pages and the catalog endpoint show the current values, and this post reflects them as of 2026-09-30.