FLUX 3 pixel-exact local edits vs the Sume mask_url edit

Black Forest Labs' release notes for October 1, 2026 describe FLUX 3 Image with pixel-exact local edits: mark the element to change and keep the rest as it was, across several elements in one request. On the Sume image API, a documented mask edit is mask_url, listed for ChatGPT Image 2.5. Whether FLUX 3 is in the catalog is something you read from GET /v1/images/models, not from this post.

What the vendor says

From BFL's release notes.

The Sume route for a local edit

The Image API docs say ChatGPT Image 2.5 supports up to 16 references and an optional mask_url, a public HTTPS mask. Pair it with aspect_ratio: "auto" so the output keeps the source shape. For several elements, send one mask that covers them all, or edit one element per call from the original.

Edit with a mask

Replace the URLs with your own public files.

import os, requests

r = requests.post(
    "https://api.sume.com/v1/images",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
    timeout=60,
    json={
        "model": "openai/gpt-image-2.5",
        "prompt": "Recolour the masked jacket to forest green; leave all else unchanged.",
        "aspect_ratio": "auto",
        "mask_url": "https://example.com/jacket-mask.png",
        "input_references": [
            {"type": "image_url", "image_url": {"url": "https://example.com/photo.jpg"}}
        ],
    },
)
print(r.status_code, r.json())

Limits

Pixel-exact is the vendor's claim for its own model. Do not assume the same of another model; compare the unmasked area of an output with the source before shipping. Sume does not publish an equivalent guarantee.