mask_url returns 400 on Nano Banana 2 and Ideogram 4.5: your options

Region editing is the headline of two launches. FLUX 3 Image's docs describe edits by boxes on a 0 to 1000 grid, and Midjourney's V8.2 edit model can inpaint and outpaint in its editor (Black Forest Labs docs, Midjourney updates feed, both read 2026-10-04). If you carry a mask workflow over to the Sume API, the first thing you will hit is a 400.

Why you get 400

The Image API docs list mask_url as an edit mask URL that only ChatGPT Image 2.5 (both variants) supports. The catalog is the parameter allowlist: if the selected model does not list a parameter, Sume rejects it with 400 unsupported_parameter and does not silently drop it. That is deliberate, so a mask never goes unapplied without you knowing.

Three ways forward

The third option is the only one that guarantees unchanged pixels, because your code does the copy. The other two rely on the model.

• Use ChatGPT Image 2.5 with mask_url. The mask is guidance, so verify the result.
• Keep the model and describe the region in words: "change only the mug on the left third; keep everything else identical".
• Edit, then composite the changed region back over the original in your own code, so pixels outside your box come from the source.

A request that is allowed

Send the source in input_references and the mask in mask_url, both as public HTTPS URLs, and keep the mask the same size as the source.

import os
import requests

body = {
    "model": "openai/gpt-image-2.5",
    "prompt": "Replace only the masked area with a ceramic mug. Keep everything else identical.",
    "input_references": [{"type": "image_url", "image_url": {"url": "https://example.com/desk.png"}}],
    "mask_url": "https://example.com/desk-mask.png",
}
r = requests.post(
    "https://api.sume.com/v1/images",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
    json=body,
    timeout=60,
)
print(r.status_code, r.text[:300])