MAI-Image-2.6 returns base64 PNG; Sume returns a hosted image URL

MAI image models on Foundry return a JSON object whose data[].b64_json holds a base64 PNG, and the output format is always PNG (Microsoft Learn, read 2026-10-01). Sume's Image API returns data[].url, a Sume-hosted signed URL, with media_type per image, and the format can be png, jpeg or webp depending on what the model lists. So the save step differs: decode bytes for MAI, download a URL for Sume.

Sume keeps responses small on purpose: the Image API reference says media is mirrored and URLs are returned instead of inline base64.

What does each response look like?

Same job, different handling.

How do I save a Sume image to disk?

Check the status code first. 200 is the image response and 202 is a job envelope, so branch before reading data[0].url:

import os
import requests

r = requests.post(
    "https://api.sume.com/v1/images",
    headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
    json={"model": "openai/gpt-image-2.5", "prompt": "a lighthouse at dusk"},
    timeout=60,
)
if r.status_code != 200:
    raise SystemExit(f"not an image response: {r.status_code} {r.text[:200]}")
item = r.json()["data"][0]
ext = item["media_type"].split("/")[-1]
with open(f"out.{ext}", "wb") as f:
    f.write(requests.get(item["url"], timeout=60).content)
print("saved", ext)

Why keep your own copy?

A signed URL is a delivery link, not an archive. Download the file into your own storage as soon as the job finishes, and store the job id so you can fetch the result again from the jobs API, whose artifacts also live on media.sume.com. Jobs and results shows the artifact shape for a completed job.

Limits

MAI is not in Sume's image list as of this post; the table compares two APIs' conventions. PNG is lossless and large, so a page full of MAI outputs weighs more than the same page in WebP, and you would convert after download. On Sume, ask for output_format only when the model lists it: an unsupported value is rejected rather than ignored.