Omni Flash edit in sync mode: it waits 30 seconds, then you poll

If you send an Omni edit with mode: "sync", Sume holds the request for at most 30 seconds, wait_timeout_seconds max, and then returns the job envelope whether the job is done or not. The docs say what to do when it is not terminal: poll, and do not resubmit. A second submit is a second paid job, unless you reuse the same Idempotency-Key.

What each mode returns

The 30 seconds is a wait budget and not a job duration. Video work can run longer, and the wait can be shorter when waiter capacity is not there. So write the client to accept either outcome: a finished job in the first reply, or an unfinished one with a status URL.

A client that polls

The sample submits an edit in sync mode and polls if needed. It reads the envelope defensively, because I did not confirm every key name against a live response, and it sends one fixed Idempotency-Key so a retry cannot make a second job.

import json, os, time, urllib.request

BASE = "https://api.sume.com"
H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}",
     "Content-Type": "application/json"}

def call(url, body=None, extra=None):
    req = urllib.request.Request(url, headers={**H, **(extra or {})},
        data=json.dumps(body).encode() if body else None)
    with urllib.request.urlopen(req, timeout=60) as r:
        return json.load(r)

body = {"model": "gemini-omni-flash-1.1", "mode": "sync",
        "prompt": "Replace the bottle with an apple. Keep everything else the same.",
        "video_url": "https://example.com/clip.mp4", "resolution": "720p"}
env = call(f"{BASE}/v1/video-router/generate", body,
           {"Idempotency-Key": "edit-demo-001"})
job = env.get("job", env)
status_url = job.get("status_url") or f"{BASE}/v1/jobs/{job['id']}/status"
while not job.get("terminal"):
    time.sleep(5)
    job = call(status_url)
    job = job.get("job", job)
print(job.get("status"), job.get("result_url"))

Before production

Add a deadline to the loop for production, and read result_url only when result_ready is true. If the process dies mid-wait, restart it with the job id from your logs and poll that job; never POST the edit again.