Edit a video with a prompt API: Python script for Omni Flash 1.1

To edit a video with a text prompt over an API on Sume, POST the source clip as video_url and your instruction as prompt to /v1/video-router/generate with model: "gemini-omni-flash-1.1". The script below submits that request, polls the job until it is terminal, and prints the result. It uses only the Python standard library.

Which route takes an edit

Google's Gemini API guide names Gemini Omni Flash as its default video model and describes it as a model for video generation and conversational video editing. Sume exposes the edit mode through the Video Router route, not through the OpenRouter-shaped /v1/videos body, which is why the path below is /video-router/generate.

The script

Router responses use Sume's { "data": ... } envelope, so the script unwraps it. The job id is read defensively from the usual names, because this post only relies on what the jobs docs promise: a job id comes back on submit, and GET /v1/jobs/{id}/status and /result read it.

import json, os, time, urllib.request

BASE = "https://api.sume.com/v1"
KEY = os.environ["SUME_API_KEY"]


def call(method, path, body=None, extra=None):
    headers = {"Authorization": f"Bearer {KEY}", "Content-Type": "application/json", **(extra or {})}
    data = json.dumps(body).encode() if body is not None else None
    req = urllib.request.Request(BASE + path, data=data, headers=headers, method=method)
    with urllib.request.urlopen(req, timeout=60) as r:
        out = json.load(r)
    return out.get("data", out)


job = call("POST", "/video-router/generate", {
    "model": "gemini-omni-flash-1.1",
    "prompt": "Replace the bottle with an apple. Keep everything else the same.",
    "video_url": "https://example.com/clip.mp4",
    "resolution": "720p",
    "mode": "async",
}, {"Idempotency-Key": "edit-demo-001"})
job_id = job.get("request_id") or job.get("job_id") or job.get("id")

delay = 5
while True:
    status = call("GET", f"/jobs/{job_id}/status")
    if status.get("terminal") or status.get("status") in ("completed", "failed", "canceled"):
        break
    time.sleep(delay)
    delay = min(delay * 2, 60)

print(status.get("status"))
if status.get("status") == "completed":
    print(json.dumps(call("GET", f"/jobs/{job_id}/result"), indent=2))

What to leave out

Keep the request small. video_url is the edit source and cannot be sent with image_url, end_image_url or any reference_*_urls. Do not send aspect_ratio, which is a 400 on edit, and do not send duration expecting it to cut the clip: the source sets the output. resolution is optional and defaults to 720p.