Veo 3.1 deletes clips after 2 days: archive them, and the Sume path

Veo 3.1 deletes generated videos 2 days after they are made. Google's Veo guide says: "Generated videos are stored on the server for 2 days, after which they are removed." A pipeline that stores only the Google file reference will lose clips on day 3, so download each clip when the job completes and keep your own copy.

Sume does not carry Veo 3.1. The Sume video catalog has one Google video model, gemini-omni-flash-1.1, and it follows a different download path (below).

What Google documents for Veo 3.1

The 2-day window is the only retention figure in the Veo guide. Extension is the one feature that moves it.

The archive pattern

Treat the Google file as a handoff, not storage. When the job finishes, copy the bytes to your own bucket in the same worker, and save your own object key in your database. Do not save only the Google reference.

If you extend a clip, Google resets the timer on the extended video. Still archive every clip you want to keep, including the intermediate ones, as you make them.

The same step on Sume

Sume video jobs are asynchronous. You submit to POST /v1/videos, poll GET /v1/videos/{id}, and read unsigned_urls when the status is completed. Each URL is the content endpoint, GET /v1/videos/{id}/content?index=0. The Sume docs send the API key as a Bearer header on that call, and the script below does the same.

import asyncio, os, requests

H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}

async def main():
    job_id = "job_01HXYZ"  # the id from POST /v1/videos
    url = f"https://api.sume.com/v1/videos/{job_id}"
    while True:
        r = await asyncio.to_thread(requests.get, url, headers=H)
        job = r.json()
        if job["status"] == "completed":
            break
        if job["status"] in ("failed", "cancelled"):
            raise SystemExit(job.get("error", job["status"]))
        await asyncio.sleep(30)
    clip = await asyncio.to_thread(requests.get, job["unsigned_urls"][0], headers=H)
    with open(f"{job_id}.mp4", "wb") as f:
        f.write(clip.content)

asyncio.run(main())

What differs between the two

The Sume docs do not state how long a generated clip stays available. The docs do say that expired is a status in the enum for wire compatibility, but Sume does not expire jobs now, so the API never emits it. Treat the retention window as unspecified and copy the file to your own storage in either case.