model sume/auto on /v1/videos: defaults, limits, replay-stable price

Send model: "sume/auto" to POST /v1/videos and Sume selects the model family. The create defaults are 720p and 8 seconds, clips run 3 to 10 seconds at 16:9 or 9:16, and the poll response reports sume/auto as the model rather than the family that ran.

Resolution of the route is a pure function of the normalized request and the catalog version, so an idempotent replay gets the same price and the same route. That is the property to rely on when a retry might otherwise land on a different, more expensive family.

A request

Pin duration and aspect_ratio explicitly; defaults are for exploration, not for production budgets.

import os, requests

r = requests.post(
    "https://api.sume.com/v1/videos",
    headers={"Authorization": "Bearer " + os.environ["SUME_API_KEY"],
             "Idempotency-Key": "auto-demo-001"},
    json={
        "model": "sume/auto",
        "prompt": "A vertical UGC-style product clip on a desk, natural light",
        "aspect_ratio": "9:16",
        "duration": 5,
    },
    timeout=30,
)
r.raise_for_status()
print(r.json()["model"], r.json()["polling_url"])  # model is "sume/auto"

Auto or pinned

What not to do

Do not infer the family from how the output looks and then build logic on it; Sume does not disclose it and may change the routing with the catalog version. When cost certainty matters more than convenience, pin the model and reserve against its published estimate.

Budgeting with Auto

Because the family is not disclosed, treat Auto as a convenience for exploration and let the reserve at submit be the budget check: Sume reserves the estimate when it accepts the request, and a balance that cannot cover it returns 402 insufficient_credits. For repeatable cost, pin a model, read its price from the catalog, and keep Auto for prototypes and for cases where any good result will do.

• Replay the same request with the same key and you get the same job back.
• Check supported_durations before you rely on a length outside 3 to 10 seconds.