Batch the shots of a micro-drama episode in Python, safe to re-run

Give every shot a stable Idempotency-Key built from season, episode, shot number and a prompt version, and a re-run of your script returns the same job instead of creating a second paid one. The script below submits three 30 second Seedance 2.5 shots for one episode and writes each job id to a ledger file.

Episodic series are the shape that Metricool says TikTok's The Next Episode program is aimed at, so a repeatable batch is worth setting up.

The script

It uses plain requests, so no async code. POST /v1/videos returns the job envelope with id, polling_url and status. Set SUME_API_KEY first. Bump the v1 in the key when you intentionally change a prompt.

import os, json, requests
API = "https://api.sume.com/v1/videos"
HEAD = {"Authorization": "Bearer " + os.environ["SUME_API_KEY"]}
SHOTS = [
    "Rain on a night bus, a woman reads a message and goes pale",
    "She steps off the bus and sees a man waiting under a streetlight",
    "He holds out her lost phone, screen lit, a call incoming",
]
def submit(ep, n, prompt):
    body = {"model": "seedance-2.5", "prompt": prompt, "duration": 30,
            "resolution": "720p", "aspect_ratio": "9:16"}
    key = "s1-e%d-shot%d-v1" % (ep, n)
    r = requests.post(API, json=body, timeout=60,
                      headers={**HEAD, "Idempotency-Key": key})
    r.raise_for_status()
    return r.json()["id"]
ledger = {}
for n, prompt in enumerate(SHOTS, 1):
    ledger[n] = submit(1, n, prompt)
    print("shot", n, ledger[n])
json.dump(ledger, open("ep1-ledger.json", "w"))

What the keys buy you

Writes accept an Idempotency-Key so a retry of the same request is not treated as a new one. This is the habit the jobs page asks for: do not resubmit a paid request just because a local process timed out. Keep the key stable per shot and change it only when you mean to pay again.

Cost of the batch

Three 30 second shots at 720p and 9:16 are 90 seconds at $0.5778 per second, so $52.00. Each shot is its own job, and workspace concurrency limits apply when workers pick them up.

Next steps and limits

Poll the returned polling_url (GET /v1/videos/{id}) or GET /v1/jobs/{id}/status with backoff until a terminal status, then read the result for the video URL. For push instead of polling, pass a callback_url or see the webhooks guide. The script has no retry logic on purpose: a 4xx should stop and be read.

It does not pass first frames; add a frame_images entry per shot if you chain from a still.