Help-center article from a product video: one still per step

To turn a product walkthrough into a help-center article, run Video inspect with segmentation.mode: "sentence" for sentence-level timing, choose the sentences that start a step, then pass those start times to Video frames at[] for one still each. Frames are unbilled, and a call takes up to 24 times from a clip of at most 300 seconds.

Sume extracts the pictures and the timed text. It does not write the article, and it does not know which sentence is a step. You pick those, or write the rule.

Why two calls?

Inspect returns stills too, but at times it chooses: 8 mid-bin samples by default, or frames.at[] you supply, resized to a 768 long edge unless you set max_edge. Video frames is the exact-time route, returns source-size images by default and accepts max_edge from 16 to 2160. For documentation screenshots you want legible UI text, so use the frames route and omit the clamp.

The 300-second cap on frames is lower than the 1800 seconds inspect allows. A 12-minute tutorial can be transcribed whole but needs cutting into pieces of at most 300 seconds for frame extraction, which Video trim does.

What does the frames call look like?

Say the transcript shows steps starting at 4.2, 19.8, 41.0 and 63.5 seconds. Request PNG for crisp text. Idempotency-Key is sent on REST so a retry does not queue a second extract, and submit is always 202; this route does not honour mode: sync.

import json, os, time, urllib.request

key = os.environ.get("SUME_API_KEY")
if not key:
    raise SystemExit("set SUME_API_KEY")
H = {"Authorization": f"Bearer {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)
    return json.load(urllib.request.urlopen(req))

start = call("https://api.sume.com/v1/video-frames", {
    "video_url": "https://media.sume.com/artifacts/artf_demo/setup.mp4",
    "at": [4.2, 19.8, 41.0, 63.5], "format": "png"},
    {"Idempotency-Key": "help-setup-frames-1"})
rid = start["request_id"]
while True:
    r = call(f"https://api.sume.com/v1/video-frames/{rid}")
    if r["video_frames"]["resource_status"] == "ready":
        break
    time.sleep(2)
for f in r["video_frames"]["frames"]:
    print(f["t"], f["url"])

The loop has no timeout or failed-status branch, so add both before production use.

What can go wrong?

Is the result ready to publish?

Not without a review. Screenshots show whatever was on screen at that instant, including a half-loaded page or a notification banner. Look at each still, and nudge the time by a second if the UI has not settled. Frames are durable media.sume.com artifacts, so you can reference or download them. Read Jobs and results for the polling model.

Also keep the original recording out of the article if it contains real customer data. Extracting a still does not redact anything.