Episode chapter timestamps and Reel cutdowns from one transcript

To make a chapter list and Reel cutdowns from one long episode, run Sume video inspect with transcribe: true and sentence segmentation, read the timestamps, mark where each topic starts, then cut each short clip with video trim. Instagram's own episodic series does the first half of this: its announcement page lists timestamped segments for key discussion points.

The format example comes from Instagram's Close Friends Only announcement, read on 2026-10-03. It is about a specific show, so only the structure, not the content, is used here.

What does Instagram's example show?

The page says the series is Instagram's original show and that an episode streams on Instagram, YouTube, Spotify, iHeartRadio and Apple Podcasts. It lists timestamped segments highlighting key discussion points, with times ranging from 2:46 to 15:00, which makes a long episode easy to navigate. That is the pattern: one long file, a chapter list, and the same material on several platforms.

The page does not say how Instagram built the timestamps, and it gives no Reel specs or performance data.

How do I get timestamps from a transcript?

Video inspect reads one clip on the Sume media host (source up to 1800 seconds). With transcribe: true it runs speech-to-text at $0.01 per audio minute per the docs, and segmentation.mode: "sentence" adds gapless sentence segments. The duration_seconds hint is capped at 600 seconds; if you omit it, 1 minute is reserved. For an episode longer than 10 minutes, set the hint close to the real length and confirm live pricing in GET /v1/catalog.

Check probe.has_audio first. A silent clip fails with inspect_source_has_no_audio. Frame sampling can be switched off with frames: false when you only need the transcript.

curl -X POST https://api.sume.com/v1/video-inspect \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: episode-12-inspect" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/artf_demo/episode.mp4",
    "frames": false,
    "transcribe": true,
    "language_code": "en",
    "duration_seconds": 600,
    "segmentation": { "mode": "sentence" }
  }'

How do I turn the list into chapters and Reels?

Read the sentence segments and write down the start time of each new topic. Those start times are your chapter list; you write the titles yourself, since Sume's inspect route returns transcript and stills, not topic labels. Then pick two or three strong passages for Reels.

What are the limits?

Video trim returns at most 900 seconds per output and accepts a source up to 1800 seconds, so a long episode must be split. Audio detach returns the audio of one hosted video as a durable wav by default, or mp3. Timestamps from the transcript follow the spoken words, so check any cut by watching it before you post.

Sume does not publish the chapter list to YouTube, Spotify or Instagram. Add it to each platform yourself. Poll every job as described in Jobs and results.