Webinar to Shorts: pick moments by transcript, not fixed cuts

To turn a webinar into Shorts that do not look mass-produced, choose the moments from the transcript instead of cutting every 60 seconds. Run Video inspect with transcribe: true and segmentation.mode: "sentence", read the sentence segments for complete thoughts, then trim each chosen range and caption it. A fixed slice starts and ends mid-sentence and gives every clip the same rhythm, which is the pattern YouTube's monetization policy calls content that feels interchangeable from video to video (read 2026-10-03).

Step 1: get sentence segments

Video inspect reads one clip on your workspace's media.sume.com. The webinar must be uploaded or imported first, and it must be 1800 seconds or shorter. With transcribe: true it runs Sume STT on the audio at $0.01 per audio minute; omit duration_seconds and the reservation is one minute, and the documented maximum hint is 600 seconds. frames: false skips the stills if you only want the transcript.

Step 2: choose the ranges

Each segment carries start and end times. Read them as candidate boundaries and choose spans that are one complete thought: a question and its answer, a claim and its example. Selection is an editorial or agent step, not an API call. If you use a Sume agent for it, give it the segments and a rule such as one point per clip and no clip over 60 seconds.

Step 3: trim and caption each range

Video trim takes video_url, start and exactly one of end or duration, and returns a new MP4. Frame-accurate precision: "exact" is the default. Then Video captions burns subtitles; leaving style omitted lets the wording decide the look, and you can pass your own cues for a per-clip hook line.

Where this stops

Sume does not choose the best moments for you in this flow; the transcript tells you what was said, not what was good. Confirm current rates in GET /v1/catalog. For the whole pipeline end to end, including delivery, see Webinar recording to Shorts.