Production · 3 / 7 · 4 min

Cutting long videos into shorts: how moment detection works

What happens to your video after upload, how the service chooses moments and which settings affect the result.

Clip cutting is the main format for anyone who already has recordings of livestreams, webinars and interviews. This lesson explains what happens inside and how to get more good clips.

What happens after upload

  1. Downloading and checking the source. The service reads duration, resolution and the audio track.
  2. Speech recognition. You get a transcript with exact timing for every word — used both for choosing moments and for subtitles.
  3. Finding the best moments. The whole transcript is analysed to find fragments that make sense without context.
  4. Reframing. Horizontal video is converted to vertical 9:16.
  5. Rendering. Subtitles, banner, filter, music, intro and outro are applied according to your brand settings.
  6. Platform captions. Each clip gets a title, description and hashtags within platform limits.

How moments are chosen

A good clip is a complete thought with a strong opening. Fragments are scored on several signals:

  • A hook in the first seconds: a question, a surprising claim, a number, “the biggest mistake”, “nobody tells you that…”.
  • Completeness: the fragment has a beginning, development and conclusion and doesn’t stop mid-sentence.
  • Independence: the viewer doesn’t need to know what came earlier.
  • Density: few pauses, repetitions and filler.
  • Trend signals: if your trend settings contain topics that are growing now, fragments on those topics get a small boost.

Each clip gets a score from 0 to 100 and a short explanation. The score helps sorting; it is not a guarantee of views.

Job settings

Setting What it does Where to start
Number of clips How many top moments to cut 6–10 per hour of footage
Min and max duration Clip length bounds 20–45 seconds
Language Speech language for recognition Detected automatically
Reframing Centre, smart or blurred background Smart for talking heads
Subtitles Style and word highlight From brand settings
Auto titles Hook text in the first seconds On

Reframing modes

  • Centre — crops the middle of the frame. Fine if the speaker always stays centred.
  • Smart — tracks the speaker and follows them smoothly. Best for interviews and livestreams.
  • Blurred background — fits the whole horizontal frame into vertical with a blurred copy above and below. For slides and screens where nothing may be cut.

How to get more good clips

  • Clean audio matters more than picture. A lavalier or desk microphone gives more accurate recognition and therefore more precise clip boundaries.
  • Answer questions. The question-and-answer format cuts best. In livestreams, read the viewer’s question aloud before answering.
  • Speak in conclusions. Phrases like “so, the key point:” and “three things that…” create natural starting points.
  • Avoid pre-edited footage. If music already runs under the speech, recognition suffers.

If you don’t like the result

Every clip has a “Redo” button for subtitles, titles and descriptions (with your hint), music and filter, or the cover. Only the chosen part is redone. If the moment itself is wrong, reject the clip and run the cut again with different settings — for example, longer fragments.

Limitations

  • Only cut videos you have rights to. Don’t upload other creators’ videos without permission.
  • Source files are kept for a limited time and then deleted automatically; finished clips are kept longer. See the Terms of Use for periods.
  • Overlapping speakers are recognised less accurately; clip boundaries may need a manual check.

Next steps