Find the moment before you touch the timeline
Read the transcript, not the waveform. A strong short-form clip is almost always a complete thought: a question and its answer, a claim and the reason for it, a story with a beginning and an end. Scrub through looking for those, and mark the timestamp where the thought STARTS — usually a few seconds earlier than you think, because the sentence that sets it up is what makes the payoff land. If you cannot summarise the clip in one sentence, it is not a clip yet.
Cut on the thought, not on a duration
Fixed-length cutting is why most auto-generated clips feel wrong. Sixty seconds is a limit, not a target. A clip that makes its point in twenty-two seconds and stops is stronger than the same point padded to fifty-nine. Let the end of the idea decide the end of the clip, then check the first three seconds separately: that is where someone decides whether to keep watching, and a clip that opens on "…and so what I was saying is" has already lost.
Crop to the speaker, not to the centre
A 16:9 recording cropped straight down the middle to 9:16 frequently lands on an empty lectern while the speaker stands to one side. Crop to where the face actually is, and hold that framing rather than letting it drift. If the source is a two-camera edit, check the crop across a shot change — the framing that works on the wide shot usually does not work on the close-up.
Captions are not optional and they are not an afterthought
Most short-form is watched muted, so the captions ARE the video for a large share of viewers. Burn them in rather than relying on platform auto-captions: auto-captions are inconsistent, they are styled by the platform rather than by you, and on a lecture they mishear names and terminology constantly. Time them to the words as spoken, and keep them inside the safe zone so a platform button does not sit on top of the line.
What DeenClipped automates here
It transcribes your selected stretch with word-level timings, has a self-hosted model score the candidate moments and say why it picked each one, crops to the speaker with face detection, burns the captions in, and puts every clip in a review queue for you to approve or reject. It does not remove the judgement — you still decide what publishes — it removes the scrubbing.
Questions
How long should a Short be?
As long as the idea takes and no longer. The platform limits are the ceiling, not the goal. A complete twenty-second point outperforms the same point stretched to a minute, because watch-through matters more than duration and a padded ending is where people leave.
Should I process the whole video?
Usually not. Most long recordings have a few minutes worth clipping and a lot of introduction, questions and repetition. Picking the stretch first saves you the review time and, in DeenClipped, the tokens: you are charged per source minute you select, not per minute of the file.
Do I need to re-edit for each platform?
Not if you keep captions inside a conservative safe zone. The platforms cover different parts of the frame with their own interface, so text that clears the worst case on each edge survives on all of them. See the safe zones guide for where those edges are.
Why do my auto-generated clips feel like they start in the middle?
Because they were cut on a timer rather than on the transcript. Any tool that slices at fixed intervals will land mid-sentence most of the time. The fix is to cut where a thought begins and ends, which requires reading the words — either yourself or with something that scores moments from the transcript.
Basic includes the whole workflow.
Import a source, generate clips, review every one and publish to your own connected channels. Upgrade when you need more.