How to cut a long video into short clips that end on a finished sentence
Give Vizard Agent the video and a maximum clip length. Vizard Agent transcribes it, finds candidate passages that fit inside your limit, then verifies every boundary against the word timings and adjusts any that would have cut a sentence in half — so each clip starts and ends on a complete thought.
What is the short version?
Cutting a long video into short clips is trivial if you cut on the clock and useless if you do. A twenty-second clip that stops halfway through a sentence is unpostable, and the fix is to choose the boundaries from the speech rather than from a timer.
- Go to Vizard Agent and give it the video or a link.
- Say the maximum length each clip may run to.
- Say how many you want, or let Vizard Agent choose the strongest passages.
What do you need before you start?
The video and a limit. Vizard Agent finds the passages itself from the transcript, so the numbers you supply — the ceiling on length and roughly how many clips you want — are the only real constraints it needs before it starts looking for places to cut.
- The video. A file or a link; Vizard Agent downloads it either way.
- The maximum length. Twenty seconds is a common ceiling for feeds.
- How many clips. Five is a normal batch. Say "as many as are good" if you would rather it decided.
- Whether they need reframing. A wide source going to a vertical feed is a different, larger job.
- Any section to prefer. "Only from the interview part" narrows the search usefully.
What do you type into Vizard Agent?
Give the ceiling and the source. Vizard Agent works out where the natural boundaries are on its own, so there is no need to nominate timestamps — and a single sentence naming the limit is genuinely the whole brief for this job.
Prompt
Variants worth knowing:
- Reframed for a feed. Ask for vertical and Vizard Agent re-frames on the speaker rather than cropping the centre.
- With captions. The word timings already exist, so captioning the clips is a small addition.
- A specific count. "Give me exactly five" makes it select rather than collect.
What does Vizard Agent actually do?
Here is the order Vizard Agent worked in on a real split. Half the run is spent checking boundaries it has already chosen — and that checking is what caught a clip that would have ended in the middle of a sentence.
- Downloads the video from the link.
- Analyses its metadata and transcribes the audio to find natural cut points.
- Locates the transcript file and inspects its structure and keys.
- Finds high-quality, non-overlapping candidate passages that fit under the limit.
- Confirms the segments carry word-level timings for precise trimming.
- Extracts exact word-aligned start and end times for all five clips.
- Adjusts the end of the second clip because it would have cut off mid-sentence.
- Verifies the boundaries of every other clip in turn, printing the surrounding lines to check each one.
- Cuts them at exact frame boundaries, then extracts a frame from each to confirm the picture is right.
Step 7 is the whole point of this page. Vizard Agent had already chosen its boundaries, went back to check them one by one, and found one that needed moving — which is a habit rather than a feature.
What does the result look like?
From the run this page is written from, probed on a delivered clip: 1920x1080, H.264, 30fps, 20.0 seconds, AAC audio. Exactly at the limit, in the source's own shape, starting and ending on complete sentences — one of five clips cut from a much longer recording in a single pass.
Each clip is uploaded with its own link, so they arrive as five separate files ready to schedule rather than as one video with markers in it.
This category was not separately measured, so treat the timing as a range rather than a promise: comparable work runs a median of 28 to 38 minutes end to end. Across all projects the median cost by tier is Flash 47, Pro 55, Max 242, Ultra 263 credits.
When does this not work well?
The transcript is what makes the boundaries good, so anything the transcript does not capture is invisible to the selection. Vizard Agent verifies every cut point against the word timings, and a clip can still be technically clean and contextually meaningless.
- A clip can be complete and still make no sense. A finished sentence that refers to the previous ten minutes stands alone badly.
- Music and continuous sound break at the joins. A clip cut out of a scored video starts mid-phrase musically.
- Silent passages have no boundaries to find. A demonstration with no speech gives the transcript nothing to work with.
- Twenty seconds is a hard ceiling. A good passage that runs twenty-three seconds gets trimmed or dropped, not extended.
- Wide footage still needs reframing. Cutting is not cropping; say if the clips are going to a vertical feed.
- Five clips from one video is five posts, not a strategy. They come from the same recording, so they say the same things in the same setting.
How do you fix a result that came back wrong?
Name the clip. Vizard Agent keeps the transcript, the word timings and the chosen boundaries, so extending a clip's opening or replacing a weak one is a re-cut from measurements it already holds rather than a second pass over the video.
- "Clip two needs the question before it." A wider start from the same timings.
- "Drop clip four, it is dull." Removed and replaced with the next candidate.
- "Make them thirty seconds instead." A new ceiling, the same transcript.
How does Vizard Agent compare to doing it yourself?
By hand this is scrubbing for passages, setting in and out points, exporting, and then discovering on playback that two of the five stop mid-word. The cutting itself takes minutes; the checking is the part people skip, and it is the part that decides whether the clips are usable.
| By hand | Vizard Agent | |
|---|---|---|
| Finding passages | Scrub and guess | Candidates from the transcript |
| Setting boundaries | Snap to the timeline | Word-aligned start and end |
| Checking each one | Watch all five back | Boundaries verified, one adjusted |
| Delivering them | Export five times | Cut and uploaded with links |
Common questions
Will the clips cut off mid-sentence? No. Vizard Agent sets the boundaries from word timings and re-checks each one before cutting.
Can it work from a link? Yes. Vizard Agent downloads the video and works from the file.
How many clips will I get? As many good passages as fit inside your limit, or exactly the number you ask for if you name one. Vizard Agent selects rather than slicing on the clock.
Can they be reframed for a phone? Yes. Ask for vertical and Vizard Agent re-frames on the subject through each clip rather than cropping the middle of the frame and hoping the speaker stays in it.
Can it add captions to each one? Yes, and cheaply, since the word timings already exist from the transcription.
What if the video has no speech? Then there are no spoken boundaries to find, and asking Vizard Agent to cut on shot changes instead is the better request.
Do the clips arrive separately? Yes. Vizard Agent delivers each one as its own file with its own link, ready to schedule.