The quick answer: one long file to a folder of vertical clips
You recorded forty minutes, posted one clip from it, and the rest sits on a drive. Cutting it up feels like a second job. The Clip Finder does the hunting: it scores the best 15, 30 or 60 second moments in the file and shows you why. The Video Repurposer then turns each pick into a vertical video with captions. Your video is never uploaded.
- Find the clips. Drop the recording into the Clip Finder, choose Transcript scan, pick 15, 30 or 60 seconds and press Find the best clips.
- Review them. Each card shows a time range, a score, the reasons and part of the transcript. Press Preview and adjust to nudge the start and end, then tick the ones you would really post.
- Export or hand off. Download the ticked clips, or press Make it vertical to send one straight to the Repurposer.
- Reframe and caption. Choose 9:16, keep Follow faces on, leave captions on and create the video.
- Check the safe zones. Scrub the result in the safe zone checker, then post.
Step 1: choose moments worth posting
No hidden model decides what is good. The Clip Finder builds many candidate windows about as long as you asked, starts and ends each at a sentence boundary, and scores it out of 100 with signals you can read:
- Opening line, 25 points. A question, a number, or a hook word such as secret or how to. Starting on so or but counts against it.
- Speech density, 20 points. Words per second compared with the rest of the file.
- Energy and peaks, 20 points. How loud and emphatic the voice is, again compared with the rest of the file.
- Few long pauses, 20 points. Pauses of nearly a second or more cost points.
- Clean ending, 15 points. A clip that ends on a complete sentence scores in full.
The detail that surprises people: the speech density and energy signals compare a clip with the other windows in the same file. A 78 in one recording says little about how it compares with a 78 in another, and no score predicts views. Hook words are matched in English only, while question marks and digits count in every language the speech model supports.
Use a Transcript scan whenever there is speech. The first one downloads a speech model (from about 41 MB) that your browser keeps. A Quick scan only listens, so it needs no download and finishes sooner, but it cannot judge what is said. It suits music and very long files.
Then use your own judgment: a clip that stands alone beats a higher scoring one that needs context. The top picks never overlap, so you will not get ten versions of the same great five minutes.
Fast exports are quickest, but a copy without re-encoding must begin on a keyframe, so a clip can start slightly early. Choose Precise when the exact start matters, and fine-tune any clip in Trim Video.
Step 2 (talking heads only): tighten each clip
If a clip is someone talking to camera, run it through the Auto Editor. It cuts pauses and the filler words a speech model can find (um, uh and a few more), alternates between full frame and a 110% zoom at each cut so jump cuts read as deliberate, and can level the sound to -14 LUFS. Every cut shows on a timeline first, so click any you want to keep.
Two cautions. It misses some ums, because speech models often leave them out of the transcript, so play the edit with Skip removed parts on. And leave its captions off: burned in now, they sit on a landscape frame that the next step crops. Caption once, at the end. The full method is in removing pauses from talking videos. Run this on the short clip, not the forty minute file, and skip it for interviews and screen recordings. Recording fresh talking-head clips rather than mining an old file? Reading to camera with a teleprompter means fewer pauses and ums to cut afterward.
Step 3: vertical framing that keeps the face in shot
A 16:9 video cut to 9:16 loses about two thirds of the picture, so the crop position matters (the numbers are in aspect ratios explained). With Follow faces on, the Repurposer samples about four small frames per second, finds faces with an open-source detector that runs in your browser and follows the largest, steadiest one, easing the crop so it does not jitter. If no face fits comfortably, it keeps the whole picture over a blurred background and tells you.
| Footage | Setting to try | Why |
|---|---|---|
| One person talking to camera | Follow faces | The crop tracks the face |
| Two people, or a panel | Fit with blurred background | It follows one face and does not know who is speaking |
| Slides or a screen recording | Blurred background, zoomed 120 to 140% in Resize for Social | Zooming makes text bigger on a phone but crops the sides, so check the edges |
Tick only the shapes you will post. Each shape is rendered separately, one after another, so three take longer than one, although finding faces and transcribing only happen once. Sizes and current platform limits are in the guide to vertical video.
Step 4: captions you have proofread
With captions on, the Repurposer transcribes the clip on your device and burns in bold captions that highlight each word as it is spoken, sized and placed for each shape. You also get an SRT file. The speech model downloads once (about 77 MB on a processor, 206 MB on a graphics card, at the default accuracy).
Speech recognition still stumbles on names and jargon. To fix a word before it is burned in, switch captions off in the Repurposer, open the vertical MP4 in Auto Subtitles, which has a full editor, correct the lines and export from there. Style advice is in captions people actually read, and the whole subtitle workflow is in adding subtitles to a video for free.
Step 5: check the safe zones
Each app draws its own buttons and text over your video. Open the finished file in the safe zone checker with All platforms selected. On a 1080 by 1920 frame, that combined safe area runs from about 270 to about 1250 pixels from the top, roughly 14% to 65% of the height.
The Repurposer places captions about 70% of the way down, lower than the bottom of that combined area. That may be fine for your app, but look. If captions touch a zone, run the Repurposer again with captions off, caption that vertical MP4 in Auto Subtitles and raise them with the Position slider. The zones are approximations, so leave margin.
How long it takes (an estimate)
These are planning numbers for ten clips from a forty minute talking video, not benchmarks. A graphics card speeds up transcription, and a processor alone is slower.
| Stage | What you do | Rough time |
|---|---|---|
| Scan | Start it and leave it. A quick scan takes a small fraction of the recording's length, a transcript scan longer | Mostly waiting |
| Review | Preview each card, nudge the edges, tick the keepers | 15 to 25 minutes |
| Tighten (optional) | Auto Editor on each talking clip | 2 to 4 minutes a clip |
| Reframe, caption, check | One Repurposer run per clip, proofreading, one safe zone scrub | 4 to 7 minutes a clip |
Expect around two hours the first time and nearer one hour with a routine, much of it waiting rather than working.
Podcasts, and what goes wrong
The Clip Finder also opens MP3, M4A, WAV and FLAC files. For a vertical video from audio, Podcast Studio makes 9:16 audiogram clips with optional captions. The episode workflow is in making a podcast in one browser tab.
- The crop follows the wrong person or misses a small face. It tracks one face at a time. Pick Center or blurred background instead.
- Captions twice, or captions cut off. They were burned in at an earlier step. Turn them on in the last one only.
- A clip opens with a stray half second. That is a Fast cut starting on the previous keyframe. Re-export it as Precise.
- The file is too big for where you post. Run it through the video compressor last.