kings video-edit turns raw phone or smart-glasses clips into one vertical video for Stories or
Reels: clips joined in recording order, sped up, and captions transcribed on the machine and
burned in. Nothing is cut — boring stretches can be fast-forwarded instead, and a long result can
be split into parts that fit a platform's length limit.
# Files, folders, Samsung Quick Share links or shared OneDrive folders — clips are sorted by recording time
kings video-edit ~/Downloads/clip1.mp4 ~/Downloads/clip2.mp4 --name Trip
kings video-edit https://quickshare.samsungcloud.com/AbC123 --name Trip --preview
# Fast-forward a walk in clip 1 and split into parts of up to a minute (Stories)
kings video-edit chalet.mp4 --name Chalet --fast 1:0:12-0:40x6 --max-length 60A shared OneDrive (personal) folder is read without signing in, through the anonymous token the OneDrive web page itself uses for visitors — undocumented, so it may change. OneDrive serves each connection at well under 1 MB/s, so files come down as 16 parallel byte ranges.
To turn one download into several videos, list it once and pass the files from its .work/raw/
to separate projects — their transcripts are reused, not redone.
The project lands in ~/Desktop/<name>/ (KINGS_VIDEO_DIR in config.env changes the base):
<name>.mp4, <name>-preview.mp4 with --preview, and a .work/ folder with the downloads,
transcripts and caption images. The caption blocks are printed with their times, for review.
The first run creates a virtualenv in ~/.kingsScripts/venvs/video and downloads the Whisper model
(mlx-community/whisper-large-v3-turbo, about 1.6 GB).
| Option | Meaning |
|---|---|
--name NAME |
Project folder and file name (required) |
--speed X |
Playback speed, default 1.25 — audio pitch is kept |
--lang CODE |
Spoken language for Whisper, default pt |
--no-captions |
Join and speed up only |
--mute |
No audio and no captions — for footage that gets music when posted (combine with --speed 1 to keep the real pace) |
--fast CLIP:START-END[xF] |
Fast-forwards that stretch of clip number CLIP (from 1, as --list numbers them) at Fx, default 4; silent and without captions. Times in source seconds or m:ss. Repeatable |
--max-length S |
Splits the result into <name>-partN.mp4 (replacing <name>.mp4) of similar length, none over S seconds, each cut in a pause before a caption block |
--only PATTERN |
From a OneDrive folder, downloads only the files whose name matches (20261104_22*); repeatable |
--list |
Only prints the clips in order, with recording date and length |
--transcribe-only |
Stop after writing the transcripts, to fix them before the first render |
--preview |
Also write a 720p copy under 30 MB |
--out DIR |
Base folder instead of KINGS_VIDEO_DIR |
Each clip gets .work/<clip>.words.json, one word per line with its time:
[
{"start": 0.0, "end": 0.42, "text": "Esse"},
{"start": 0.42, "end": 0.61, "text": "daqui"},Fix a misheard word in text, or delete the lines of a sentence Whisper invented at the end of a
clip (a common hallucination over silence). Then run with just the name — the clip list and the
transcripts are reused, only the render runs again:
kings video-edit --name TripDeleting a .words.json transcribes that clip again.
| Frame | 1080×1920, center crop of the source, 30 fps |
| Video | H.264 high, CRF 20 |
| Audio | AAC 160 kbps: 80 Hz high-pass and a mild FFT denoise, then one fixed gain to −16 LUFS with a peak limiter — no dynamic normalization, which would lift the hiss in the pauses |
| Fast-forward | Same picture pipeline, audio silenced, no captions |
| Captions | Up to two lines per block, white bold with black outline, lower third; a block ends at a pause over 0.8 s or a sentence end, and stays up across pauses under 0.6 s |
Captions are full-frame PNGs overlaid through ffmpeg's concat demuxer, so ffmpeg builds without
libass or drawtext (Homebrew's default) work.