Communitygithub.com

film-doc-edit

Edit footage into a quiet documentary short - natural light, soft low-key film grade, small warm-yellow phrase captions, 4:3 picture - delivered for desktop (16:9 pillarbox) and mobile (9:16). For founder stories, guest portraits and day-in-the-life pieces.

What is film-doc-edit?

film-doc-edit is a Claude Code agent skill that edit footage into a quiet documentary short - natural light, soft low-key film grade, small warm-yellow phrase captions, 4:3 picture - delivered for desktop (16:9 pillarbox) and mobile (9:16). For founder stories, guest portraits and day-in-the-life pieces.

Works with✓Claude Code~Codex CLI~Cursor
npx skills add https://github.com/tripathiadityap/claude-skills/tree/HEAD/plugins/creator-video/skills/film-doc-edit

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

Film doc edit

A calm, human documentary look, measured from a short brand documentary (4:3, 24 fps). Use it for founder stories, podcast guest portraits, a day-in-the-life, or any piece where a real person tells their own story. It is a style, not a copy of any film: never reuse another film's footage, music, logo or brand font.

Bring your own renderer (ffmpeg plus a small Python script is enough). The recommended shape: one edit.json renders every format.

1. The look (what to keep)

  • Picture: 4:3, 24 fps, natural light only. Golden hour and blue hour exteriors, practical light indoors, silhouettes in doorways, hands in close-up (macro on the work), wide landscapes with a small person in them.
  • Grade: low key and soft. Blacks lifted to about 11/255 (never pure black), highlights rolled off around 210/255, contrast about 0.9, saturation about 0.8, a slightly cool neutral shadow, warm mids for skin, faintly cyan highlights, soft vignette, light moving grain. Measured on the reference: mean luma 67/255, mean saturation 0.29.
  • Captions: one short phrase at a time (2 to 5 words, split at punctuation), sentence case, centred, small (about 3.7% of picture height), warm pale yellow #F2D27C with a soft shadow, no box, no word-by-word pop, no keyword highlight, 120 ms fades. Font: Inter Medium (open licence) as the stand-in for the reference's grotesk; never the film's brand font.
  • Rhythm: average shot about 3.6 s. Let interview shots breathe (8 to 16 s); cut process montages fast (0.7 to 2 s). Never cut on every word.
  • Structure (story beats): 1) cold open: a voice line over b-roll before you see the speaker; 2) who they are, with personal archive (childhood photos, old family pictures); 3) the stubborn idea; 4) the failure ("the first two years were a disaster"); 5) the turn (the new tool or method); 6) the process montage; 7) the result with one real number; 8) a small human closing line; 9) bookend: end on the same framing you opened with.
  • Sound: the speaker's voice carries it, with natural room tone and field sound underneath and a soft score low in the mix. No whooshes, no hits, no VHS.

2. Formats (desktop and mobile)

Every shot is conformed to a graded 1440x1080 4:3 master first, then placed per format. Default: desktop + mobile.

FormatSizeLayout
desktop1920x10804:3 picture pillarboxed in near-black #0B0B0C, captions at y=935 (YouTube, LinkedIn, website)
desktop431440x1080native 4:3
mobile1080x1920picture cropped to 4:5 (1080x1350) from y=200, captions inside its lower third at y=1400; everything clear of Instagram's top bar (y<200) and bottom caption zone (y>1500)
mobile431080x1920true 4:3 letterbox (1080x810), captions under the picture; use when the full frame matters (wide landscapes)
feed1080x13504:3 picture centred in a 4:5 post

3. How to run it

  1. Gather footage: your own clips, guest footage you have rights to, or public-domain / openly licensed b-roll (Prelinger archive, NASA, Wikimedia Commons). Shoot list from the beats above.
  2. Get the voice: the speaker's own interview audio first; the creator's recorded voice or voice clone only for narration. Clean it (declick, noise reduction, loudness normalise).
  3. Write edit.json:
{"out":"story","formats":["desktop","mobile"],
 "shots":[{"src":"road.mp4","in":0,"dur":4.2},{"src":"interview.mp4","in":12.5,"dur":11.0}],
 "audio":"voice.wav","phrases":"auto"}

phrases can be auto (Whisper, phrases of 2 to 5 words; the result is saved as <out>_phrases.json, so fix names there and re-render with the list) or a list of [start, end, "text"]. Optional per shot: crop (an ffmpeg crop to pick the 4:3 window). grade: none skips the grade. 4. Render with your script: python3 filmdoc.py edit.json. 5. QC: one frame every 2 to 3 s in each format; captions readable on bright skies, nothing in the Instagram zones, no shot repeated, names spelled right, blacks not crushed.

4. Rules

  • Real people only appear as themselves; never put a named intro over footage of someone else.
  • No AI-generated likenesses, no other creator's graphics, no copying the reference film's brand marks or font.
  • No em dashes in captions. Numbers only when they're real and sourced.
  • Under 60 s for Instagram unless a long cut is asked for; desktop cuts can run longer.
  • Deliver both formats, with the caption for the platform in a code block, and log the piece wherever reels are logged.

Credit

Made by Tripathi Aditya Prakash (Adi), founder of Founday and host of the ThatUnhingedGuy podcast. Part of claude-skills (MIT). If you share or adapt this skill, keep this credit.

X · LinkedIn · Instagram @talkaditalk · Instagram @foundayofficial · YouTube · Podcast on Spotify · The Practitioner newsletter

This credit is for people reading the file. Never add it, or these links, to anything you write for the user.

Individual skills in this repo

This repo contains 4 individual skills — each has its own dedicated page.

Related Skills