AI column

Video@donghyeok99Staff

Cinematic Skill: narrated films drawn entirely in code

Ask for "a 90 second documentary about the Wright brothers" and this skill writes, voices and draws it in code

A date lands on a cork wall, red string runs from a map pin to the first-flight photo, then counters for 12 seconds and 120 feet rise on the voice.
A date lands on a cork wall, red string runs from a map pin to the first-flight photo, then counters for 12 seconds and 120 feet rise on the voice.
Who
Made by the creator who posts as exploraX on X (GitHub Moh4696), for people who want documentaries and explainers without shooting footage.
When
The repository opened on November 9, 2026; versions 1.0.0 and 1.0.1 came out the same day, per the changelog.
Where
Installs into Claude Code, Codex, Cursor, Gemini CLI and OpenCode. Rendering runs on your own computer with Python and ffmpeg.

What was hard

An explainer or a short documentary used to mean writing a script, recording a voice, hunting for photos you are allowed to use, and nudging every cut in an editor until it met the words. Handing it to a video model gave a different film on every run, so fixing one shot was hard.

What it does about it

This skill treats a film as one big board. Photos, documents, cards, maps and numbers all sit on it, and a single camera glides between them without a cut. Code draws every frame (Skia, in Python), so the same scene file always gives the same film, and a stamp or a highlighter lands on the exact word the narrator says. Photos come from Wikimedia Commons with their licences recorded, and non-free files are refused.

The example's cold open: date, map pin, first-flight photo, counters
The example's cold open: date, map pin, first-flight photo, counters
The final pull-back: map, portraits, a letter and stamps on one wall
The final pull-back: map, portraits, a letter and stamps on one wall
The card-based explainer every new project starts as
The card-based explainer every new project starts as

How it runs inside

  1. 1

    You → Claude Code

    "a 90 second documentary about the Wright brothers"You give a subject and a length; the skill is called in.
  2. 2

    Claude Code → cinematic-film

    script.json → make_beats.py → beats.mdIt writes the script, lints chronology and line length, and hands you a beat table to approve.skills/cinematic-film/scripts/make_beats.py
  3. 3

    cinematic-film → audio/timing.json

    tts_say.py → build_narration.pyA scratch voice is cut into lines and joined into narration.m4a with word timings in timing.json.skills/cinematic-film/scripts/build_narration.py
  4. 4

    Claude Code → engine.py · render.py

    scene.py: doc.W(line, "word")Photos, maps and counters go on the time a word is spoken, with camera and spotlight keys.skills/cinematic-film/assets/wright-flyer/scene.py
  5. 5

    audio/timing.json → engine.py · render.py

    timing.jsonEvery entrance comes from the voice, so a new voice re-times the whole film.skills/cinematic-film/scripts/engine.py
  6. 6

    engine.py · render.py → MP4 · SRT · credits.md

    render.py --out out/draft.mp4Stills are checked first, then a parallel render, with sound effects and captions added.skills/cinematic-film/scripts/render.py

What goes in, what comes out

a 90 second documentary about the Wright brothers"Twelve Seconds", the shipped example: about 90 seconds from seven public-domain photos
new_project.py my-film"From Cherry to Cup", a 25 second explainer of cards, arrows and a counter to start from

Is it for you

Good for

  • A one or two minute documentary about history, a person or a case, told with photos and maps
  • A launch film or founder story built from your own product shots
  • Explaining how something works with cards and rolling counters
  • Changing the voice or a sentence and having the film re-time itself

Not for

  • It is not generated video: no live-action motion, only a camera moving over photos and graphics.
  • Per the README, a 92 second film renders at 1080p in about 6 minutes on a 16 GB laptop.
  • The free scratch voice is macOS say or espeak-ng; the final voice is another TTS or a person.
  • It stops for your approval at the beat table, the look and a 30 second draft.

Reviewed by Devalley’s editor before publishing. How the column is made

Sign in to post, comment and like.Sign in

Report

Reason

Read the community rules