Skip to content
Cameron Mills

All projects

A video and design engine that turns sources, a script and a voice into a finished short

Digital media system

Sources, script, voice and timing drive a beat-aware edit engine that makes the short films.

Its parts

Dashed lines join each part to Digital media system. Arrows run from the part that feeds to the part it feeds, labelled with what passes between them.

  • Sources: Feeds and pages read into a ledger where every number is checked. Feeds Green Label content engine (sources).
  • Green Label content engine: One sourced spec in, a newsletter, a post and a short video out. Feeds Voice (script), Captions and export (video). Fed by Sources (sources).
  • Voice: Narration on my own machine, with room for a recorded voice. Feeds Timing (voice). Fed by Green Label content engine (script), One Paper Further (narration).
  • Timing: Word times from the voice and a beat grid from the music. Feeds Design library (timing). Fed by Voice (voice), Fontainebleau ad (beat grid).
  • Design library: A library of learned edit styles and the engine that applies them. Feeds Captions and export (edit), Fontainebleau ad (style). Fed by Timing (timing), Creative apps (visuals).
  • Creative apps: Open-source photo, vector, layout, video and motion apps my agents can drive. Feeds Design library (visuals).
  • Captions and export: Captions drawn frame by frame, then one encode and a self-check. Fed by Green Label content engine (video), Design library (edit).
  • Fontainebleau ad: An ad cut to a reference clip's beats, with every sound synthesised. Feeds Timing (beat grid). Fed by Design library (style).
  • One Paper Further: A short film that explains one AI research paper, and where it stops. Feeds Voice (narration).

The film it made about itself

How it works

How sources, a script and a voice become a finished short, with the edit following the audio. Select a part to open it.

Problem

A short film needs many separate jobs done well: sourcing, a script, a voice, timing, pictures, an edit, captions and an export. Doing each by hand each time is slow, and an editor's plugins are not something I can rerun or review.

Approach

Treat each job as a stage with a file in and a file out, and let the edit follow the audio. The engine takes any clips and any song and returns a video in one of the learned styles, with cuts, moves, transitions, sound effects and captions placed on the beats.

Architecture

Each stage hands the next a file. Sources go to a ledger, the ledger to a script, the script to a voice, the voice to word times, and the word times and the beat grid to the edit plan, which the renderer turns into pictures, captions and sound.

What I built

The design library holds the styles and the edit engine. The Green Label content engine turns one sourced spec into a newsletter, a post and a vertical video, and nothing is published without my review. The Fontainebleau ad took its cuts from a reference clip and used synthesised sound effects. Open-source creative apps supply photos, vectors, layouts and motion.

Results

One engine now makes both the Green Label weekly briefing video and the Fontainebleau ad, and the engine has made an explainer film about itself.

Stack

  • Python
  • ffmpeg
  • Pillow

How these figures are sourced

The outcome figure is agent-reported: it is taken from a test log or report written by the agents that built and ran the system. The status was read from the design library and Green Label repositories. Every figure and claim on this page is tied to a source file or a CV line in the site's content.