Skip to content
All work

2026 · Design, engineering, packaging

Video Maker AI

A desktop pipeline that turns a topic and a runtime into a finished video: script, neural voiceover, generated frames, word aligned captions and karaoke highlighting.

A closer look

Video Maker AI

A video pipeline that lives in a terminal

Repository

Video-Maker-AI

pushed 12 Nov 2026

Commits in the last year

4

Language bytes

  • Python256k
  • JavaScript64k
  • HTML35k
  • Shell6k
  • Batchfile3k
  • PowerShell3k

The problem

Making a narrated video means moving between a script editor, a voice tool, an image generator, a subtitle editor and an encoder. Each one has its own export format, and the timings only line up in the last tool, by hand.

What I built

  1. 01

    A single control panel launched from one file scans the workspace and reports what is ready: which audio file, how many images, whether the script exists and whether FFmpeg is found.

  2. 02

    The script stage asks for a topic and an exact runtime, then writes narration to that length rather than truncating afterwards.

  3. 03

    WhisperX transcribes the voiceover and forces alignment against the script, which gives every word a start and end time.

  4. 04

    Captions are burned in with the spoken word highlighted in yellow, so timing comes from the audio rather than from a guess.

  5. 05

    Images come from either a CLI generator or a bulk Flow session, then a sanitiser renames them to a clean sequence so the track order is unambiguous.

  6. 06

    The render step applies Ken Burns motion per frame, letterboxes mixed aspect ratios and encodes at the chosen shape for landscape, shorts or square.

What it changed

  • One launcher replaces five tools and their export formats.

  • Captions are word aligned by construction, not adjusted by hand.

  • Every stage can be rerun alone, so a wrong frame costs seconds rather than a whole render.