TikTok Motion Factory (Non-AI Remotion Pipeline)

A fully automated, deterministic TikTok motion graphics pipeline using Remotion + signal processing only. No AI is used anywhere in the workflow.

Quickstart (<=10 commands)

1. git clone <this-repo>
2. cd video-workflow
3. python3 -m pip install -r tools/requirements.txt
4. npm install
5. cd remotion && npm install && cd ..
6. cp /path/to/your/audio.mp3 input/
7. ./run.sh --dur 10 --title "My Title"
8. open assets/renders/<job>.mp4

Smoke Test

This will fail gracefully if there is no audio in input/:

./run.sh --dur 5

Workflow

./run.sh performs:

1. Dependency checks (tools/doctor.sh)
2. Audio prep (trim + loudness normalization -> wav)
3. Cue detection (librosa onset strength + peak picking)
4. Props generation for Remotion
5. Headless render to vertical 1080x1920 MP4
6. Writes job notes and hashes

Usage

./run.sh --audio input/myfile.mp3 --start 00:12.0 --dur 10 --title "My Title" --slug my_slug

• If --audio is omitted, the newest file in input/ is used.
• Output MP4 goes to assets/renders/<job>.mp4
• Job folder and logs go to jobs/<job>/

Codex Skill

This repository is the Codex skill directory (root). The canonical skill file is SKILL.md. For compatibility, a copy also lives at .codex/skills/tiktok-motion-factory/SKILL.md.

Example invocation:

$tiktok-motion-factory

The skill will ask for an audio path (or use the newest file in input/), then run ./run.sh end-to-end.

Repository Layout

.codex/skills/tiktok-motion-factory/SKILL.md
remotion/
  src/Root.tsx
  src/index.ts
tools/
  prep_audio.sh
  analyze_cues.py
  make_props.ts
  doctor.sh
input/
assets/
  audio/
  renders/
jobs/

Notes

• No downloading from TikTok/Spotify/YouTube/etc. Provide local audio only.
• Compatible with macOS and Linux.
• If you pass a URL, the pipeline will refuse and ask for a local file.