claude-video-skills

Video-editing automation for Claude Code, built on Remotion. Drop a pre-cut clip (or a voiceover + idea) into projects//raw/, tell Claude what you’re making, and it takes it from there - transcription, a beat-timed graphics plan, captions, b-roll, zooms, and a full audio score, end to end.

Start here: /make-video

The unified entry point is the make-video router skill (.claude/skills/make-video/SKILL.md). Invoke it (or just ask Claude Code to “make a video”) and it asks what you’re making, then runs the right pipeline end-to-end:

Per-video runbook: EXECUTION.md · install + brand setup: SETUP.md · session state: WORKLOG.md · agent conventions: CLAUDE.md.

Not one brand - yours

This isn’t skinned to a single creator’s look. Routes A and B ship with an example “Visual OS” preset so they render something reasonable out of the box, but before your first real render, run the Brand setup flow in SETUP.md: a short back-and-forth (primary color, a second accent, fonts, anything to avoid) that Claude uses to derive a full, coherent token set and write it into the templates - once, not per video.

---

Install

# system tools
brew install node ffmpeg

# python env (shared across all engines)
python3.11 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

# secrets - copy the template and add your key(s)
cp .env.example .env
# then edit .env and set ONE of:
#   ASSEMBLYAI_API_KEY=your_key_here
#   OPENAI_API_KEY=your_key_here

All external-tool keys go in the repo-root .env file (never in code, never committed - .env is gitignored, .env.example is the committed template). Every script loads it automatically, so you don’t export anything in your shell. Full install (including each engine’s own npm install) and the one-time brand setup: SETUP.md.

Transcription options

• Cloud (default): AssemblyAI or OpenAI Whisper - fast, no heavy install, needs one API key. AssemblyAI tags filler words with timestamps. Audio is uploaded to the provider, so avoid this for sensitive clips.
• Alternative: WhisperX - fully local, no API/key, but installs PyTorch and runs on CPU (slower on Mac). Uncomment whisperx in requirements.txt.

Faceless motion-graphics reels (motion-design-edit)

For reels with no camera at all - cream paper grid, punch-color cards, taped proof-card shots, timed to a voiceover - use the motion-design-edit skill (.claude/skills/motion-design-edit/). It’s a separate, self-contained skill (own Remotion project, own render pipeline) built for the “paper-reel” style. Pair it with reel-script (.claude/skills/reel-script/) to draft the voiceover script first, in the matching hook/credibility/reframe/ numbered-tips/CTA structure.

See .claude/skills/motion-design-edit/knowledge/paper_style.md for the full design system (colors, the three graphic kinds - pr_card / pr_punch / pr_proof - and the 5-part story structure) and the skill’s own SKILL.md for the render pipeline. Full step-by-step workflow: see the “Faceless motion-graphics workflow” section in EXECUTION.md; install details in SETUP.md.

Supports two selectable visual styles per project (Paper Brand - the original cream/ink/orange reference-reel look, default - or your configured Visual-OS-style brand) and a recurring-series pattern (STYLE.md creative contract + the shared scripts/series_pipeline.sh) for projects that ship more than one episode - see EXECUTION.md’s “Visual style” and “Recurring series” subsections.

Assets - none ship, and nothing breaks

This repo is code only. No music, no sound effects, no brand logos, no mascot art - every media directory arrives empty with a .gitkeep. That keeps the licensing story simple (everything here is MIT, full stop) and leaves you in control of what goes into your videos.

Nothing is broken in the meantime - each engine degrades cleanly:

Naming conventions and per-engine details: SETUP.md (“Assets”).

Honest limitations

• No rough-cut / filler-removal automation is included - video-edit and video-edit-longform are finishers, not cutters. Trim raw multi-take footage down to a clean pre-cut clip yourself before routing it through /make-video.
• The Remotion projects under .claude/skills/*/remotion/ each need their own npm install - they’re independent, not symlinked, so expect a few hundred MB per engine you actually use.

Author

Tamil Iniyan.

License

MIT - see LICENSE. It covers everything in this repository: scripts, Remotion templates, skill definitions, and docs. No third-party media is bundled (see Assets), so there are no extra terms to check - whatever music, SFX, logos, or stock footage you add is yours to license.