Video Production Agent Skill

Turn an AI agent into a full-stack video producer: research, storyboard, narration, animated visuals, render, QC, archive, and delivery.

This is built for requests like:

"Make a video explaining this research paper."
"Make a product walkthrough with narration and animated diagrams."
"Make a daily summary video with sources and caveats."
"Go all out on an explainer with Manim visuals and narration."

Install with an AI agent

Want an AI coding agent (Claude Code, OpenClaw, Gemini CLI, etc.) to install this for you? Just point it at INSTALL_FOR_AGENTS.md - a step-by-step guide that walks the agent through adding this skill (plus its scripts and references) to its skills directory and checking dependencies. No manual setup required.

Prefer to install it yourself? See the Install section below.

Showcase

These are real generated videos created with the skill workflow for this repository update. Each one includes narration, visuals, source notes, and verified audio/video streams.

Repo intro	GenLIP paper showcase	ClawMark paper showcase

Full MP4 / verification	Full MP4 / sources	Full MP4 / sources

Repo intro: 226-second narrated walkthrough of what this AgentSkill does and how agents use it.
GenLIP / Let ViT Speak: 356-second narrated showcase for a recent multimodal generative pretraining paper.
ClawMark: 331-second narrated showcase for a recent multi-turn, multi-day coworker-agent benchmark paper.

Lightweight Example Clips

These short clips are compact examples for quick README browsing, not the full generated showcase videos.

Research paper explainer	Product walkthrough	Current-topic briefing

10-second MP4	10-second MP4	10-second MP4

What it does

video-production gives agents a repeatable production workflow:

Create a durable job workspace.
Pull or extract source material.
Build a source-grounded production brief.
Write segmented narration.
Generate TTS voiceover.
Produce Manim/ffmpeg/PIL visuals.
Sync animation timing to audio.
Render and mux final MP4.
QC frames and audio stream.
Archive and optionally host the deliverable.

Repository layout

skills/video-production/
|-- SKILL.md
|-- scripts/
|   |-- init_video_job.py
|   |-- synth_kokoro.py
|   |-- archive_video.py
|   `-- host_video.py
`-- references/
    |-- production-standards.md
    `-- source-playbooks.md

dist/
|-- video-production.skill
`-- video-production.zip

examples/videos/
|-- paper-explainer-demo.mp4 / .gif
|-- product-walkthrough-demo.mp4 / .gif
`-- current-topic-demo.mp4 / .gif

Install

Using an AI agent? Point it at INSTALL_FOR_AGENTS.md and it will handle everything below automatically.

OpenClaw

git clone https://github.com/Aaryan-Kapoor/video-production-skill.git
mkdir -p /path/to/openclaw-workspace/skills
cp -R video-production-skill/skills/video-production /path/to/openclaw-workspace/skills/

Start a new OpenClaw session so the skill snapshot reloads.

AgentSkills / Claude-style upload

Use either packaged artifact:

Both archives contain a root video-production/ folder with SKILL.md, scripts/, and references/.

Gemini / generic agents

Load skills/video-production/SKILL.md and keep the sibling scripts/ and references/ directories available.

gemini skills install ./skills/video-production

Requirements

Required:

Python 3.10+
ffmpeg + ffprobe

Recommended:

Manim for animated diagrams/equations
Kokoro TTS for the built-in synth_kokoro.py path, or another TTS stack such as OpenAI TTS, ElevenLabs, or Piper
poppler-utils for paper/PDF extraction
Pillow for image/crop workflows
Tailscale for private one-file hosting fallback

Set VIDEO_PRODUCTION_ROOT to control where generated jobs live. Default: ~/video-productions.

Quick test

export VIDEO_PRODUCTION_ROOT="$PWD/.video-productions"
python skills/video-production/scripts/init_video_job.py "Attention explainer" --prompt "Make a short video about attention"

Kokoro TTS check

If Kokoro is installed in the active environment:

python skills/video-production/scripts/synth_kokoro.py --check

Or point the skill at a Kokoro virtualenv:

KOKORO_PYTHON=/path/to/kokoro/.venv/bin/python \
  python skills/video-production/scripts/synth_kokoro.py --check

Package locally:

./scripts/package.sh

Safety

This skill can lead agents to run local scripts and media commands. Review the skill before enabling it, keep secrets out of the skill folder, and only host files you intend to share.

License

MIT

Video Production Skill

README