mujoco-explainer-videolisted
Install: claude install-skill Yash-Kavaiya/mujoco-explainer-skills
# MuJoCo Explainer Video
Turn a concept into a narrated-looking explainer video: **storyboard YAML → MuJoCo offscreen
render → PIL overlays → ffmpeg H.264**. One command, deterministic output, built for an agent
to iterate on by *looking at* a contact sheet.
```
skills/mujoco-explainer-video/
├── scripts/make_video.py # CLI entry point
├── scripts/mjexplainer/ # pipeline.py · sim.py · graphics.py
├── assets/scenes/*.xml # camera-ready scene library (referenced as @scenes/…)
├── assets/scenes/controllers/ # lqr_balance.py (generic LQR from mjd_transitionFD)
├── references/storyboard-spec.md # every YAML key, with defaults
└── templates/storyboard.yaml # copy-and-edit starter
```
## Setup (once)
```bash
pip install -r requirements.txt # mujoco>=3.2 imageio imageio-ffmpeg pillow numpy pyyaml (+ pyttsx3 for --tts)
```
Headless Linux / CI: `export MUJOCO_GL=egl` (NVIDIA) or `MUJOCO_GL=osmesa` (CPU). Windows/macOS need nothing.
## Workflow
1. **Plan the story.** If the user gave only a topic, use the `mujoco-physics-storyboarding`
skill to pick 4–8 shots (hook → concept → sim → contrast → takeaway). Each sim shot should
prove exactly one claim the caption states.
2. **Choose or build scenes.** Check `assets/scenes/` first (run
`python ../mujoco-scene-authoring/scripts/validate_scene.py <xml>` to list its bodies/sites/joints).
New models: follow the `mujoco-scene-authoring` skill, and keep the shared visual style.
3. **Wri