video-animatic
Use when building an animation, motion graphic, kinetic type, animated chart from data, or animatic with code and rendering it to MP4 frame-exact. Compiles a scene.v1 JSON into a seekable HTML timeline, lints readability, flashes and contrast, renders with Playwright, and exports clean keyframes for AI shot upgrades.
Try your first task
A source-grounded starter task has not been verified for this skill yet.
SKILL.md
name: video-animaticdescription: Use when building an animation, motion graphic, kinetic type, animated chart from data, or animatic with code and rendering it to MP4 frame-exact. Compiles a scene.v1 JSON into a seekable HTML timeline, lints readability, flashes and contrast, renders with Playwright, and exports clean keyframes for AI shot upgrades.argument-hint: "[scene.json]"license: MITmetadata: version: 1.0.0 build_pattern: "scene.v1 JSON → one self-contained HTML page with window.__seek(t) → deterministic Playwright frame capture → ffmpeg. Five stdlib scripts; Playwright is the only install." distinct_from: "demo-video (screenshot walkthroughs); markdown-html md-slides (presentations, not video); cinematic-shot-prompting (prompts for generated clips)"Video Animatic
Portability: Python stdlib scripts + Playwright (
pip install playwright && python -m playwright install chromium, or Node Playwright) + ffmpeg. Free, offline with--offline.
The film is data. scene.json describes shots and layers in normalized
coordinates; timeline_compiler.py turns it into one HTML page whose
window.__seek(t) draws the exact frame at time t. Nothing runs on a wall
clock, so frame 214 is the same pixels every render, and one changed shot
re-renders alone.
Workflow
S=${CLAUDE_PLUGIN_ROOT}/skills/video-animatic/scripts
python3 $S/motion_linter.py --scene film/scene.json # exit 4 = fix first
python3 $S/timeline_compiler.py --scene film/scene.json --out film/animatic.html
# open film/animatic.html in a browser: play, scrub, review
python3 $S/frame_renderer.py --html film/animatic.html --fps 12 --mp4 film/animatic.mp4 # review cut
python3 $S/frame_renderer.py --html film/animatic.html --shots s3 # re-render one shot after an edit
python3 $S/timeline_compiler.py --scene film/scene.json --aspect 9:16 --out film/vertical.html # same film, vertical
Always lint the target aspect too (motion_linter.py --aspect 9:16): text that
fits 16:9 often leaves the safe area when re-framed.
scene.v1 in one screen
{"schema": "scene.v1",
"meta": {"title": "Launch", "aspect": "16:9", "fps": 30, "target_duration": 12, "mode": "greybox"},
"theme": {"bg": "#0b0d12", "fg": "#f5f7fa", "accent": "#5b8cff", "font_heading": "Inter"},
"shots": [{"id": "s1", "duration": 4, "transition": "cut",
"camera": {"move": "push-in", "amount": 0.06},
"vo": "One line of narration.",
"layers": [{"type": "text", "text": "Headline", "style": "title", "x": 0.5, "y": 0.45, "in": 0.2, "anim": "words"}]}]}
| Field | Values |
|---|---|
layer type | text, shape, image, chart (bar/line, data or csv), counter, lower_third, plate (full-frame footage slot) |
anim | fade, rise, slide, scale, type, words, draw, grow, kenburns, none |
camera.move | static, push-in, pull-out, pan-left/right, tilt-up/down, drift |
transition | cut, fade (crossfade), dip (through background) |
mode | greybox (images and plates as labelled boxes) or final (real media; plate src may be an MP4) |
Text styles: title, subtitle, body, caption, kicker. Colours take a
hex value or a theme key (fg, accent, muted). fixed: true keeps a layer
out of the camera move.
Linter (exit 4 blocks)
ML1 total vs target length · ML2 text leaves before it can be read · ML3 voice-over longer than its shot · ML4 more than 3 flashes per second (WCAG 2.3.1) · ML5 text contrast under 4.5:1 · ML6 layer timed outside its shot · ML7 upgrade shot without a prompt. Warnings: ML8 safe area, ML9 pacing, ML10 more than two fonts.
Keyframes and the upgrade manifest
For shots marked upgrade: true, give the plate a storyboard still as src,
compile with --mode final, then:
python3 $S/frame_renderer.py --html film/final.html --out film/keyframes --keyframes
python3 $S/shot_manifest.py --scene film/scene.json --keyframes film/keyframes --out film/shots.json
shots.json (shots.v1) holds per shot: prompt, exact duration, aspect, a
camera-intent sentence, first/last keyframes and what stays as code overlay.
It names no provider and calls nothing; video-director quotes and sends it.
A greybox placeholder is never sent as a keyframe, because it carries no framing.
Render only pages you compiled yourself: frame_renderer.py opens the HTML in
Chromium from file://, where its script can read local files. The render
browser blocks every remote request except Google Fonts and reports what it
blocked; pass --allow-remote only for media URLs you trust.
Design rules
- One idea per shot; one heading face and one body face.
- Motion explains, never decorates: every move should point at what matters.
- Numbers count up, bars grow, lines draw; the viewer watches the change, not a slide.
- Keep the plate prompt free of text: titles and figures stay code.
See motion_craft.md for timing, easing, accessibility and render sources.