explainer_animation v1.0.0

name: "explainer_animation" description: "Storyboard-to-render pipeline for simple explainer animations with ffmpeg: kinetic typography captions, procedural backgrounds, Ken Burns stills, and voiceover-timed scenes. Use when asked to make an explainer animation, animate text over video, or build a narrated slideshow-style video without a GUI editor."

Explainer Animation

Purpose

Turn a storyboard (scenes + beats + captions + voiceover) into a finished narrated animation video using only ffmpeg — no GUI editor, no assets required beyond stills and VO audio.

Workflow

1. Storyboard. Write one scene per JSON object (see references/scene-pipeline.md and the template at assets/sample-storyboard.json). Each scene: name, duration, bg, optional captions, optional vo.
2. Voiceover first. Get VO audio and its word timestamps before locking durations (see references/vo-timing.md). Scene duration = VO segment length + 0.5–1.0s padding. Text-driven layouts follow the VO, never the reverse.
3. Render. Run bin/explainer.py storyboard.json -o out.mp4. It renders each scene (background → captions → effects), concatenates, then mixes VO + optional music bed with loudness normalization.
4. Verify. ffprobe the total duration against the storyboard sum; spot-check 2–3 frames per scene (text present, positioned, on time). Fix the storyboard, never hand-edit intermediates.

Output Contract

• One MP4 per storyboard: <name>.mp4, 1280×720, 30fps, H.264 + AAC, yuv420p.
• Captions visible only inside their start → start+duration windows, entry animation complete within 0.4s.
• VO audible, music bed ducked, peaks within streaming loudness (loudnorm I=-16 LUFS).

Operating Rules

1. Never hand-edit the filtergraph in bin/ at request time — fix the storyboard JSON instead; the renderer is the single source of truth.
2. Caption text is written via textfile= (never inline text=), so colons/apostrophes in copy cannot break the filtergraph.
3. Dynamic alpha only: :alpha='expr'. Never fontcolor=white@'expr' — that form fails for runtime expressions.
4. Never use dynamic fontsize='expr' (drawtext scale tricks) — one build family has segfaulted on it; simulate scale with alpha flashes and position snaps (see references/text-recipes.md).
5. All factual claims about ffmpeg options inside this skill were verified on ffmpeg 8.1.2 (2026-09-20); if a recipe fails on another build, simplify the filtergraph and note it in the storyboard log.