Flat Infographic Explainer
Create an original, voice-led explainer using a continuous visual world and semantic animation. Reuse the production grammar, not the reference video's footage, wording, or exact illustrations.
Route the request
- Read references/style-system.md for every project.
- For a Chinese legal voiceover targeting 16:9 horizontal delivery, or any explicit request to follow the "11月15日模式", also read references/legal-horizontal-mode.md.
- When matching or extending the proven Nov15 implementation, read references/nov15-case-study.md and inspect the repository files it names if they exist.
- Preserve the supplied narration verbatim unless the user explicitly asks for rewriting. Transcript correction may fix ASR homophones and punctuation only when confirmed by the audio and context.
Workflow
- Lock the inputs: voice track, transcript, target aspect ratio, frame rate, duration, palette preference, legal sources, and delivery codec. Derive the composition duration from the actual audio.
- Split narration at semantic turns, not fixed time intervals. Give each scene one visual model such as threshold, comparison, flow, hierarchy, evidence cards, scale, checklist, or conclusion.
- Produce a scene plan containing narration range, visual claim, objects, motion beats, caption phrase, and evidence note. Flag any unsupported factual or legal claim instead of inventing a source.
- Keep the background field, environmental line art, top context bar, ground line, and caption rail persistent. Let foreground objects enter, transform, connect, and leave; never present a sequence of static title cards.
- Map important nouns to visible objects and verbs or relationships to motion. Reveal an object near the word it represents, and animate only relationships explained by the narration.
- Build with the repository's Remotion stack. Use
useCurrentFrame(),interpolate(),spring(), and deterministic SVG/DOM animation. Do not use CSS transitions, random values, wall-clock timers, or an additional animation framework by default. - Keep narration captions to one semantic line. Separate captions from diagram labels, highlight only meaningful keywords, and keep each caption visible through the end of the spoken phrase.
- From the repository root, validate the timing manifest with
python .codex/skills/flat-infographic-explainer/scripts/validate_timeline.py <manifest.json>, run the project checks, inspect representative stills, render a low-bitrate preview, and only then render the final master.
Non-negotiable constraints
- Do not import copyrighted reference footage or trace its illustrations into reusable assets.
- Do not invent facts, statistics, citations, case details, or transcript text.
- Do not silently rewrite a legal proposition to make the animation easier. Surface conflicts between narration and authoritative sources for review.
- Do not add animation merely to keep the frame busy. Each primary motion must explain, compare, route, accumulate, block, balance, or conclude.
- Do not force beige or any other single palette. Preserve color roles and contrast while adapting the hue family to the brief.
- Keep scene data, captions, design tokens, and render components separate so later voice tracks can reuse the same primitives.
- Do not add background music unless the user requests it and the asset rights are clear.
Definition of done
- The audio duration and composition duration agree without clipping.
- Scenes continuously cover the timeline and every scene has a meaningful entrance, internal change, and exit or handoff.
- Captions are reviewed against the audio, single-line, non-overlapping, within the safe area, and free of legal-term homophone errors.
- Legal claims and examples have evidence notes; case outcomes are not presented as automatic sentencing formulas.
- Key stills show no overflow, dead frames, accidental blank states, or unsafe subtitle placement.
- The final file is probed for resolution, frame rate, duration, audio, pixel format, and color metadata, and a short QA note records the result.
Expected artifacts
- Reviewed transcript and timing data.
- Scene plan and legal-source review where applicable.
- Reusable theme, stage, typography, motion, figure, flow, and panel primitives.
- Timing manifest, validation output, representative stills, preview render, final render, and QA report.