Turn tabular data into a narrated, animated data video. Core artifact: DVSpec, a renderer-agnostic screenplay for scenes, data bindings, narration, and animation. Render with any stack; Remotion is reference path.
Use
- CSV / Excel / structured table -> multi-scene narrated video
- Pasted data -> animated chart or short data story
- Existing DVSpec / render -> refine, debug, or improve
- Goal only, no render requested -> produce DVSpec plan
Do not use for raw video editing, non-data text-to-video, or live dashboards.
Core Rules
- Plan first. DVSpec before render.
- Data leads story. Pick narrative pattern from data shape, not topic.
- Narration leads viewer eye. Animation follows narration timing.
- Read only needed rule files. Heavy details live under
rules/. - Before calling work done: run
rules/self-review.md.
Router
| Task | Read |
|---|---|
| Profile data, find candidate insights | rules/data-analysis.md |
| Choose narrative pattern, scenes | rules/scene-planning.md |
| Pick chart type, axes, scales | rules/chart-selection.md |
| Author or validate DVSpec | rules/dvspec.md |
| Choose visual system, tokens, motion defaults | rules/design-system.md |
| Pick style pack | rules/styles/README.md |
| Render DVSpec to video with Remotion | rules/remotion-integration.md |
| Add TTS, audio-driven duration, subtitles | rules/voiceover.md |
| Write narration script | rules/narration.md |
| Refine existing output | rules/refinement.md |
| Diagnose bad output | rules/anti-patterns.md |
| Final quality gate | rules/self-review.md |
Workflows
Full video:
data-analysis -> scene-planning -> chart-selection -> dvspec -> design-system -> styles -> remotion-integration -> voiceover -> narration -> self-review
DVSpec only:
data-analysis -> scene-planning -> chart-selection -> dvspec
Single animated chart:
data-analysis -> chart-selection -> dvspec -> design-system -> remotion-integration
Refinement:
refinement -> changed domain rule(s) -> self-review
Narrative Patterns
| Pattern | Use when | Climax |
|---|---|---|
freytag_default |
One main finding builds over time | 60-75% |
hook_then_evidence |
One striking number should open | first chart scene |
comparison_driven |
A vs B is story spine | largest contrast |
time_driven |
chronological change drives story | sharpest shift |
drill_down |
total -> segment -> cause | deepest zoom |
See rules/scene-planning.md before locking pattern.