Mission
Turn a topic, authorized presenter recording, or approved photo/product into a bold editorial explainer with a coherent paper-collage visual language. The workflow separates script, beat design, keyframe approval, animation, audio, and assembly so each layer can be reviewed independently.
This is a visual treatment—not a claim that the source style, generated scene, or narration is associated with any particular publisher or creator.
Input modes
- Topic mode: research or supplied facts become a script, beat plan, collage keyframes, animated coverage, narration, music, and captions.
- Presenter mode: an authorized talking-head source supplies narration and timing; the workflow creates collage coverage around it only when the host exposes a compatible edit path.
- Anchor mode: an authorized photo or product remains a recognizable, clearly labeled collage element across generated beats.
Collect topic, audience, factual sources, language, duration, ratio, visual palette, collage materials, CTA, audio, rights/consent, and review owner. If a topic involves current events, medicine, finance, or a real person, require source review and do not use generated scenes as evidence.
Phase 1: script and beat layer
Create a beat table:
| Beat | Narration | Visual idea | Collage elements | Motion |
|---|---|---|---|---|
| 1 | hook and question | establishing poster | subject, title-safe space | entrance/reveal |
| 2 | supported explanation | editorial metaphor or diagram | cut-outs, labels | parallax/pan |
| 3 | consequence or contrast | close detail or comparison | texture, arrows, props | push/transition |
| 4 | conclusion/CTA | final lock-up | approved logo/copy | settle/hold |
Keep the spoken script separate from on-screen text. Mark claims as sourced, proposed, or approved. A curiosity hook is not permission to exaggerate or withhold material context.
Phase 2: visual system and keyframes
Define a compact visual system: paper grain, torn edges, cut-out silhouettes,
tape/shadow treatment, high-contrast palette, typography-safe regions, and
allowed photographic versus illustrated elements. Generate one representative
style frame first, then create beat keyframes through the image agent or
available media.generate_image capability.
Keep recurring people, products, and symbols anchored to approved references. Review the keyframe board for beat order, legibility, identity, product truth, and claim context before animating the full sequence.
Phase 3: motion, audio, and assembly
- Animate each approved keyframe with short, physically coherent motion: parallax, paper-layer drift, camera push, reveal, or controlled pan. Use MODELS.md and a live schema check for duration, ratio, and image input.
- Generate or accept narration. Use the adjacent voice agent or an explicitly available speech capability; record voice consent and provenance.
- Generate or accept music separately. Use the music workflow in muapi-video-tools.md, retain music rights, and keep narration intelligible through the mix.
- Poll all jobs and combine approved clips in beat order. Add captions only after the timing is stable; proofread every factual term, name, and number.
- Run video QA for collage consistency, temporal artifacts, audio, text, captions, factual labeling, and export.
Do not silently replace a failed beat, turn a source presenter into an unconsented likeness transfer, or imply that a collage depiction is archival footage.
Output contract
Return script/source notes, beat table, visual system, keyframe IDs, animation
jobs, audio assets, prediction IDs, assembled output URL, caption status, QA,
rights/disclosure state, and failed branches. Save permitted artifacts under
.video/ using reports-and-receipts.md.
This skill creates a draft package only; it does not publish or certify facts.