Curiosity 3D Explainer

Mission

Produce a fast, understandable 3D explainer from one topic or question. Use a curiosity loop to organize attention, but keep every factual statement source-grounded and clearly separate from creative staging. The workflow builds reusable character/object references before generating dependent shots.

Required inputs

• one topic or question and the intended audience;
• reliable sources or permission to draft a research brief;
• desired duration, language, ratio, resolution, narration, music, and captions;
• recurring character, anatomical, product, or object descriptions;
• visual style, motion intensity, CTA, variant count, and review owner; and
• rights/consent for people, voices, brands, reference images, and music.

For medical, scientific, financial, or safety topics, require source review and an appropriate disclaimer. Generated anatomy or process animation is illustrative unless verified by a subject-matter owner.

Phase 1: curiosity script and beat breakdown

Draft a compact structure:

1. open with a precise question or observable setup;
2. introduce the subject and establish the viewer's mental model;
3. reveal one supported mechanism or surprising contrast per beat;
4. resolve the question without manufacturing a twist; and
5. close with the requested CTA or a clear ending.

Create a beat record for each shot containing narration, factual source, subject/action, camera, environment, cross-section or detail requirement, character/object IDs, duration, and acceptance checks. Freeze the script before generating a large batch.

Phase 2: character and asset anchors

Generate or accept turnaround/reference sheets for recurring characters, products, anatomical parts, or props. Use the adjacent image agent or an available image capability. Review identity, proportions, materials, labels, and style before using a sheet as a parent.

Create one keyframe per beat, reusing the approved anchors. Keep a continuity sheet for wardrobe, colors, orientation, time, world rules, and camera direction. Present keyframes for approval before motion rendering.

Phase 3: motion, voice, and assembly

1. Choose a live image-to-video or reference path from MODELS.md, preserving shot order and model duration constraints. Use muapi-video-tools.md for exact fields.
2. Upload local anchors and record roles, rights, and returned URLs.
3. Animate each approved keyframe with a single clear action and camera move. Avoid complex transformations, contradictory movement, and fake factual capture.
4. Generate or accept narration and music as separate assets. Use the adjacent voice agent or available capabilities only with appropriate consent and licensing. Preserve pronunciation and exact approved wording.
5. Combine clips in beat order, mix audio, and generate captions after timing stabilizes. Poll and record every prediction ID.
6. Run video QA for story order, continuity, temporal stability, factual labeling, audio, captions, and export.

Output contract

Return the source/evidence brief, script version, beat table, character/object sheet IDs, keyframes, animation jobs, voice/music provenance, prediction IDs, assembled output URL, caption status, QA findings, and failed branches. Save permitted plans and receipts under .video/ using reports-and-receipts.md.

Curiosity, 3D rendering, or a generated voice is not evidence of a real event or expert conclusion. The result is draft-only and never published here.