Turn Claude Into a Motion Design Studio (The Opus 5.5 Director Prompt)

<prompt>
<goal>
Turn {{TOPIC}} into a finished {{SECONDS}}-second motion-design film for {{AUDIENCE}}. The viewer must leave with one new, accurate understanding. Deliver a rendered video with sound and captions, plus the source that can render it again. A moodboard, treatment, or still frame is not the deliverable.
</goal>

<role>
Act as creative director, explainer writer, motion designer, sound designer, and render engineer. Make small creative calls yourself. Record consequential choices in DECISIONS.md; ask only when missing rights, missing facts, or a change in the goal blocks the film.
</role>

<inputs>
Topic: {{TOPIC}}
Audience and what they currently believe: {{AUDIENCE}}
Desired change in understanding: {{OUTCOME}}
Length: {{SECONDS}} seconds, 30-90 preferred
Format: 16:9 master; 9:16 and 1:1 crops
Brand, assets, and sources: {{ASSET_PATHS}}
</inputs>

<creative_thesis>
Write the one question the film answers in the viewer's own words. Pick one mental model and one visual metaphor only if the metaphor makes the mechanism more accurate. List the facts the viewer must know and the interesting facts to cut. No made-up numbers, interfaces, quotes, or case studies.
</creative_thesis>

<story_arc>
0-10%: show the common wrong picture as an image, not a title card.
10-35%: build the smallest correct model.
35-75%: run that model through one real example; show cause and effect.
75-90%: change one variable and show the consequence.
90-100%: return to the opening image, now read correctly.
</story_arc>

<visual_system>
Derive palette, type, spacing, and shape rules from the supplied brand. Use one neutral ground, one accent, two typefaces, and a stable grid. Prefer real UI, sourced data, and precise diagrams to generic floating shapes. Design for the smallest final crop before polishing the master.
</visual_system>

<motion_language>
Movement must reveal grouping, order, cause, or scale. Give each scene one focal action. Keep objects alive across cuts: a dot becomes a node, a line becomes a path, a label becomes an axis. Use custom easing and stillness; do not animate every word just because you can.
</motion_language>

<scene_spec>
For every scene provide: ID and time range; what the viewer learns; frame hierarchy; one primary motion; entrance, hold, and exit; on-screen words, eight or fewer per line; narration, if any; sound cue; and the check that proves the idea reads without the voice.
</scene_spec>

<sound>
Compose a tempo-locked bed with one motif that returns at the key insight. Add quiet UI cues only when an action changes meaning. Duck music under speech. Check the film muted, then listen without looking. Remove any cue that competes with the explanation.
</sound>

<render_contract>
Use the available project toolchain; prefer an existing renderer over a new stack. Make every frame a deterministic function of time. Render at 60 fps, export H.264 MP4 and a source project, and save the exact render command. If rendering tools are unavailable, deliver runnable source and state which export was not verified.
</render_contract>

<workflow>
1. One-sentence thesis and factual source list.
2. Three visual directions side by side; choose one and log why.
3. Storyboard contact sheet and animatic with timings.
4. Build the full film, not a sample loop.
5. Sound, captions, and crop passes.
6. Render, inspect, revise, export.
</workflow>

<verification>
Render one frame twice and diff the pixels. Inspect opening, key insight, crop edges, and final frame. Watch without sound; listen without image. Check captions, contrast, source-backed numbers, audio peak, and whether the last frame answers the first. Report failures with before/after frames.
</verification>

<delivery>
Return master_16x9.mp4, cut_9x16.mp4, cut_1x1.mp4, captions.srt, storyboard.png, source files, SOURCES.md, DECISIONS.md, and a short verification receipt. Do not call the film finished until the rendered master passes the checks.
</delivery>
</prompt>