ROLE You are a motion design director from a premium ad agen...

Prompt

ROLE You are a motion design director from a premium ad agency plus a creative technologist. Your answers get implemented by an autonomous editing agent, so every recommendation must be concrete: schemas, numbers, thresholds. No mood boards, no vague "make it feel premium", no invented tool flags — if unsure something exists, write "verify availability". PROJECT CONTEXT An autonomous short-form editing agent ("Pro Editor Brain") builds 1080×1920 vertical videos (15–40 s, H.264/AAC 30 fps) from raw or pre-edited footage plus a campaign brief. Stack: Python 3.11+, Pillow, CairoSVG (SVG→PNG with alpha), FFmpeg overlay/filtergraph; no After Effects. Taste target: premium ad-agency polish — designed motion systems, brand-locked typography and color — fused with viral short-form mechanics. Already implemented (do NOT re-propose): five fixed motion recipes (gradient hero word, glassmorphic stat card, angled kinetic word cluster, progress badge, closing stamp); design-token concept (max 2 font families, neutral+accent palette, type hierarchy, motion vocabulary, SFX vocabulary); bounding-box math with rotation bounds and collision checks (temporal+spatial+semantic); safe-area exclusion masks with honest "provisional platform profile" labeling; frame-accurate timing and easing; campaign end-card gates; caption rules; QC gates. What is missing: the agent can EXECUTE fixed recipes but cannot DESIGN. There is no method for reading raw footage and deciding WHERE graphics add communication value, WHAT to design, and HOW to keep 30+ clips from one campaign coherent yet non-templated. YOUR TASK — design the MOTION GRAPHICS DESIGN ENGINE Answer all 8 questions with tables/schemas inline. End every answer with 2–6 lines starting with "RULE:" — imperative prompt-ready directives. Q1. OPPORTUNITY TAXONOMY Define 10–12 graphic moment classes for vertical short-form (e.g., hook title, context setter, stat/claim card, process/step badge, kinetic emphasis, callout/arrow annotation, comparison, quote, location/time stamp, transition graphic, end CTA card). For each: communication purpose, ideal duration range, placement region, and 2–3 trigger conditions observable from transcript/audio/shot analysis. Q2. MOMENT DETECTION WITHOUT HALLUCINATION How should the agent infer "this is a stat moment" / "this is a quote moment" from transcript + audio events + shot type + empty frame regions — with an evidence requirement per class (minimum confidence to propose, what blocks it), so graphics never appear without a real referent in the footage? Q3. DESIGN SYSTEM AND VARIETY Style-frame approach before batch production: tokens (type scale for 1080×1920, grid/margins, palette derivation from brand assets incl. logo color extraction), 1-coherent-language rule. Then VARIETY-WITHIN-SYSTEM: concrete permutation dimensions (layout archetype, alignment, entrance direction, density, accent usage) so 30+ clips stay fresh without breaking brand coherence. Include "when to deliberately break the grid". Q4. ANIMATION CHOREOGRAPHY STANDARDS Per element class: entrance/exit duration ranges (frames at 30 fps), stagger intervals, easing recommendations, reading-time math (words/sec for on-screen text), hold-minimum rules, exit discipline (clean before cut vs carry-through), and sound-sync conventions (which events get audio anchors). Q5. ASSET PIPELINE Recommend the primary + fallback path for precision, alpha correctness, and performance in our stack: (a) SVG→CairoSVG→PNG-with-alpha→FFmpeg overlay loop; (b) FFmpeg drawtext/drawbox native; (c) Python-rendered PNG sequences with alpha composited via overlay. Include: text-shaping risks (kerning, variable fonts), HiDPI/anti-aliasing practice, color-space consistency, and how to spec a keyframe-driven overlay renderer that reads a JSON animation and emits exact frames. Q6. ANTI-AI-TELLS FOR GRAPHICS 12 "AI graphic slop" patterns (the motion-graphics equivalents of machine-made video) with detection heuristic + agency-grade alternative each (e.g., every element same entrance, centered-everything, gradient-on-everything, overshoot on every object…). Q7. QC FOR GRAPHICS Measurable checks: transformed bounds vs safe areas, collision rules, contrast/readability thresholds at phone viewing size, frame-step review checklist (which frames matter: entrances, extrema, exits), and honest limits (what needs human eyes). Q8. STARTER ARCHETYPE LIBRARY Spec 10–12 graphic archetypes BEYOND our existing five. Per archetype: purpose, trigger, parameter schema (JSON fields with types/ranges/defaults), animation outline (key events in frames), failure fallback (reduced-motion or omit). OUTPUT FORMAT Sections Q1–Q8, tables/schemas inline, RULE: lines per section, and a final IMPLEMENTATION ORDER (5 steps with acceptance tests).