Visual Design

The unit is a time-coded shot sequence

A frame’s visual layer is a sequence of time windows paced to the voiceover, not a bag of effect tags. The failure that reads as PowerPoint is front-loading: the agent rushes the whole canvas on screen in the first ~25%, and then it just sits (the old representation — a flat set of effect names + a prose note — fired everything at entrance and left the rest empty). A time-coded shot sequence written against the VO makes that impossible: each window states what is on screen and what is moving, and nothing appears before the voiceover reaches it.

Write each frame as a handful of windows cued by the spoken line:

• Each Scene line names what’s on screen, what moves in this window, and where it sits (layout, inline). Times are real seconds across the frame’s duration.
• Pace reveals to the voiceover; never front-load. This is the core anti-PowerPoint mechanism (→ motion-language.md Part 2 Rule 2). At t=0 show only what the VO is saying then; reveal each further piece — a line, a card, even an h1 — when the VO names it, spreading reveals across the shot and especially the back ~50%. The window count = the number of spoken cues the line calls for — a two-beat line is two windows, a five-feature list is five or six. There is no fixed count and no mandatory “middle” act; the only sin is dumping everything up front.
• End on a held read. Once the content has resolved it holds and reads — prefer stillness to bad motion: no forced camera drift, no lazy breathing, no back-half pan/push; at most a subtle jitter keeps it alive (→ motion-language.md). On a short shot the final reveal and the hold are the same window — the hold is not a separate mandatory act. Only the final frame has a real exit; every other frame’s exit is the harness transition (story’s transition_in).
• A deliberately held frame — content already revealed, now reading still — is legitimate and often right (a climax, a breather). The failure is never “too still”; it is front-loaded-then-frozen (everything dumped by ~25%, nothing cued to the VO). Place held beats deliberately for rhythm so the video isn’t uniformly busy (allocate them in ## Video direction). Reveal pacing + holds + the idle budget → motion-language.md.

Pick the shape — instantiate a blueprint

Don’t invent each shot from scratch. The frame’s role (its type / beat) points to a proven shape:

1. Match the role to a blueprint. Open ../hyperframes-animation/blueprints-index.md, find the frame’s role in the role→blueprint menu, and pick the blueprint whose intent fits this beat (story may already have named a candidate id — confirm or override it). Read that blueprints/<id>.md: it is a short, product-agnostic, time-coded shot template with [slots] and a named signature move (the thing that makes the shape itself — the SVG ring, the push-THROUGH, the in-place token swap).

2. Instantiate its [slots] with THIS product’s content — three postures:

   • Reproduce — the blueprint fits the beat and your content maps onto its slots cleanly. Fill every [slot] with this product’s word / asset / stat and follow its Scene timing. Write the resulting Scene lines.
   • Adapt — the structure fits but the content / asset-count / surface doesn’t (or you want a fresher surface to avoid templating). State what you keep / what you change in one line, then write the adapted Scene lines. You may extend or vary; you may never drop the signature move (drop it and you picked the wrong blueprint), and you keep the reveals paced to the VO — never collapse the shape to a single front-loaded dump.
   • Compose — no blueprint fits the beat. Build the shot from the motion vocabulary in motion-language.md: still pace the reveals to the VO across the shot, never fire everything at t=0. Mark it blueprint: compose.

3. Keep the signature move. Whichever posture, the blueprint’s signature move (named in its file) is the spine of the shot — it usually lands on the shot’s key reveal. Carry it through.

The blueprint’s own Scene lines, motion vocabulary, and rule mapping are your raw material; you are choosing a shape and casting this product into it, not copying an engineering spec.

What you add to each frame

Story-design’s ## Frame N block already carries the narrative + asset_candidates. You append the shot. Story’s scene / voiceover / transition_in / role fields stay untouched.

The lightweight tags:

• blueprint: — the id you instantiated (with (Reproduce) / (Adapt)), or compose. One id per frame.
• focal: — which existing candidate is the hero of this beat.
• roles: — each candidate’s role: cutout (foreground subject, lay text around it) · background (full-bleed, dim 30–50%) · supporting (secondary). You consume the candidates story chose — never add, swap, or drop one (coverage is story’s call; if a frame truly has the wrong candidates, flag it back, don’t reach into capture/).
• sfx: — name the sound the beat wants (an impact for a slam, a whoosh for a push, a riser into a reveal). The audio script’s fetch-sfx pass retrieves it and the assembler mounts it at the root — you only name it, never embed an <audio> element.

Layout is stated INLINE in each Scene line — name the template, density, depth, and hierarchy as part of “where it sits” (Centered, ~50% of frame, asymmetric 60/40, 3 depth layers), drawing on the Layout vocabulary below; never write px / scale / shadow recipes (the worker writes those).

Motion is named INLINE in each Scene line — name the move from motion-language.md‘s vocabulary (ring sweeps, pushes THROUGH, count-up, glow blooms) and let it settle on a long-tail curve (power3 default — smooth beats bouncy; see motion-language.md). Never write ease curves / ms / stagger (those resolve in Step 5 from this skill’s local ../hyperframes-animation/rules/).

Layout — named inline per Scene

State each Scene’s layout as part of “where it sits.” If the blueprint already implies a composition (a ring around a center, stations on a wide canvas, two cards from opposite wings), that wins — describe it directly; the vocabulary below is for composing freely or a generic beat, not a menu you must pick from. Never write px / scale / shadow (the worker does). One frame’s layout can EVOLVE across its Scenes (Scene 1 centered hero → Scene 2 rearranges to a grid).

• Framing vocabulary — centered (hero / climax) · rule-of-thirds · split-screen (comparison) · layered-depth (immersive) · asymmetric 60/40 or 70/30 (editorial) · triptych (three panels) · full-width strip. Vary the framing across the video so it doesn’t read as one repeated template — let the beat decide, not a quota.
• Density — primary visual ≥ 40% of canvas; ≥ 3 depth layers (background + midground + foreground); never a lone small cluster floating in empty space. Squint test: after blur you can still pick out the #1 element.
• Hierarchy — combine ≥ 2 of size (3:1) / weight (800 vs 400) / contrast / position (upper-third is golden) / motion, so one element clearly dominates.
• Depth — layer 2–3 of: size, blur, opacity gradient, overlap, shadow-stack.
• Don’t show: nav bars, footers, scrollbars, real cursors / browser chrome, generic decorative shapes standing in for a real asset, floating bokeh / purple-blue “AI” gradients — unless it’s an intentional UI-demo reconstruction.

## Video direction — write the invariants ONCE

The whole video shares one look and one motion grammar. Write a ## Video direction block ONCE at the top of STORYBOARD.md so every frame inherits it and per-frame Scene lines carry only the delta. This block is load-bearing — it is what binds many independent shots into one film. Keep it.

• palette system — from frame.md: which roles map to which hues. Never invent.
• motion grammar + reveal model — long-tail eases (power3 default, smooth over bouncy) + the VO-paced reveal model every frame follows (reveal each piece on its spoken cue; never front-load) + what may stay alive during a hold (subtle jitter at most; no lazy breathing) (→ motion-language.md).
• rhythm / held-frame allocation — name the held / breather frames (often before a climax) so the video varies its energy: most frames reveal to the VO, a few hold still (a held read beats bad motion; the anti-monotony discipline; → motion-language.md).
• negative list — what never appears: off-brand textures / effects the pack forbids, plus both motion failure modes — slideshow (front-load then freeze) and screensaver (everything floating independently) (→ motion-language.md).

Do not repeat these per frame — restating video-level rules in every frame is exactly the bloat this layer prevents.

Palette & type — from frame.md, never invented

• Palette — frame.md (the adopted pack) is the color truth; apply its roles per frame. Generic basics (one accent, tint neutrals, avoid pure #000/#fff) → hyperframes-creative/references/house-style.md.
• Type — fonts resolve via frame.md‘s type tokens; reference them by role (display / body / mono / the pack’s ramp), never by raw family or px. Generic typography craft (embedded fonts, dark-bg optical compensation, tabular-nums) → hyperframes-creative/references/typography.md.

Caption-band keep-out (plan side)

The bottom ~17% of the canvas is reserved for the caption pill. Plan every frame’s content into the top ~83% so nothing important lands in the band (the worker enforces the pixel cutoff; you plan the layout). Holds even when captions are disabled — bottom-edge consistency.

Where the detail lives

Before you finish — checklist

• ## Video direction written once at the top (palette · motion grammar + shot model + idle budget · stillness allocation · negative list incl. both failure modes); per-frame entries are deltas, not restatements.
• Every frame is a time-coded shot sequence with real second windows across its duration — not a tag bag.
• No frame front-loads — at t=0 only what the VO is saying enters; each further piece reveals on its spoken cue, across the back ~50%. Window count follows the VO, not a fixed number.
• Every frame names an blueprint: id (Reproduce / Adapt) or compose; an Adapt states keep/change and keeps the signature move; nothing collapses to a single front-loaded dump — reveals stay paced to the VO.
• Held frames are deliberate — allocated in Video direction for rhythm; a held read is fine (prefer stillness to bad motion), but no frame may be front-loaded-then-frozen.
• Each frame’s asset_candidates have a focal + per-candidate roles; none added, swapped, or dropped.
• Layout named inline per Scene (template / density / depth / hierarchy — the Layout vocabulary here); motion named inline per Scene from the vocabulary (motion-language.md). No px / ease curves / ms / JS.
• Content planned into the top ~83% (caption band clear).
• Palette / type pulled from frame.md by role — nothing invented.
• You wrote no HTML and never read capture/.