README

This replaces the template grab-bag. Six deep languages × scene adaptation beats 54 shallow presets — every render is already fitted to its footage.

Category lock (deliveries field, enforced by the compilers)

Every classic DNA’s home is Cinematic (column) — that is where all ten were built and validated. (Standard/rail mode was retired 2026-06-12; the verbatim-rail need is served by the anchor theme. The old rail combos are archived outside this repo and are not distributed with the skill.)

The ten

fx fields (hero block) the engine understands

entrance: emergence | settle | slam | rise | wipe-up | flicker-on | streak · glow (0–0.5) · echoes: [{dx,dy,color,opacity,blur}] (converge-on-land duplicates) · sheen: true (gradient sweep — pair with wordCss background-clip) · wordCss (per-word treatment; required for background-clip:text, which can’t clip through composited children) · ripple (px — landing bump, captions only) · loom / breathe (hold-life) · letterBlur · bodyLayer (top-level: default layer for narration lines).

How to pick (agent)

1. Read safe-zones.json → heroAnchor.bandLuma + palette.temperature.
2. Bright band (>150) → ink (never fight a bright scene with cream/screen).
3. Else pick by content register: poetic/warm → cream · introspective/luxe → editorial · product/tech → keynote · serious/interview → documentary · hype/social → loud.
4. State the pick + why in one line; the user decides (SKILL.md Step 0 still applies).

Authoring

• Cinematic mode: cinematic.json → "dna": "<name>" (drop the template field).
• Locked per DNA: family, palette scheme, blend, motion curves, hero orchestration.
• Open per group (unchanged): size / weight / style / case / spacing + planes.
• var(--accent) is available in per-group CSS — it resolves to the scene-sampled accent.

What the engine generates from the DNA (no authoring needed)

• Hero three-act: co-visible captions dim 0.35s before the hero lands (setup) → entrance with per-letter stagger, amplitude ∝ spoken loudness (impact) → slow breathe
  • glow until exit (afterglow).
• Scene optics: depth-match blur on embed captions, light-direction contact shadow.
• All deterministic — measured from files, no randomness.

Climax selection — what earns the promotion

Pick the payoff, not the topic: the words you would quote when paraphrasing the beat. “The stars were only limited by the pixel size” — the payoff is pixel size (the surprising claim), not “stars” (the subject). Test: would the word alone, big on screen, make the listener nod? A topic word makes them wait.

• Phrases are legal (match: "pixel size" / "skin food"): a peak is a semantic unit, not necessarily one token. 2–4 words max — beyond that it’s a sentence, not a peak.
• Prefer clause-final — nothing left dangling to place after the lift.
• Never strand a determiner: lifting “stars” out of “The stars…” leaves an orphan “The” that vanishes mid-air. The compiler auto-absorbs a line-leading article into the promoted phrase; mid-line, match the phrase including its article.
• Sizes are TARGETED by the compiler, not just floored — author sizes are hints:
  • apex = frame event: under 88% of usable width → raised toward a 93% fill. Std height cap = sizeRange[1] × 1.25 (≤46cqh; formal register ×1.0) — a std apex bursts a rail, not a body composition. Cinematic cap = sizeRange[1].
  • short words fill with TRACKING: when the height cap binds before the width target (SHINE — 5 glyphs), the compiler letterspaces up to +0.32em until the word owns ~88% of the width (film-title craft: HER / DUNE). Long words fill by scale, short words by air — impact is WIDTH-led either way.
  • minor = damped beat in the apex family: max(3× rail, 0.55× apex-final), width-capped. A peak that reads as a label is a bug.
  • lockup context is RATIO-LOCKED: kicker/tail = 0.26× hero (clamped 0.05–0.085·h), recomputed after any width-fit raise. Don’t hand-tune context sizes — size the hero; the context follows like a poster system.

Climax placement — the lockup is the default, NOT the only composition

A hero line takes "placement": "subject" | "column":

• subject (default) — the ORBIT lockup centered ON the subject. NOT a centered sandwich: context anchors to the hero’s EDGES like a poster kicker/tagline — pre-context flush-left off the hero’s top-left corner, post-context flush-right off its bottom-right — so the small lines land BESIDE the head/shoulder (on the scene), never on the face, and therefore EMBED (bg) like everything else. The eye reads a diagonal: kicker ↘ HERO ↘ tail. Only the hero crosses the silhouette (the matte-occlusion showcase). The context’s job is reading continuity — it gets the margins; the hero’s job is the cinematic event — it gets the subject. Right when the peak is THE dramatic moment: cream / loud / glitch / neon / velocity lean here.
• column — the apex stays IN its column’s reading flow and BURSTS it (oversized, full hero motion + dim/glow privileges, but composed within the sentence’s home zone; spills toward the subject naturally when oversized). Right when the content is a continuous read and the column is the narrative’s home: keynotes, tutorials, explainers, documentary. Registers: keynote / documentary / editorial lean here.

Applying ONE composition to every peak is monoculture — the same failure the motion review caught. Pick per beat: a 15s explainer might run column peaks throughout and save the subject lockup for the single apex. (A third pattern — the editorial split, bg context upper + fg hero lower-third — lives in references/composition-craft.md and is the next placement candidate.)

The motion language (five layers, each DNA answers all five)

A DNA’s animation is not a parameter tweak on a shared fade — it’s a distinct physics:

Shipped signatures: cream = light condenses (blur-settle, drift-up exits) · ink = letterpress (scale-press, no float, dead-still) · editorial = the pen glides (x-glide with the italic, cascade-out page turns) · keynote = surgical reveal (burn words + line wipe + hero wipe-up, exits never move) · documentary = burn and hold (motion’s absence IS the language) · loud = percussion (back-eased punches, alternating tilt, hero ripple that bumps every caption — never the footage).

Rules inherited from hyperframes motion-principles: exits faster than entrances; vary ease families BETWEEN DNAs (not within one); .out enters, .in leaves; never two transform tweens on one element in the same window (wipe-up is clip-path so the afterglow scale channel stays free; loom subsumes breathe).

Adding a DNA

Copy an existing dna/<name>.json, change the voice, keep the schema. The engine (modes/cinematic/engine.html) and both compilers consume it as-is. A DNA earns its place by being a distinct voice with a reason to exist — not a recolor.