Cinematic landing

Cinematic Landing — design system + composition rules

This is a fat skill: the markdown is most of the value. The runtime code under components/landing/cinematic/ is a few hundred lines of CSS + four small React primitives. Everything that makes a cinematic landing feel cinematic — the design tokens, the motion timings, the copy patterns, the Pexels query strategy — lives here. Smarter Claude models produce better landings from the same spec without any code change.

The skill auto-applies to two archetypes during create_full_workspace:

• cinematic-aspirational — coaching, medspa, wellness, fitness, salons, premium lifestyle, anywhere the customer buys the dream before the service
• technical-restrained — marketing/dev agencies, consultancies, technical SaaS, fractional executives, anywhere the buyer is sophisticated and the work itself signals quality

For other archetypes the hero falls back to the previous variant (left-aligned-asymmetric, split-screen-50-50, cinematic-fullbleed, founder-portrait). A craft-roofer doesn't need a cinematic video hero; their proof is the work itself.

The four pillars

A page is "cinematic" when all four are present. Each is independent — adding three of four feels off; adding all four feels like Aura.

1. Looping MP4 background (full-bleed)

• Sourced from Pexels via searchPexelsVideo() in lib/assets/pexels.ts
• One call per workspace creation, URL persisted into the hero JSONB (the rendered page hits Pexels' CDN directly forever after — no live API calls at render time)
• HD MP4 ≤1080p, ≤30s duration, landscape orientation
• Wrapped in the FadingVideo component (rAF-driven 500ms crossfade at start + 0.55s lead-out before end → seamless manual loop, no CSS jank)
• A subtle bottom-up gradient overlay (from-black/30 via-black/10 to-black/55) for headline legibility — Aura ran no overlay but our copy is longer and the safety margin is worth it
• Bottom-right Video by NAME on Pexels attribution pill (required by Pexels licence; rendered in text-white/60 so it doesn't fight the composition)

2. Liquid-glass chrome

Two variants of the same glass treatment, defined in components/landing/cinematic/styles.css:

• .cin-liquid-glass — subtle (4px blur) for badges, secondary CTA pills, chip tags, card surfaces
• .cin-liquid-glass-strong — heavy (50px blur) for the primary CTA only

The glass effect comes from a ::before pseudo-element with a vertical white-fade gradient masked to a 1.4px border. Pure CSS, zero JS.

3. Instrument Serif italic typography

The headline (and only the headline) renders in Instrument Serif. Italic treatment on the shiny word (see below) is what gives the cinematic genre its signature look. Body copy stays in the page's archetype-default body font (typically Geist) for legibility.

4. Framer Motion entrance choreography

Layered staggered entrances:

1. Badge: 0ms, fade up y:10 → 0, duration 0.5s
2. Headline: 100ms, word-by-word blur-in via BlurText (3-step keyframe blur 10→5→0, opacity 0→0.5→1, y 50→-5→0, per-word stagger 100ms)
3. Subhead: 500ms, fade up y:16 → 0, duration 0.6s
4. Proof tile: 600ms
5. CTAs: 700ms

Past the entrance, the only running animation is the shiny-word gradient shimmer (6s linear infinite, pure CSS).

Headline copy rules (cinematic-aware)

The hero block's universal Hormozi rules still apply (quantification, specificity, no throat-clearing). On top of those:

• Pick one shiny word. The gradient-shiny treatment works best on exactly one word per headline. Two words feels noisy; zero words feels like a missed opportunity. Pick the outcome noun — what the visitor actually wants to walk away with.
• Lean italic-friendly. Instrument Serif italic is gorgeous on nouns and verbs; awkward on numbers and acronyms. "Pipeline" reads beautifully italicized; "401k" doesn't. If the proof IS a number, put it in the kicker or subhead, not the headline word that gets the shiny treatment.
• Two-line headlines work best. The Aura reference uses "Your email."
  • "Revitalized." Two lines of ~3 words each beats one line of 6 words in the cinematic genre. Both lines render via BlurText as a single string; the layout flex-wraps naturally.

Pexels query patterns (per archetype)

The Pexels API returns its best match for a 2-5 word query. Match the operator's outcome, not their literal vertical.

cinematic-aspirational (coach, medspa, wellness, lifestyle)

Niche	Query
Fitness coach	sunset beach running
X / social growth coach	phone scrolling social media
Executive / career coach	city skyline office window
Medspa / dermatology	spa water reflection slow
Yoga studio	yoga sunrise studio
Wellness / holistic	candle flame slow motion
Luxury salon	hair flowing slow motion

technical-restrained (agency, consultancy, B2B)

Niche	Query
Design / creative agency	abstract design motion graphics
Dev agency / engineering	code on screen close up
Marketing agency	macbook typing close up
Brand studio	neon city night drive
Strategy consultancy	team meeting office cinematic
Fractional exec	executive office window light

The LLM may also pick its own query if it has better niche insight than this table. The orchestrator hands a hint via the prompt; the LLM's heroVideo_query overrides it when provided.

Soft-fail strategy

Every external dependency in the cinematic stack soft-fails:

• No PEXELS_API_KEY → no video → hero renders branded gradient empty state (deep navy radial + giant ghost word) which still looks intentional
• Pexels rate-limit / 5xx → same fallback
• Pexels returns 0 results → same fallback
• Video URL 404s on the user's browser → <video> element shows the poster image + subtle gradient; nothing breaks
• Missing shinyWord → headline renders fully white (still beautiful)

The cinematic landing is strictly additive to the existing four hero variants — never replacing them, only opting in when archetype + Pexels both cooperate.

What this skill is NOT (yet)

Phase 1 ships the hero. Phase 2 will extend the cinematic system to:

• about — liquid-glass card stack with founder photo + body
• services — Aura "Capabilities" grid (liquid-glass cards, icon top-left, tag pills top-right, title + body bottom, min-height 360px)
• testimonials — figure cards with quote + author + company uppercased
• pricing — giant watermark headline + glass tier cards with white border + Pro tier highlighted
• cta — final liquid-glass full-bleed radial-glow CTA

Phase 3:

• Marketplace listing — fork this skill, customize the palette + queries
• Light-mode variant (cinematic structure on white for builders who want the choreography without the dark canvas)
• Per-niche curated query libraries (top 20 verticals with hand-picked Pexels IDs, no live search needed)

Verification checklist (operator side)

When this skill applies, the rendered page should:

• Show a looping video background that crossfades smoothly between loop cycles (no hard cut at the seam)
• Display the Pexels photographer credit in the bottom-right
• Render the headline in Instrument Serif italic with one word animated in the cyan→navy gradient shimmer
• Have a white pill primary CTA with a chevron that shifts on hover
• Have liquid-glass secondary CTA + badge + risk-reversal chips
• Render correctly on mobile (no horizontal scroll, headline word-wraps, video still plays — iOS Safari requires playsInline + muted, both set)
• Keep Lighthouse perf ≥80 (video is the heaviest asset; we cap at 1080p HD MP4 ≤30s precisely to stay under budget)

Antifragility note

This skill is ~90% markdown (this doc + the orchestrator prompt + the SKILL.md schema). The ~10% of code is:

• One 200-line Pexels client (lib/assets/pexels.ts)
• One 100-line FadingVideo component
• One 60-line BlurText component
• One 30-line AppleButton component
• One 200-line hero-cinematic-aura.tsx
• ~150 lines of CSS

When Opus 5 ships, it produces better headlines, better shiny-word picks, better Pexels queries from this same spec — zero code change required. The harness stays thin; the skill stays fat; the look gets better with every model upgrade.