Parallax landing page

01What is it?
Build a scroll-driven cinematic landing page from a short video. The user provides a 5, 15 second video (often AI-generated); this skill extracts every frame at HD JPEG quality, then produces a single-hero HTML page. The value is a focused slice of conversion optimization judgment, useful when several similar skills cover the same ground.
02Inputs
Context for conversion optimization: your goals, audience, constraints, and any source material the skill asks for.
03Output
A ready-to-use result for conversion optimization: the analysis, copy, or recommendations the agent produces.
Install-only

Install as a package

Installs this one skill package for your coding agent, including any supporting files that skill ships with — not every skill in the repository. Read the tutorial.

Terminal
$ npx skills add hoodini/ai-agents-skills --skill parallax-landing-page

Skill instructions

The instruction file for this skill. The skill also includes other files you need to install to use it.

SKILL.md

Parallax Landing Page

Turn a short video into a one-screen cinematic landing page where the user's scroll gesture scrubs through the video frame-by-frame while five dramatic text scenes crossfade in and out. The document never actually scrolls — wheel/touch/keyboard input is intercepted in JS and converted into virtual frame progress.

This skill encodes the patterns proven in the bundled reference at ../../examples/parasites/ (the github / lion / hope landings — three working pages built with this exact skill). Treat that folder as the ground truth if anything here is ambiguous.

Where this skill sits in the YUV.AI pyramid

parallax-landing-page is in the middle tier of the YUV.AI skills pyramid, alongside yuv-design-system, yuv-decks, yuv-viral-video, and video-to-landing-page. The top-tier orchestrator yuv-pilot routes here whenever a 5–15s video clip needs to become a single-hero scroll-scrub landing.

This skill brings its own visual language (Anton + Caveat + Inter, gold/amber/accent/cream accent palette) — it does NOT call into yuv-design-system. The two systems coexist: yuv-design-system owns the brand palette across web/app/deck surfaces, parallax-landing-page owns the cinematic scroll-scrub aesthetic specifically. For YUV.AI parallax landings, you can OPTIONALLY add a phoenix watermark + the canonical Linktree URL in the footer area as a brand throughline — see yuv-design-system/assets/ for the watermark file. Do NOT swap the Anton/Caveat type stack for the design system's Neon palette — that would lose the signature look.

What you're building

Each landing is a single locked viewport with:

  • A canvas filling the screen, drawing one frame at a time (cover-fit, DPR ≤ 2)
  • 5 absolutely-positioned text scenes (kicker + uppercase Anton headline + Caveat handwritten line), each with its own scroll-progress window
  • A fixed nav with PARALLAX // showcase logo and links to sibling landings
  • A loader that preloads every frame before letting the user interact
  • A vignette + film grain overlay
  • A bottom scroll-hint that fades out once the user starts scrolling
  • A bottom-right frame counter (e.g. 073 / 145)
  • A final scene that is the CTA section — no separate .end section below

Total weight at q:v 2: roughly 120–220 KB per frame × frame count. A 6-second 24fps clip = 145 frames ≈ 25 MB. A 10-second clip = 241 frames ≈ 53 MB.

Save location

Default: ~/Documents/yuv-projects/landings/<slug>/ — always save standalone landings here so you can find them again. Override only if the user explicitly picks a different location OR if Mode B (integrate into the existing examples/parasites/ showcase) is selected.

mkdir -p ~/Documents/yuv-projects/landings
cd ~/Documents/yuv-projects/landings
# The skill creates <slug>/ here as it extracts frames and builds the HTML.

Final path (Mode A — standalone): ~/Documents/yuv-projects/landings/<slug>/. Final path (Mode B — integrate): the existing showcase directory the user chose.

Tell the user the final path at the end of every build.


Workflow

There are four phases. Don't skip phases — each one feeds the next.

Phase 1 — Discover

Confirm with the user:

  1. Video path — where the source MP4 lives.
  2. Slug — short folder/file name (e.g. hope, marcus, desk). One word, lowercase, no spaces. This becomes both the frame folder name (<slug>/) and the HTML filename (<slug>.html).
  3. Theme / topic — what is this landing about? A product launch? A personal moment? A wildlife encounter? You need this to write the copy.
  4. Project mode — standalone (brand new folder) or integrate into the existing parasites showcase at examples/parasites/ (top of this repo). If the user says "add it to the showcase", they mean Mode B below.
  5. Accent color — for the script/CTA highlight: gold (#f5b042), amber (#ff8a3d), accent (pink #ff4d6d), cream (#f5e9d4), or something new (add it to :root in style.css and write a new .script.<name> rule).

If the user hasn't specified some of these, propose defaults and move on — don't stall on questions they can answer once they see the result.

Phase 2 — Extract frames

Run the bundled extraction script. It probes the video with ffprobe and extracts every frame at native resolution with -q:v 2 (near-lossless JPEG), naming them frame-001.jpg through frame-NNN.jpg (3-digit zero-padded).

python <skill>/scripts/extract_frames.py <video_path> <output_folder>

Example:

python C:/Users/User/.claude/skills/parallax-landing-page/scripts/extract_frames.py \
    C:/Users/User/Documents/parasites/hope/hope.mp4 \
    C:/Users/User/Documents/parasites/hope

The script prints a JSON metadata block to stdout. Capture it — you need frame_count and suggested_scroll_budget for the HTML.

Why -q:v 2: ezgif and similar tools re-compress aggressively (~55 KB/frame). Native ffmpeg at q:v 2 gives ~120–220 KB/frame at the same dimensions — visibly sharper. Do not downscale; never upscale (it inflates files without adding detail).

Why the scroll-budget formula (≈26 px per frame): keeps the tactile "pixels of scroll per frame advanced" constant across clip lengths. 145 frames → ~4200, 241 frames → ~6300, 90 frames → ~2500 (floor). The script computes it for you and clamps to [2500, 8000].

Phase 3 — Author the page

Read the template at <skill>/assets/landing-page-template.html. It is a complete working file with {{PLACEHOLDER}} markers and `` comments next to creative bits. Write a new file at <target>/<slug>.html with everything substituted.

Substitution checklist (every placeholder must be replaced):

PlaceholderWhat to putExample
{{PAGE_TITLE}}HTML <title>Hope — The Fastest Heart in the Negev
{{PAGE_DESCRIPTION}}meta description, one sentenceA moment with Hope, the cheetah of Midbarium.
{{SLUG}}folder name (no path, no slash)hope
{{FRAME_COUNT}}integer frame count241
{{FRAME_COUNTER_TEXT}}001 / NNN with matching width001 / 241
{{SCROLL_BUDGET}}integer from the extractor JSON6300
{{PREVIEW_FRAME_MID}}zero-padded middle frame121
{{PREVIEW_FRAME_END}}zero-padded last frame241
{{LOADER_TITLE}}one word, uppercaseHOPE
{{LOADER_SCRIPT}}handwritten line ending in summoning the runner…
{{NAV_LINKS_HTML}}full <a> tag block for navsee below
{{SCENE_N_*}}kicker, headline, handwritten line for each of 5 scenessee copy guide
{{SCROLL_HINT_TEXT}}2–4 word uppercase promptSCROLL TO RUN
{{CTA_BUTTONS_HTML}}2–4 <a class="cta"> tagssee below
{{ACCENT_CLASS}}accent color suffix for .script.X and .cta.Xamber

Read references/copy-guide.md for the cinematic copy patterns (kicker phrasing, headline length, handwritten line tone). The copy is where this skill earns its keep — generic copy will sink an otherwise beautiful animation.

Phase 4 — Wire up assets

If the target folder already has style.css and parallax.js (showcase integration), do nothing — they're shared. Otherwise copy both from <skill>/assets/ into the target folder:

cp <skill>/assets/style.css <target>/style.css
cp <skill>/assets/parallax.js <target>/parallax.js

If integrating into an existing showcase, also do the steps in references/showcase-integration.md — add a card to index.html, add a link to every existing landing's navbar, and add the chain CTA on the previous landing's final scene.

Tell the user how to view:

Open <target>/<slug>.html via python -m http.server 8000 from the parent folder (not file:// — Chromium can taint the canvas with local image origins).

File contract

<target>/
├── <slug>/                  # frame folder
│   ├── <slug>.mp4           # the source (preserved)
│   ├── frame-001.jpg
│   ├── frame-002.jpg
│   └── … frame-NNN.jpg
├── <slug>.html              # the landing page
├── style.css                # shared (copy if standalone)
├── parallax.js              # shared (copy if standalone)
└── index.html               # the hub (only if standalone or integration)

The frame folder name and the HTML filename MUST share the same slug — the folder option in the page's new ParallaxPage({folder: '<slug>', ...}) call is the literal folder path.

The ParallaxPage API (already in parallax.js, do not redefine)

new ParallaxPage({
  folder:       '<slug>',     // folder containing frame-001.jpg, etc.
  frameCount:   145,          // integer
  prefix:       'frame-',     // always 'frame-' for this skill
  scrollBudget: 4200,         // pixels of accumulated input for full pass
  // optional: padWidth (default 3), ext (default 'jpg')
})

The class auto-attaches body.scrub-page (which locks scroll via CSS), preloads all frames, runs an RAF-driven lerp (0.22 factor) toward the scroll-derived target frame, and crossfades scenes whose data-start/data-end ranges include the current progress.

Hard constraints (the non-negotiables)

These are baked into the reference and break the experience if violated:

  • 5 scenes, fixed progress windows: 0.00–0.20, 0.22–0.40, 0.42–0.60, 0.62–0.80, 0.82–1.00. The 0.02 gap between scenes is intentional — it's a moment of "just the image, no text" between beats. The hero (scene 1) and final (scene 5) use data-fade="0.30" and data-fade="0.18" respectively so they hold longer at the edges; middle scenes use the default 0.25.
  • No .end section beneath the scrub. The final scene contains the CTAs. The whole page is one locked viewport.
  • Anton + Caveat + Inter — uppercase Anton for everything bold, Caveat for handwritten accents (gold/amber/cream), Inter for any body lead text. Never substitute, never add a fourth font.
  • Off-white #f5f1ea, never #fff — already in CSS as var(--text).
  • Frame names are frame-NNN.jpg zero-padded to the width of the total (3 digits is the standard since clips < 1000 frames; the JS auto-computes counter width from the total).
  • The slug is the folder name AND the HTML filename basenamehope/, hope.html. No exceptions; the integration steps assume this.

Where to look when you get stuck

  • The reference showcase bundled in this repo: examples/parasites/ — github.html, lion.html, hope.html, style.css, parallax.js, index.html. If your output diverges visibly from those, you've drifted; reread.
  • references/copy-guide.md — how to write the 5 scenes so they actually feel cinematic.
  • references/showcase-integration.md — exact edits needed when adding to an existing hub.
  • assets/landing-page-template.html — the canonical structure with placeholders.
  • assets/hub-template.html — only used when creating a brand-new showcase (3 cards from scratch).

Two real example invocations

Standalone: user has a single clip desk.mp4 and wants a landing page.

  1. Ask for slug + theme. Run extractor on desk.mp4./desk/frame-NNN.jpg.
  2. Copy style.css + parallax.js to project root.
  3. Author desk.html from template. Nav has just one link: itself.
  4. Done. Tell user to serve with python -m http.server.

Showcase add: user adds hope.mp4 to the existing parasites folder.

  1. Slug = hope. Extract → hope/frame-NNN.jpg. (Reuse existing style.css + parallax.js.)
  2. Author hope.html from template.
  3. Run showcase integration: add a third card to index.html, append a Hope link to the navbar in github.html and lion.html, chain the previous landing's final scene with a Next: Hope → CTA.
  4. If introducing a new accent color, add it to :root and add .script.X + .cta.X rules in style.css.

Supporting file: references/copy-guide.md

Copy Guide — writing the five scenes

The animation is half the work. The copy is the other half. Generic copy will sink an otherwise beautiful landing. Read this before you write a single word of headline.

The structural rule

Each of the 5 scenes is exactly three lines:

[KICKER]            ← 1–3 words, all caps, Anton, letter-spaced
[HEADLINE]          ← 2–4 words, broken across 2 lines with <br>
[HANDWRITTEN LINE]  ← 4–8 words, one full sentence, ends with a soft punctuation

The kicker frames the moment ("Chapter 02", "— A True Story —", "01 — Workflow"). The headline is the punch. The handwritten line is the emotional follow-through — the line that earns the trust the headline asked for.

Length matters because the typography is huge. A 6-word headline at 200px breaks the layout. Keep it tight.

The five-beat arc

This structure is borrowed from short documentary editing — you have ~25 seconds of attention; spend each beat well.

SceneBeatJob
1HookIntroduce the subject. Name it. State its essence in one phrase.
2OriginWhere does it come from? What is its world?
3StakesWhy does it matter? Where does the story take place?
4MomentThe personal turn — the speaker's act, the decision, the proximity.
5Resolution + CTAWhat happened. Then the call to action.

You're not writing about a product. You're writing about a moment that includes a product.

Voice rules

  • Anton headlines are commands or noun-phrases, never sentences. "Born to Run." "Built for Flow." "Meet Marcus." Not "He was born to run." The handwritten line is where sentences live.
  • Handwritten lines are colloquial, almost spoken. "the white lion of the Negev." "an AI that finally ships with you." "and knelt with food in my hands." Lowercase first word; trail off with a period (or only for the loader script).
  • Avoid adjectives in the headline. The image already shows the adjective. "Beautiful Cheetah" is wasted ink. "Born to Run" lets the image do the work.
  • Specifics beat abstractions. "Be'er Sheva" beats "the desert". "macOS · Windows · Linux" beats "every platform". "70 mph" beats "fast".

Three real examples — read these carefully

GitHub Desktop (product)

SCENE 1
— A NEW CHAPTER —
GitHub
Desktop
the new build. native. fast. yours.

SCENE 2
01 — WORKFLOW
Where
Code Lives.
branches, PRs, reviews — at your fingertips.

SCENE 3
02 — PERFORMANCE
Built
For Flow.
native speed. zero friction. all platforms.

SCENE 4
03 — INTELLIGENCE
Copilot
Inside.
an AI that finally ships with you.

SCENE 5 (CTA)
— NOW AVAILABLE —
Ready?
macOS · Windows · Linux.
[ Download for macOS ] [ Windows ] [ Linux ] [ Next: Marcus → ]

Marcus the white lion (personal moment, wildlife)

SCENE 1
— A TRUE STORY —
Meet
Marcus.
the white lion of the Negev.

SCENE 2
CHAPTER 01
Born
of the Desert.
a coat like snow over a land that burns.

SCENE 3
CHAPTER 02
Midbarium.
Be'er Sheva.
where wilderness meets wonder.

SCENE 4
CHAPTER 03
I Crossed
the Fence.
and knelt with food in my hands.

SCENE 5
— THE LAST FRAME —
And the
King Ate.
a moment frozen in time.
[ Visit Midbarium ] [ Open in Maps ] [ Next: Hope → ] [ ← Back ]

Hope the cheetah (speed, motion, intimacy)

SCENE 1
— A TRUE STORY —
Meet
Hope.
the fastest heart in the Negev.

SCENE 2
CHAPTER 01
Born
to Run.
seventy miles an hour of wild grace.

SCENE 3
CHAPTER 02
Midbarium.
Be'er Sheva.
where the desert opens — she answers.

SCENE 4
CHAPTER 03
I Came
Close.
she looked at me — and let me stay.

SCENE 5
— THE LAST FRAME —
A Second
of Stillness.
with the fastest cat alive.
[ Visit Midbarium ] [ Open in Maps ] [ ← Back to showcase ]

The accent color picks a tone

  • gold (#f5b042) — warmth, reverence, mythic ("Marcus")
  • amber (#ff8a3d) — heat, motion, vitality ("Hope")
  • accent (pink #ff4d6d) — product/tech, modern, slightly playful ("GitHub")
  • cream (#f5e9d4) — neutral, used inside scenes 2 and 4 for variety so every line isn't the same color

The convention in the references: scenes 1, 3, 5 use the page's signature accent; scenes 2 and 4 use cream for visual rhythm. Follow that unless you have a reason not to.

Loader script tone

The loader sits on screen for about 1–3 seconds while frames preload. The handwritten line beneath the giant title should be a tiny hook — present-tense, ending in .

Good examples (from the reference):

  • loading the new build…
  • summoning the king of the Negev…
  • summoning the runner…

Bad: Please wait while frames load. (boring, breaks tone)

Scroll hint tone

Short. Uppercase. 2–4 words. Often a verb tied to the subject:

  • SCROLL TO PLAY (GitHub — software metaphor)
  • SCROLL TO ENTER (Marcus — entering a kingdom)
  • SCROLL TO RUN (Hope — running with the cheetah)

If you can't find a verb that suits the subject, fall back to SCROLL TO BEGIN.

CTAs in the final scene

2–4 CTAs, the first one solid or accent-colored, the rest outlined. Keep them short — 2–4 words each. End with a chain link (Next: X →) if this is part of a showcase, and a back-to-hub link (← Back to showcase).

Available CTA classes:

  • cta — outlined off-white (default)
  • cta solid — filled off-white on bg
  • cta gold — outlined gold (good for Midbarium / nature subjects)
  • cta amber — outlined amber (Hope-style)

If introducing a new accent, add .cta.<name> rules to style.css in the same pattern as .cta.gold.


Supporting file: references/showcase-integration.md

Showcase Integration — adding a new landing to an existing project

Use this when the target folder already has an index.html (hub) and at least one other landing (e.g. github.html + lion.html). You're not creating from scratch — you're growing what's there.

There are four edits, all small. Do them in this order so the user can preview after each step if they want.

1. Add a card to index.html

Inside <div class="hub-cards">, add a new <a class="hub-card {{slug}}"> block alongside the existing cards. The grid is already repeat(3, 1fr) and collapses gracefully — you don't need to touch it.

Pattern (copy and adapt):

<a href="{{slug}}.html" class="hub-card {{slug}}">
  <div class="hub-card-bg" style="background-image: url('{{slug}}/frame-{{mid}}.jpg');"></div>
  <span class="dot"></span>
  <span class="hub-card-arrow">Enter →</span>
  <div class="hub-card-content">
    <span class="kicker">{{NN}} — {{CATEGORY}}</span>
    <h2>{{HEADLINE_LINE_1}}<br />{{HEADLINE_LINE_2}}</h2>
    <p class="script" style="color: var(--{{accent}});">{{HANDWRITTEN_LINE}}</p>
  </div>
</a>

Notes:

  • {{mid}} is the middle frame number (e.g. 121 for a 241-frame clip) — gives the card a more visually arresting still than frame-001.
  • {{NN}} is the position in the showcase: 01, 02, 03, etc. Match the order they appear.
  • {{CATEGORY}} is a one-word taxonomy (Product, Wild, Speed, Story, Field, etc.).
  • The style="color: var(--{{accent}});" on the handwritten line ties the card to its landing page's accent color.

Also bump the hub copy to reflect the new count:

<div class="hub-eyebrow">{{N}} Stories · One Scroll</div>
<span class="script">{{n}} short films, one page each.</span>

Spell N as a word ("Three", "Four", "Five") to match the existing typography rhythm.

Add the preload link in <head>:

<link rel="preload" as="image" href="{{slug}}/frame-{{mid}}.jpg" />

2. Add a dot accent style to style.css

If the new landing uses an existing accent color (gold, amber), reuse the dot rule. If it's a new color, add both the variable and the dot rule.

Add the color to :root (next to the existing --accent, --gold, --amber):

--{{newcolor}}: #HEXVAL;

Add the dot rule right after the existing .hub-card.lion .dot / .hub-card.hope .dot lines:

.hub-card.{{slug}} .dot {
  background: var(--{{newcolor}});
  box-shadow: 0 0 0 4px rgba(R, G, B, 0.18);
}

The RGB triplet should match the hex. (E.g., #ff8a3d255, 138, 61.)

And if you want the handwritten lines on that landing to use this color, also add:

.script.{{newcolor}} { color: var(--{{newcolor}}); }
.cta.{{newcolor}}    { border-color: var(--{{newcolor}}); color: var(--{{newcolor}}); }
.cta.{{newcolor}}:hover { background: var(--{{newcolor}}); color: var(--bg); }

3. Add the new landing to every existing nav

In index.html, github.html, lion.html, and any other landing, find the .nav-links block and add:

<a href="{{slug}}.html">{{Title}}</a>

On the new landing's own page, the link gets class="active". On every other page, no class.

Maintain a stable order across all pages — easiest is alphabetical or "by show-order" (i.e. matching the hub card order). Either is fine; just be consistent so the nav doesn't jitter as the user clicks around.

4. Chain the previous landing's final scene

In the immediately previous landing (the one with the highest existing number), update the end-actions block inside its scene 5 to add a chain link to the new landing.

Before:

<div class="end-actions">
  <a class="cta gold" href="...">Visit Midbarium</a>
  <a class="cta" href="...">Open in Maps</a>
  <a class="cta" href="index.html">← Back to showcase</a>
</div>

After:

<div class="end-actions">
  <a class="cta gold" href="...">Visit Midbarium</a>
  <a class="cta" href="...">Open in Maps</a>
  <a class="cta" href="{{slug}}.html">Next: {{Title}} →</a>
  <a class="cta" href="index.html">← Back to showcase</a>
</div>

On the new landing's own scene 5 CTA block, end with ← Back to showcase and (if appropriate) a chain forward to whatever you'd want next.

Verification checklist

After all four edits, the user should be able to:

  1. Open index.html → see N cards in the grid, all with the right backgrounds and accent dots.
  2. Click each card → land on the right page, with the loader title matching.
  3. Use the nav from any page → reach any other page; the active page is underlined.
  4. Scroll to the end of any non-final landing → see a Next → CTA pointing to a real page.
  5. Scroll to the end of the new landing → see the back-to-showcase CTA.

If any of those breaks, you missed an edit. Re-check nav-links parity across files.

Common slips to avoid

  • Forgetting to add the slug to the preload list in index.html. Cards without a preload still load, just a beat later.
  • Mixing accent colors on a single card. The card BG accent (dot + script color) should match the landing's interior accent — otherwise the hub feels disconnected from the page.
  • Reusing a frame number that's the wrong padding. Always pad to 3 digits (frame-073.jpg, not frame-73.jpg).
  • Forgetting to set class="active" on the current page's nav link. Subtle but breaks the "where am I" affordance.
  • Adding a new accent variable but forgetting the .script.<name> and .cta.<name> rules. The page renders but the accent never appears anywhere except the dot.

How do I install Parallax landing page in Cursor, Claude Code, or Codex?

Run npx skills add hoodini/ai-agents-skills --skill parallax-landing-page in the project where you want it, then ask your agent for the skill by name. The --skill flag installs only Parallax landing page, not every skill in the repository.

Where does Parallax landing page come from and what license is it under?

Parallax landing page comes from the hoodini/ai-agents-skills repository on GitHub. That repository has 274 GitHub stars. No license was detected on the source repository, so check with the author before redistributing it.

Prefer plain text? Read the Parallax landing page guide as markdown.