convez.ai
SkillDesenvolvimento WebIntermediário
MustBeSimo

Cinematic Scroll

MustBeSimo/cinematic-scroll-skill

Websites cinematográficos scroll-driven gerados por IA em minutos

Ver no GitHub
InicialPluginScore · de 0 a 100

Skill para agentes de IA que capacita a construção de websites scroll-driven com estética cinematográfica. Gera sistemas visuais, storyboards de movimento, parallax multi-camadas, capítulos fixos e templates Next.js prontos para produção, partindo apenas de uma descrição textual.

Instalação

$npx convez-ai MustBeSimo/cinematic-scroll-skill

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
54
Forks
4
Acompanhando
0
Saúde
42%
Último commit
há 1 semana
Versão
v3.0.1
Issues abertas
2
Contribuidores
3

Casos de uso

  • Páginas de lançamento de produtos com movimento fluido
  • Sites editoriais e portfolios com efeitos parallax avançados
  • Histórias visuais multi-capítulo com navegação fixa
  • Demos e showcases com 3D tilt e revelaç ões sincronizadas ao scroll

Veredito

Skill bem arquitetada que democratiza a criação de experiências scroll complexas, reduzindo significativamente o tempo entre briefing visual e implementação. Pronto para produção, agnóstico a agente, e com excelente documentação e exemplos executáveis.

README

A tradução para o português deste README ainda está na fila.

Cinematic Scroll

One sentence in. A cinematic, scroll-driven website out.

npm version Install via skills.sh Works with Claude · Cursor · Hermes License: MIT GitHub stars

👉 Visit the live landing page — built with the skill itself. It adapts to your GitHub theme: dark → Petroleum Editorial, light → Swiss Museum. ▶ Watch the 30-second demo → — one sentence in, a cinematic scroll site out.

↑ The motion is the skill — pinned chapters, multi-depth parallax, mask/stagger title reveals, a tracking index rail. The look is whatever you ask for. This clip happens to be a Renaissance editorial; the same grammar drives a brutalist drop, a neon Gen-Z launch, a noir game page, or your brand. Scroll it live →

A free, MIT-licensed craft skill that gives any coding agent — Claude, Cursor, Hermes, OpenClaw — the taste to build cinematic, scroll-driven websites. Describe the aesthetic — palette, mood, references — and get a visual system, motion storyboard, pinned chapters, multi-depth parallax, 3D tilt, and full release pages art-directed to match. The motion is the constant · the look is yours · the agent is your choice. It's a skill, not a plugin — the craft travels with you across every agent.

Free for any use, personal or commercial (MIT). Actively developed — built from production work and shipped open source. Issues, PRs, and showcase submissions welcome — I collect what people build.

Built by Simone Leonelli · simone@w230.net


Get started — two paths

Pick how you want to build:

Mode A: Single scroll section — one runnable .html file, no build step, no keys. Perfect for a hero chapter or one-off section.
Mode B: Full release site — complete Next.js project, tested templates, optional AI image generation pipeline. Best for product launches and multi-chapter stories.


Install

Pick whichever channel fits your client. All four install the same skill.

Claude Code — plugin marketplace (recommended for Claude Code)

/plugin marketplace add MustBeSimo/cinematic-scroll-skill
/plugin install cinematic-scroll@mustbesimo

Installs as a namespaced plugin and stays updatable with /plugin marketplace update mustbesimo.

Any client — npx installer

npx cinematic-scroll-skill          # copies the skill into ~/.claude/skills/cinematic-scroll
npx cinematic-scroll-skill --dir .cursor/skills   # or a custom skills directory

Any client — git clone

git clone https://github.com/MustBeSimo/cinematic-scroll-skill ~/.claude/skills/cinematic-scroll

Registries

npx skills add MustBeSimo/cinematic-scroll-skill   # skills.sh

Also listed on agentskills.io (search "cinematic-scroll").

Platform-specific

Paths vary by client — see COMPATIBILITY.md for step-by-step instructions:

  • Claude Desktop — Settings → Capabilities → Skills → Upload
  • Cursor — drop into .cursor/skills/ (or npx cinematic-scroll-skill --dir .cursor/skills)
  • Hermes Agent — git clone to ~/.hermes/skills/
  • OpenClaw — openclaw skills install git:MustBeSimo/cinematic-scroll-skill@v2.1.0

Quick start

After installing, describe what you want to build in chat — see examples/PROMPTS.md for 20+ copy-paste examples across aesthetic worlds.


Live examples — five scrollable worlds, one grammar

Same engine, deliberately clashing aesthetics — five single, build-free index.html files (GitHub-Pages-native) running the skill's Mode A grammar: vanilla JS on requestAnimationFrame with optional GSAP/ScrollTrigger enhancements. All five retain a dependency-free core; Noir, Luxe, and Pop progressively enhance one showcase beat with deferred GSAP + ScrollTrigger (loaded from CDN with vanilla fallback). All five render fully with zero image files (CSS-only placeholders that upgrade when you add stills). Proof the look is a variable, not a default.

Same motion grammar; any aesthetic. These five show different visual directions the skill can art-direct — change the copy, palette, and references, and the same engine produces any world you describe. This isn't an exhaustive list of styles (the styling is infinite). It's a proof that the cinematic motion is the constant, and the look is whatever you ask for.

Aesthetic directions

Beyond these five live examples, the skill can art-direct itself into any visual world. Below are six concept style directions to illustrate the range:

Six style directions generated by the skill's own fal.ai pipeline (or bring your own images / go CSS-only at $0). These are concept stills that show possible aesthetic ranges — the five live examples above are the scrollable proofs.

Running locally

python3 -m http.server 8099   # then open /examples/renaissance/ · /studio/ · /noir/ · /luxe/ · /pop/

Under the motion, every chapter ships with:

Cinematic depth5–7 parallax layers per chapter, perspective camera, dolly-back transitions
Editorial typeOversized titles with word-stagger / clip-path mask / letter-spacing-scrub reveals
Atmosphere morphsBackgrounds crossfade between chapter color-worlds as you scroll
Image pipelineOptional: fal.ai-generated heroes (FLUX.2, Nano Banana, Imagen); required: bring your own images or render CSS-only visuals. Generated assets remain subject to your input rights and model-provider terms — review output before commercial deployment.
Bulletproof basicsReduced-motion fallback, iOS video safety, mobile-stacked layout, transform/opacity-only core hot paths; optional GSAP showcase enhancements in selected examples; no WebGL required. Validate performance on target devices before production.

✦ New: one choreography, two media

A single scroll-choreography.json now compiles to the website and its launch film — same beats, same easings, same depth choreography. Rebrand your site, and the video rebrands itself.

node compile-choreography.mjs scene.json --target web     # → GSAP ScrollTrigger page
node compile-choreography.mjs scene.json --target video   # → paused timeline for HyperFrames / Remotion
node compile-choreography.mjs scene.json --harness --out preview.html   # → watch it move, zero install
node compile-choreography.mjs scene.json --target hyperframes           # → render-ready composition → MP4

The web target is scroll-driven and responsive; the video target is a deterministic 16:9 timeline for HTML-to-video renderers — directed by FRAME.md, the brand spec that translates this design system for the frame. The --harness flag emits a self-contained preview HTML with play/scrub controls so you can watch any choreography in a plain browser.

Same DOM contract ([data-chapter] / [data-layer] / [data-title]) serves both targets. Full mapping table + Remotion adapter: scroll-choreography-compilation.md · mixing strategy (HyperFrames × Remotion): video/PIPELINE.md · reference film project: video/ship-in-5/


Quickstart

Mode A — instant scroll section

"Use cinematic-scroll to build a self-contained HTML pinned hero chapter for [YOUR BRAND]. Include a progress HUD."

You get one runnable .html file. Open it. Done.

Mode B — full release site

"Use cinematic-scroll to scaffold a complete Shopify-Editions-tier release page for [YOUR PRODUCT IN ONE LINE]. Demo mode first — do not require my fal.ai key. Copy all bundled templates verbatim. 8 chapters. Finish with the exact commands to run."

Then, in the scaffolded project:

npm install
npm run dev

Open http://localhost:3000 — a full 8-chapter cinematic page, CSS-only visuals, zero AI setup.

Want real generated chapter art? Add your own fal.ai key and run the command below. Generation cost varies by model and resolution — see MODELS.md and current fal.ai pricing before running a batch.

npm run setup        # interactive key wizard → writes .env.local
npm run generate     # generates all chapter heroes into public/generated/

Full walkthrough: examples/GETTING_STARTED.md. Model menu + costs: MODELS.md.


What's in the box

cinematic-scroll-skill/
├── SKILL.md                  # the agent contract (Mode A + Mode B). For Claude, not humans.
├── README.md                 # you are here
├── COMPATIBILITY.md          # platform installation & verification (Claude, Cursor, Hermes, OpenClaw)
├── LICENSE                   # MIT
├── manifest.json             # skill metadata (free)
├── MODELS.md                 # fal.ai model menu, costs, when-to-use
├── compile-choreography.mjs  # ✦ scroll-choreography.json → page / film / preview
├── scroll-choreography.json  # the declarative choreography schema + example
├── FRAME.md                  # brand spec for video agents (the frame translation)
├── taste-guardrails.md       # banned patterns, cinematic vocabulary, pacing rules
├── references/               # scroll patterns · film archetypes · performance budget
├── video/                    # PIPELINE.md (HyperFrames × Remotion) + film projects
├── examples/
│   ├── PROMPTS.md            # 20+ trigger prompts across aesthetic worlds
│   ├── GETTING_STARTED.md    # fal.ai setup, troubleshooting, queue+webhook
│   ├── KNOWN_ISSUES.md       # QA log of real failure modes + fixes
│   ├── renaissance/          # Mode A example — warm classical editorial
│   ├── studio/               # Mode A example — brutalist creative-director portfolio
│   ├── noir/                 # Mode A example — sci-fi noir (VANTASCOPE)
│   ├── luxe/                 # Mode A example — quiet luxury (Maison Solenne)
│   └── pop/                  # Mode A example — Gen-Z pop (BLOOM)
└── templates/nextjs/         # tested, copy-verbatim Next.js App Router project
    ├── app/ (+ api/fal/*, generate-edition-asset)
    ├── components/ (ChapterScene, ChapterDemoVisual, EditionsPage, SmoothScrollProvider)
    ├── lib/ (editions-manifest, fal-*, prompt-contract, use-lenis, use-device)
    ├── scripts/ (setup.mjs, generate-chapter-assets.mjs)
    └── package.json, tailwind.config.ts, tsconfig.json, …

Peer dependencies (in the consuming app)

npm install choreo-3d framer-motion gsap @gsap/react lenis @fal-ai/client @fal-ai/server-proxy

The motion primitives target the choreo-3d package, with a built-in vanilla fallback (sticky + IntersectionObserver + rAF) for sandboxes where npm packages can't be installed — identical math, same keyframes.

On GSAP: as of the 2025 Webflow acquisition, GSAP is 100% free — every former Club plugin included (SplitText, ScrollSmoother, ScrollTrigger, MorphSVG…), commercial use too. The Next.js build (Mode B) uses ScrollTrigger + SplitText for pinning and title reveals; the standalone demos retain a vanilla rAF core with optional deferred GSAP enhancements in selected showcase beats (loaded from CDN with vanilla fallback). This means Mode A files run from file:// with zero build, and gracefully fall back to CSS-only rendering if the CDN is unavailable. For low-level GSAP help, pair this with the official greensock/gsap-skills — that skill teaches the GSAP API; this one teaches the cinematic system on top.


Originality & legal

The reference direction is Shopify Editions, used only as an art-direction benchmark — chaptered release storytelling. The skill never copies Shopify's assets, logos, copy, source, or exact compositions, and never bakes readable UI text into generated images or imitates a named living artist. Generated assets may be used subject to fal.ai, model-provider and input-rights terms. Review output before commercial deployment.

License

MIT © 2026 Simone Leonelli — see LICENSE.

Built something with it? I'd genuinely love to see it: simone@w230.net

Ver o original no GitHub

Bom para

Relacionados