30x-video

agent
Security Audit
Fail
Health Pass
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 54 GitHub stars
Code Fail
  • spawnSync — Synchronous process spawning in cli/cli.mjs
  • process.env — Environment variable access in cli/cli.mjs
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

One URL in, an agency-grade launch video out. A Claude Code skill (Remotion + React) with a 16-law taste codex — 12 brands, 12 worlds, zero templates.

README.md

30x web-to-video

One URL in. An agency-grade launch video out.

▶ Watch all 12 films · npm

A Claude Code skill that turns any brand's website into a 40-second launch video —
real brand assets, a 5-act narrative, word-aligned AI voiceover, beat-synced music,
and a taste codex of 16 hard-won design laws that keeps every film premium and
every brand distinct.

12 brands tested, 12 completely different visual worlds, zero templates — plus a
handful of earlier-generation films (retired from the public wall as the codex improved,
history preserved in git). Five of the current films were built end-to-end by
autonomous agents reading nothing but the rules.

Quick start

# one-shot: harvest a brand and build the video
npx 30x-web-to-video https://your-brand.com --agent

# or: install the skill user-wide, then just talk to Claude Code
npx 30x-web-to-video --global
# → open Claude Code anywhere and say: "make a launch video for your-brand.com"

Requirements & expectations

  • macOS / Linux (Windows untested)
  • Node 20+ and Claude Code (any paid plan)
  • One full run takes ~25 minutes and a meaningful chunk of agent tokens — it is doing real work: harvesting, designing five acts, rendering proof frames, self-reviewing
  • Best results with the strongest model your plan offers — the taste codex holds the floor on any model; the ceiling scales with the model
  • ELEVENLABS_API_KEY in the project .env gets premium narration; without it, narration falls back to local TTS (works, sounds simpler)
  • ffmpeg / yt-dlp / playwright auto-install during harvest

How it works

┌──────────────┐   ┌──────────────────┐   ┌──────────────────────────┐
│   HARVEST    │ → │   TASTE CODEX    │ → │   FIVE-ACT FILM          │
│ deterministic│   │ 16 design laws   │   │ Remotion + React         │
│ colors fonts │   │ 4 evidence gates │   │ ElevenLabs word-aligned  │
│ copy logo UI │   │ Claim→World map  │   │ VO · beat-synced BGM     │
└──────────────┘   └──────────────────┘   └──────────────────────────┘
  1. Harvest — a deterministic orchestrator scrapes the real brand: computed colors,
    loaded fonts, verbatim copy, logo, product screenshots, official demo video, BGM candidates.
  2. Taste — Claude designs five acts under the codex: no expanding rings, no heartbeat
    pulses, no divider bars, no text over full-bleed motion, one world per film, real assets
    over drawn abstractions. Every rule cites the incident that created it.
  3. Evidence gates — every asset is audited frame-by-frame before use; every act renders
    proof frames that get inspected; cuts snap to voiceover phrase onsets measured from
    word-level timestamps; downloaded music is whisper-checked for spoken watermarks.

What's in the box

skills/30x-web-to-video/
├── SKILL.md                  # the workflow: 8 steps, 4 blocking gates
├── rules/                    # the taste codex — 16 laws with case law
│   ├── taste.md              #   anti-slop blacklist (rings, pulses, bars, pills…)
│   ├── composition.md        #   zero-vacuum, full-bleed moments, unity of world
│   ├── narration-sync.md     #   continuous VO, cuts snap to phrase onsets
│   ├── artifact-catalog.md   #   selection priority: semantics > real assets > rotation
│   └── …
├── scaffold/                 # Remotion project template (transitions, shaders, captions)
└── scripts/                  # orchestrator, beat-sync, audiomap, visual-audit

The showcase

12 films are live on the public wall today. Numbering follows build order (some
earlier-generation builds were retired as the codex improved — history preserved in git,
so the numbers aren't consecutive).

# Brand World
01 Happy Model — AI gateway black/green routing world, global backbone
02 Caylent — AWS consulting black/mint infra, sailboat footage, agent decision graph
04 Parker — AI banking retro serif collage, 1940s jazz (agent-built)
05 Maná — yerba mate four flavor-colored rooms (agent-built)
07 JetPartners — private aviation navy old-money, night aerials
10 Corgi — startup insurance painted skies, official mascot family (agent-built)
11 Orchid — iMessage assistant paper-to-dusk, petals converge into the mark (agent-built)
12 TheraSun — spectral window film warm-black amber, the spectrum splits
13 Idensia — brand passport deep-forest engraving world
14 Laper — screenwriting software screenplay paper, FADE IN:, film-noir jazz
15 Dusty & Co — ceramic studio risograph print world; found the real brand behind a hosted passport page (agent-built)
20 Dior cream editorial, real Atacama/Hellix fonts embedded, faithful maison grammar

▶ Watch them all

Philosophy

AI-generated video all looks the same because everyone writes "make it premium" in a prompt.
Premium is not an adjective. Premium is a few hundred specific nevers — written down,
enforced by evidence, and versioned. Every time this codex grows, the floor rises for
everyone who runs it.

License

MIT

Reviews (0)

No results found