跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
视频与音频

launch-film

Make or change a product launch film in your own brand. A film is a hook (the story before the cut) on a shared body (the name punching in, hero words cut on the voice, your product at close range, the logo lockup in near silence), built in HyperFrames from brand/brand.json and film/film.json. Holds the core, the template workflow (build, render and mix 1 hook or all of them), the generation recipes, the checks and the lessons. Use for "launch video", "launch film", "product video", "promo", "reveal", "announcement video", "new hook", "change the body", "render all launch videos".

timscheuerai · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add timscheuerai/launch-video-kit --skill launch-film

技能说明


name: launch-film description: Make or change a product launch film in your own brand. A film is a hook (the story before the cut) on a shared body (the name punching in, hero words cut on the voice, your product at close range, the logo lockup in near silence), built in HyperFrames from brand/brand.json and film/film.json. Holds the core, the template workflow (build, render and mix 1 hook or all of them), the generation recipes, the checks and the lessons. Use for "launch video", "launch film", "product video", "promo", "reveal", "announcement video", "new hook", "change the body", "render all launch videos".

Launch film

Every launch film here is a hook plus the body. The hook is the story before the cut (4 to 18 s, any idea). On the music's drop it cuts to your brand, and the body names the product, shows it at close range beat by beat, closes the features into one idea and ends on your logo in near silence. The body is 1 template shared by every hook, so a new hook is 1 folder and 1 JSON file, and a body change reaches every film.

The working project is film/. film/film.json is the body (the script lines and the beats), film/hooks/<name>/hook.json is each hook, brand/brand.json is your brand.

Start here

  1. Brand. Read brand/brand.json. If it is missing, still the example brand, or "reviewed": false, run the launch-brand skill first (it turns the website into a brand kit and a preview sheet). Every colour, font and logo in the film comes from that file, never from memory.
  2. Brand core. Read references/brand-core.md: how each brand role maps onto the film, and the defaults when the brand is thin.
  3. Story. A new film starts in launch-story (brief, spine, script, beat sheet). A new hook on an existing body can start at "Make a new hook" below.
  4. Lessons. Skim references/lessons.md before the first frame. Each line cost a round once.

The core (keep it)

  • Format: 16:9, 1920x1080, 30 fps, about 30 to 60 s with the body. Picture rendered silent from HyperFrames, sound built by our pipeline and muxed on, master at -14 LUFS integrated and -1 dBTP.
  • Look: from brand/brand.json. Statement scenes on the brand ground, headlines in ink with 1 or 2 key words under the highlight, dark hero cards on dark, the accent on grounds and on the one element per shot that matters, never on long runs of type. The display font from fonts.display. Your logo as it appears on your site.
  • Grammar: Apple and Lovable launch grammar. Full-frame hero words cut in hard on the voice (the first word snaps from 2.4x with a blur, the rest step in from 1.3x), grounds change on hard cuts, one object becomes the next (dives, collapses, morphs), the real product UI at close range, fast with room to breathe, a quiet end hold.
  • Voice: 1 narrator for the whole film. Every new line is recorded in that voice. Type lands on the narrator's words, timed by forced alignment.
  • Music: 1 track, placed so its kick (the drop) lands on the cut whatever the hook's length, looped on its bar grid to the end, faded under the logo. Sound effects come from 1 palette: the synthesised forge or your generated kit.
  • Footage: generated or your own. Fictional people, no famous faces or lookalikes, no real films or actors, no third-party brands except to name an integration. Prompts kept beside the footage as provenance (docs/rights.md).
  • Words: the product's own. Message words come from your live website (brand.json copy); the owner's copy is used verbatim. Every number must be true and provable.

What is open

Everything else is a choice per film: the hook's story, look, length, its own sound, and whether it names the product at all. Pick what serves the idea and show the look before building it all. Change the body only when asked, and then change it in the template or film.json, never in 1 hook.

Make a new hook

  1. Idea and copy. Write the idea in the owner's words. Keep their lines verbatim.
  2. Show the look first. 3 to 5 keyframes or a storyboard page before any paid video job. Frames are judged fast; a full build on a wrong look wastes a day.
  3. Picture. Type-only (free: "video": null plus a ground), your own footage, or generated footage (references/generation.md). State the credit cost of every video job and wait for a yes. Check generated faces for likeness to anyone famous.
  4. The hook folder. film/hooks/<name>/ with the picture, the voice line and hook.json (schema in references/hooks.md). A hook longer than about 4 s brings its own music bed, or it plays in silence before the drop.
  5. Build, render, mix. From film/: ./render.sh <name> (build, HyperFrames render from index-<name>.html, audio). The master lands in film/renders/.
  6. Check (below), then show it with the gap list and record the verdict in your notes.

Change the body

The body's beats, their fields and the code that draws them are in references/body.md. For any change:

  1. Change film/film.json (lines, beats, words, focus rects) or the template film/templates/body.html, never a generated film/compositions/*.html.
  2. A new or changed line: record it in the narrator's voice, re-run the voice prep so every word is re-aligned (skills/launch-voice/scripts/prep_lines.py), and the cues follow the words.
  3. Snapshot the changed seconds and look at them before rendering (references/hyperframes-notes.md).
  4. ./render.sh rebuilds, renders and mixes every hook (a few minutes a film).

Checks before anyone sees it

  • Words: whisper reads every line exactly (whisper-cli -m models/ggml-small.en.bin or larger; the base model mishears product names).
  • Loudness: -14.0 LUFS integrated, true peak at or under -1.0 dBTP (ffmpeg -af ebur128=peak=true). skills/launch-review/scripts/review.py measures both.
  • Sync: every hero word and key moment within about 0.1 s of its spoken word.
  • Picture:
阅读完整来源 ↗