motion-video

یک brief کوتاه را به ویدیوی موشن MP4 کارگردانی‌شده و آماده انتشار تبدیل می‌کند.

این skill از brief، اسکرین‌شات، لوگو، موسیقی، ویدیو یا voiceover شما یک مفهوم و روایت تصویری می‌سازد و آن را به صحنه‌های متحرک تبدیل می‌کند. صحنه‌ها با HTML/CSS/SVG ساخته می‌شوند و با Chromium و FFmpeg، فریم‌به‌فریم رندر می‌گیرند.

برای teaser محصول، معرفی قابلیت، آموزش، ویدیوی رویداد، data story، reel و kinetic typography کاربرد دارد. پیش از ساخت، منابع را بررسی می‌کند، نسبت تصویر و ریتم را انتخاب می‌کند و خروجی نهایی، نسخه فشرده و thumbnail را تحویل می‌دهد.

چه چیزی می‌گیرید

  • تبدیل brief به ایده، ساختار روایی و direction مشخص
  • صحنه‌های زنده و قابل‌ویرایش با HTML/CSS/SVG
  • رندر دقیق MP4 با Chromium و FFmpeg
  • هماهنگ‌سازی حرکت، موسیقی، voiceover و sound effects
  • بازبینی فریم‌به‌فریم، audit خروجی و thumbnail نهایی

برای چه کاری خوب است

  • تیم‌های محصول و استارتاپ‌هایی که teaser یا launch video می‌خواهند
  • ساخت explainer، app demo، tutorial و ویدیوی معرفی قابلیت
  • تولید social reel، تبلیغ، announcement یا ویدیوی رویداد
  • پروژه‌هایی که مواد اولیه مشخص دارند اما به کارگردانی و اجرای موشن نیازمندند

برای چه کاری نیست

  • ساخت پوستر، تصویر ثابت، پست گرافیکی یا slide deck
  • تولید footage لایو‌اکشن با هوش مصنوعی
  • پروژه‌های مستقل صوتی بدون خروجی ویدیویی

دربارهٔ این اسکیل

Motion Video

Direct first, then build. Never turn a plain brief straight into code.

Tool: node <skill-dir>/scripts/video.mjs <command>, run from the project directory. <skill-dir> is the absolute path of the directory this SKILL.md loaded from (for a user install, ~/.claude/skills/motion-video). Never the project directory. Setup and every flag: references/cli.md.

Workflow

  1. Sweep. Run doctor in the user's folder. Fix each MISSING line under READY, or tell the user. Plan around each LIMIT line. If it says an update check is due, tell the user once. Update only with their consent (README, Versions and updates). Plan with what exists: a software-only machine keeps 3D and blur light. An English-only voice never narrates another language.
  2. Gather facts from the user's material first. Research only the gaps. Write the facts with sources before anything else (references/direction.md, §0). Every claim on screen comes from them.
  3. Ask, when it matters. Ask once, in one message, per references/direction.md, Ask the user. Skip it for a one-shot or a run with no user.
  4. Direct. Run init <dir> in the user's working directory, never inside <skill-dir>. Replace its placeholder scene (references/building.md, Project layout). Fill <dir>/direction.md per references/direction.md and references/styles.md. Write three concepts, then pick the strongest (references/concepts.md). Pick ratio, size and fps for the platform now (references/delivery.md). Give the user a 5-line summary, then continue without waiting.
  5. Sound. Voice, music and effects combine as the video needs. Write the sonic concept and energy arc first (references/direction.md, §4). - Voice: use the user's narration as given. For text to speech, follow references/narration.md, Prepare the script for speech. - Captions: never add captions or a subtitle file without the user's yes (references/narration.md, Captions). - Music: only the user's own track. Run beats and size scenes in bars. - Motion effects: on by default. Cue them per references/sound.md, with its caps under a voice or music. - Silence: on request, "sfx": false and no voice or music.
  6. Build the scenes per references/building.md. Rebuild reference designs as live HTML. Never animate a flat screenshot of a design you can rebuild.
  7. Check each scene. After writing or editing a scene, run check --scene N. Open its still sheet with Read and fix what fails references/craft.md. Fix every finding of a pass in one edit before running again. A warning stays only for a deliberate choice.
  8. Render. Run render --draft once every scene passes, and again after each fix. Run the full render once the draft passes review. With a sound.bed, compare the no-bed copy, or tell the user to.
  9. Review. render already ran audit and wrote out/sheet.png and, with effects, out/audit.png. Read its output and open both. Never run audit alone after a render. A sound-only fix needs render again: audit reads the out/cues.json that render writes. Compare each cue's intent with its thumbnail. Fix each failing cue and false intent label, then render again. Listen to the mix when playback is available, and report when it was not.
  10. Deliver. Check out/thumbnail.png per references/craft.md, Final pass. Report the master, the -compressed.mp4 copy and the thumbnail. Make the upload copies and the listing text per references/delivery.md.

Non-negotiables

  • One visual language per video: shared type, icon family, light and camera language. Places can change by chapter inside one palette family.
  • A concept, not a template: story shape, compositions and copy system come from this product. Vary the composition from beat to beat.
  • The skill supplies tools, not the idea. Reference tables show range, never a menu. The kit is a floor. Build beyond it when the concept needs more (references/building.md, Beyond CSS).
  • No dead frame: the content carries the motion. The camera moves only with a reason, never a push-in on every scene.
  • Screen text holds still while the camera moves the world under it. World text moves with the world.
  • Every feature is shown by a verb moment, never a bullet list.
  • On-screen text passes check for size, contrast and hold time.

Routing

WhenRead
Expanding the brief into direction.mdreferences/direction.md
Choosing the concept: story shapes, beats, compositions, craft devices, design vocabularyreferences/concepts.md
Choosing or executing a visual stylereferences/styles.md
Writing scenes, the kit (icons, atmospheres, UI parts, leader lines), 3D, footage and libraries, the seek contract, the camerareferences/building.md
Sound effects, cues, intents, the mix under a voice or music, materials, key, user soundsreferences/sound.md
Voiceover, text to speech, word timing, captionsreferences/narration.md
Choosing a TTS engine for a language or machine, adding a voice or modelreferences/tts.md
Using one TTS engine: its voice, delivery and writing rulesreferences/tts/<engine>.md
Finding an icon namereferences/icons.txt (grep it)
Output for a platform (ratio, size, fps, codec, upload copies), listing text and covers, screen recordingreferences/delivery.md
Thresholds, and the visual review of sheetsreferences/craft.md
Commands, flags, video.json fields, setup errorsreferences/cli.md

Boundaries

  • Static designs (posters, social graphics, print) need a design tool. Slide decks need a slides tool.
  • Charts inside a video: one message per chart, the right chart form for the data, color that encodes meaning.
  • Live-action footage: composed from supplied footage (footage), never generated.

دانلود ZIP