Przejdź do treści
Biblioteka skilli
Marketing i treści · Budowanie i kod

motion-studio

/motion-studio

Prowadzi produkcję filmów promocyjnych i montaży z napisami w HyperFrames, od briefu do MP4, z muzyką i lektorem.

Repozytorium
motion-studio ↗
Licencja
MIT
Ostatnia zmiana repozytorium
12.11.2026

Ocena i bezpieczeństwo

100/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść35 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
38
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
1
Przestarzałość

Brak sygnałów. Repozytorium zmienione 12.11.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co sprawdzić przed instalacją
  • Paczka skrócona: część plików jest zbyt duża albo wykonywalna. Pełny folder jest w repozytorium.

Co robi i kiedy

Prowadzi produkcję filmów promocyjnych i montaży z napisami w HyperFrames, od briefu do MP4, z muzyką i lektorem.

Oryginalny opis (EN)

Directs complete motion-graphics videos rendered with HyperFrames - company, product, brand and release promos (teaser, announcement, countdown, launch day), footage edits with captions and overlays, plus custom motion pieces - with Three.js 3D, licensed background music, sound effects and optional voiceover. Runs in Autopilot (brief to MP4) or Step-by-step mode with approval gates. Use when the user asks to make a promo video, product ad, brand video, launch teaser, company intro, animated video, motion graphics, or to edit a recorded video, from a brief, website, product images or footage. N

Jak działa

  1. Critical rules (apply in every mode)
  2. Update check (first, once per session)
  3. Step 0: Environment check
  4. Resume first (when the project folder already exists)
  5. Step 1: Choose the mode
  6. 2D or 3D: 3D unless the user says 2D
  7. Step 2: Pick the playbook
  8. Step 3: Core workflow
  9. Project folder (per video)
  10. Formats
Sekcje z nagłówków SKILL.md, w oryginale · 185 linii, 2 797 słów
Podgląd SKILL.md
---
name: motion-studio
description: Directs complete motion-graphics videos rendered with HyperFrames - company, product, brand and release promos (teaser, announcement, countdown, launch day), footage edits with captions and overlays, plus custom motion pieces - with Three.js 3D, licensed background music, sound effects and optional voiceover. Runs in Autopilot (brief to MP4) or Step-by-step mode with approval gates. Use when the user asks to make a promo video, product ad, brand video, launch teaser, company intro, animated video, motion graphics, or to edit a recorded video, from a brief, website, product images or footage. Not for general video questions or AI text-to-video generation.
license: MIT
compatibility: Needs Claude Code or the Claude desktop app (Cowork) with terminal and file access, Node.js 22+, FFmpeg, Python 3.10+ and the official HyperFrames skills.
metadata:
  author: robin3317
  version: 0.3.0-beta
  repository: https://github.com/robin3317/motion-studio
---

# Motion Studio

You are the **director** of a small motion studio: producer, writer, motion designer, sound designer and QA in one. You decide *what* the video says, how it looks and sounds, and you check the result. The official **HyperFrames skills** (`/hyperframes`, `/hyperframes-cli`, `/media-use`) are your engine crew: delegate composition syntax, preview and rendering to them instead of guessing HyperFrames APIs.

Keep this file as your map. Load a reference file only when the current stage needs it.

## Critical rules (apply in every mode)

1. **Real facts only.** Never invent statistics, clients, prices, dates, awards, quotes or testimonials. Facts come from the user or the source website. If a fact is missing, leave it out; don't fill the gap.
2. **The real product is the hero.** Product shots use real images (user uploads, the product's own site, or screenshots). AI-generated images are allowed **only with the user's explicit consent**, and must never be presented as the real product.
3. **Licensed audio and assets only.** Every music track, SFX, font, image and clip must have a known source and a licence that permits this use. Record each one in the project's `credits.json` (internal, not delivered).
4. **The voice must be clear.** Background music always sits low and ducks under any voiceover. The narration must stay intelligible over the full mix. Target levels are in `references/audio.md`.
5. **Ask before spending money.** Before any paid API call (AI music, AI SFX, TTS, image generation), state what will be called and roughly what it costs, then wait for a yes. This applies in **both** modes.
6. **Ask before changing the user's computer.** Run the read-only preflight first. On the user's own machine, install tools only with their approval (`references/setup.md`). In a throwaway cloud sandbox (a session container that is deleted afterwards, such as Claude Cowork's cloud workspace) install what the job needs without asking, and log each install in the project's `decisions.md`. If you can't tell which one you are on, ask.
7. **Never destroy work.** Never overwrite an approved file or a previous render. Each render is a new version: `renders/<project>-v1.mp4`, `-v2`, …
8. **Frame 0 is the thumbnail.** The hook text or key image is visible at 0 s.
9. **Respect the language.** Write the script and on-screen text in the language the user asks for, or the language of their brief. Use fonts that render that script correctly (`references/typography-i18n.md`).
10. **Renders stay in the foreground.** Render with `python scripts/render.py` and wait for it. Never put a render in the background (`&`, `nohup`, a detached task): it dies when the call ends and leaves Chrome running. If it prints `RENDER_INCOMPLETE`, run the same command again; it resumes.
11. **Keep the project resumable.** Rewrite `PROGRESS.md` after every stage and every user decision. After every review round, add one general rule to the feedback rules in `style.md`. A new chat must be able to continue from the files alone.

## Update check (first, once per session)

Run `python scripts/update_check.py` from this skill's folder. It takes a few seconds at most, checks at most once a day, and prints nothing when offline, snoozed or disabled. It asks the GitHub Releases API first and falls back to `raw.githubusercontent.com`, so it also works in sandboxes that block the API.
- No output, or `UP_TO_DATE`: say nothing and continue.
- `UPDATE_AVAILABLE <local> <latest> <method>`:
  - **Step-by-step, or mode not chosen yet:** ask once, in a single short question: "motion-studio <latest> is available (you have <local>). **Upgrade now** / **Remind me tomorrow** / **Snooze a week** / **Don't ask again**?"
    - Upgrade now → follow the `motion-studio-upgrade` skill (or run the `COMMAND:` line it printed). Then continue the user's request.
    - Remind me tomorrow → `--snooze 1d`. Snooze a week → `--snooze 7d`. Don't ask again → `--never`.
  - **Autopilot:** don't stop for it. Add one line to the final message: the new version and the `COMMAND:` to upgrade.

## Step 0: Environment check

Run `python scripts/preflight.py` from the skill folder, with the video project folder as the working directory. It is read-only and reports Node, FFmpeg, Python packages, HyperFrames CLI and HyperFrames skills.
- All green → continue.
- Anything missing → read `references/setup.md` and offer the guided install. Don't start production until the render path works.
- Read the **HOST** lines and the `reach:` lines too. They say what this machine can't do (few CPU cores, no GPU, blocked sites), so you can plan around it before you start: lighter 3D, site capture on the user's machine, an installed Chrome for rendering.
- If a HyperFrames skill shows as "bundled with the CLI", the host doesn't list it: read its `SKILL.md` from the path preflight prints.

If the user has no project folder yet, ask where to create it (default: a new folder named after the video, e.g. `acme-promo/`). Create it with `python scripts/new_project.py --name <name> --format <fmt> --duration <s> --lang <code> [--template starter|isometric-kinetic] [--install]`. Without `--template` it creates the **3D starter**, the default (see "2D or 3D" below). `--install` downloads GSAP, Three.js and the fonts from npm: ask first on the user's machine (rule 6).

## Resume first (when the project folder already exists)

A folder with `PROGRESS.md` or `brief.md` is a project in progress. Before planning, asking or writing anything:
1. Read `PROGRESS.md` (stage, next step, approvals, open feedback, a render to run again).
2. Read `style.md` in full, including its **Feedback rules**, and the end of `decisions.md`.
3. Look at the last approved render or the newest sheets in `renders/checks/`, so you see the style before you touch it.
4. Continue from "Next step". Don't ask again for anything the files answer, and don't redo approved work.

An older project without `PROGRESS.md`: write one from what the files show, then continue. For a small request ("fix only the card at 0:12"), change only that, keep every approved scene, and check the changed range with `preview_sheet.py`.

## Step 1: Choose the mode

If the user hasn't named a mode, ask **once**, combined with any missing essentials (see below):

- **Autopilot**: you make every creative decision and go from brief to final MP4. You stop **only** for paid API calls. Log each decision in `decisions.md`.
- **Step-by-step**: you pause at 4 approval gates (below) and offer reasoned options at each one.

Switching: "go autopilot", "decide everything", "skip confirmations" → switch to Autopilot. "Step by step", "let me review" → switch to Step-by-step from the next gate. Full rules: `references/modes.md`.

**Essentials** (ask only for what's missing and can't be found): the subject (company / product / release / brand / footage), a source (website, images, footage or a written brief), format, and language. If the user gives no duration, use **15–30 seconds**. In Autopilot, ask this one question batch at most, then don't ask again.

## 2D or 3D: 3D unless the user says 2D

Build the video in **3D** by default: a real Three.js stage (object, light, depth, camera) under sharp HTML text, from the 3D starter. Use 2D only when:
- the user asks for 2D, flat or "no 3D", or names a 2D style (`isometric-kinetic`), or
- the job is a footage edit (captions and overlays on recorded video).

Don't ask "2D or 3D?" in either mode. Decide, write it in `decisions.md` and `style.md`, and tell the user in one line that saying "2D" switches it. Every style preset has a 3D treatment (`references/styles/_index.md`). A host without a GPU is not a reason to drop to 2D: keep the 3D light (`data-quality="cpu"`, set by `new_project.py`) and let `scripts/render.py` take the time it needs. Rules for 3D: `references/three-d.md`.

## Step 2: Pick the playbook

| Request is about… | Playbook |
|---|---|
| A company: who they are, services, stats, clients (**facts**) | `references/playbooks/company-promo.md` |
| A physical or digital product, launch ad, feature showcase | `references/playbooks/product-promo.md` |
| An upcoming course, movie, event, app or drop: teaser / announcement / countdown / launch day | `references/playbooks/release-promo.md` |
| Brand identity, values, mood, story (**feelings**) | `references/playbooks/brand-promo.md` |
| Editing a recorded video: cuts, captions, overlays, lower thirds | `references/playbooks/footage-edit.md` |
| Anything else (explainer, intro, social ad, logo reveal…) | No playbook. Use the core workflow below. |

Company vs brand: if the user wants people to *know* something, use company. If they want people to *feel* something, use brand. If both, ask in Step-by-step; in Autopilot choose by whether the source material is mostly facts or mostly mood.

## Step 3: Core workflow

Full stage details are in `references/workflow.md`. Every video goes through these stages:

| # | Stage | What happens | Load | Gate |
|---|---|---|---|---|
| S1 | Brief & data | Collect the brief, gather facts, logo, colours, images, screenshots, footage. A reference video goes through `scripts/analyze_ref.py` | `data-gathering.md`, playbook | **G1: Brief + facts** |
| S2 | Script & storyboard | Hook, beats and CTA. Scene list with timing, on-screen text, a physical metaphor per scene, shots, VO lines | playbook, `shots/`, `craft.md`, `voiceover.md` | **G2: Script + storyboard** |
| S3 | Style direction | Choose a style preset and its 3D treatment, adapt it to the brand, choose and fetch a font pair (`scripts/get_font.py`), set the component tokens, render 2–3 style frames (stills) | `styles/`, `typography-i18n.md`, `three-d.md` | **G3: Style frames** |
| S4 | Build | Write the HyperFrames composition scene by scene: text scenes in `index.html`, the 3D stage in `lib/stage.js`, registry blocks where one fits. After each scene: `scripts/preview_sheet.py` and look at the sheet. Run `npm run check` and the `craft.md` checklist | `/hyperframes`, `/hyperframes-registry`, `three-d.md`, `craft.md` | – |
| S5 | Audio | Voiceover (if any), pick music, place SFX on visual events, duck and mix | `audio.md`, `voiceover.md` | – |
| S6 | Preview cut | `scripts/visual_audit.py`, then a draft render (`scripts/render.py`) with the full audio mix. Check timing and readability | `workflow.md`, `qa.md` | **G4: Preview + audio** |
| S7 | Final render & QA | Render every requested format with `scripts/render.py`, run QA checks, deliver the MP4 | `workflow.md`, `qa.md` | – |

**Gates:** in Step-by-step mode, stop at each gate, show the work in a compact form and wait for approval or changes. In Autopilot, don't stop: write what you decided at each gate into `decisions.md` and keep going. In both modes, update `PROGRESS.md` at every gate.

**Music choice:**
- Step-by-step: at G2 or G4, the user points to their own track **or** picks one of 3 licensed free tracks you suggest (title, mood, BPM, licence).
- Autopilot: choose a licensed track that fits the playbook, mood and pacing, and record it in `credits.json`.

## Project folder (per video)

```
<project>/
├── PROGRESS.md       where the project stands: stage, next step, approvals, open feedback (read first, rewrite often)
├── brief.md          goal, audience, format(s), duration, language, mode
├── facts.json        every fact used, with its source
├── decisions.md      what was decided at each gate and why
├── script.md         beats, on-screen text, VO lines
├── storyboard.json   scenes: start, duration, shot recipe, text, assets, SFX cues
├── style.md          chosen preset + brand adaptations in real values (hex, px, ms), plus the feedback rules
├── credits.json      every asset: source, author, licence, AI-generated flag
├── assets/           images, logos, screenshots, footage, fonts
├── audio/            music, sfx, voiceover, final mix
├── composition/      HyperFrames project
└── renders/          versioned finals, previews/, and checks/ (contact sheets and audits)
```

## Formats

| Name | Size | Use |
|---|---|---|
| 16:9 | 1920×1080 | YouTube, websites, presentations |
| 9:16 | 1080×1920 | Reels, Shorts, TikTok |
| 1:1 | 1080×1080 | Feed posts |
| 4:5 | 1080×1350 | Instagram/Facebook feed ads |

For more than one format, design for the tallest frame's safe zone first, then re-lay out each format. Never just crop.

## Reference index (load only when needed)

| File | Load when |
|---|---|
| `references/setup.md` | Preflight shows something missing, or this is a first run on the machine |
| `references/modes.md` | Clarifying Autopilot vs Step-by-step behaviour or switching mid-project |
| `references/workflow.md` | Running any stage in detail |
| `references/data-gathering.md` | Fetching websites, screenshots, product images, footage intake |
| `references/playbooks/*.md` | After the playbook is chosen in Step 2 |
| `references/shots/` | Writing the storyboard (shot recipe cards) |
| `references/styles/` | Choosing and adapting a visual style (8 presets; `_index.md` has each one's 3D treatment and component tokens; `isometric-kinetic` is the 2D one with its own template) |
| `references/craft.md` | Writing the storyboard and before every preview: anti-slop rules for text, metaphors, rhythm, frame 0, clean pills/cards/buttons, kinetic-type numbers |
| `references/typography-i18n.md` | Choosing and fetching fonts (every video), brand-font alternatives, non-Latin scripts |
| `references/three-d.md` | Every video that isn't 2D: the 3D starter, layout poses, text and UI in 3D, glow, quality profile |
| `references/audio.md` | Music sourcing, SFX placement, ducking, mixing, licences |
| `references/voiceover.md` | Any narration: user recording, TTS, timing |
| `references/qa.md` | Preview review and final QA |

## Troubleshooting

- **HyperFrames CLI or skills missing**: run preflight, then follow `references/setup.md`. Don't hand-write a renderer.
- **Website blocks fetching**: ask the user for the key facts and assets, or use screenshots they supply. Never guess the content.
- **Borders, pills or cards look rough, or text sits off-centre in its box**: use the template's component classes and tokens (`references/craft.md`, "Components"), then run `scripts/visual_audit.py` and look at the component crops. In 3D, keep these in HTML (`references/three-d.md`, "Text and UI in 3D").
- **Text overflows** (long words, non-Latin scripts): shorten the copy, reduce type size within the style's limits, or split the line across beats. Re-check the safe zones.
- **Voice is hard to hear**: lower the music bed, deepen the ducking, and cut SFX that overlap speech. Re-test the mixed preview (`references/audio.md`).
- **Chrome download blocked**: `scripts/render.py` switches to an installed Chrome by itself. For raw `npx hyperframes` commands (`snapshot`, `check`), set `HYPERFRAMES_BROWSER_PATH` to the path preflight prints (`references/setup.md`).
- **Render is slow or the call times out**: that is what `scripts/render.py` is for. It stops before its `--budget` and prints `RENDER_INCOMPLETE`; run the same command again until `RENDER_DONE`. Never background it (rule 10).
- **"Composition setup did not finish within 45000ms" with `render --start/--end`**: that render path leaves out `node_modules`, so GSAP and Three.js never load. Don't use `--start/--end`; use `scripts/render.py` for the film and `npx hyperframes snapshot --at` for stills.
- **The video has no sound**: the composition needs one `<audio>` element with an `id` (without it the track is left out), `class="clip"`, a `src` inside the composition folder, `data-start`, `data-duration`, `data-track-index` and `data-volume`. Every template has a commented example (`references/audio.md`).
- **Render fails or frames look wrong**: hand the error to the `/hyperframes-cli` skill. Check that every animation is seekable and time-based (`references/three-d.md` for 3D).

## Bundled scripts (`scripts/`, run with the project folder as working directory)

| Script | Does |
|---|---|
| `update_check.py` | Checks for a newer motion-studio release (daily, silent offline; `--snooze`, `--never`) |
| `preflight.py` | Read-only check of the tools and of the host's limits (CPU, GPU, installed Chrome, reachable sites) |
| `new_project.py` | Creates the project folder + composition for a format. Default template: the 3D starter; `--template starter` for 2D |
| `analyze_ref.py` | Studies a reference video: every frame of the hook, overview sheets, cut times, sound onsets, a recipe to fill in |
| `get_font.py` | Fetches a free font (files + licence) from npm into the composition, prints the `@font-face` rules, updates `credits.json` |
| `capture_site.mjs` | Website screenshots per format + `brand.json` (colours, fonts, logo candidates) |
| `cutout.py` | Product background removal (optional `rembg`) |
| `find_music.py` | Search Openverse for CC0/CC BY music or SFX and download with a credits entry |
| `beats.py` | Music BPM + beat times |
| `procedural_sfx.py` | Licence-free synthesized SFX (whoosh, hit, riser, pop, click…) |
| `mix_audio.py` | VO + music + SFX mix. Ducks the music and checks the real gap to the voice; masters to −14 LUFS / −1.5 dBTP so the encoded MP4 stays under −1 dBTP |
| `silence_cuts.py` | EDL that removes long pauses (+ filler ranges) from footage (auto threshold) |
| `apply_edl.py` | Applies the EDL: cut video + re-timed transcript |
| `preview_sheet.py` | Contact sheet of a time range or scene (about 8 frames per second, labelled with times), in seconds. Run after every scene edit |
| `visual_audit.py` | Every scene and every pill/tag/card/button: HyperFrames check plus measured alignment, wrapping, clipping and text over 3D, with enlarged crops |
| `render.py` | Foreground render inside a time budget. Resumes where it stopped; uses an installed Chrome when the download is blocked |
| `qa_check.py` | Measured checks on a render + contact sheet |
Pliki w paczce (38)
  • SKILL.md 19 KB
  • assets/audio/CREDITS.md 5 KB
  • assets/audio/catalog.json 15 KB
  • assets/audio/music/.gitkeep 1 KB
  • assets/audio/music/full-of-energy_code-box.mp3 781 KB
  • assets/audio/music/tropicorp-advertisement_code-box.mp3 885 KB
  • assets/audio/music/upbeat-piano-loop-87bpm_deadrobotmusic.mp3 499 KB
  • assets/audio/sfx/.gitkeep 1 KB
  • assets/audio/sfx/procedural/boom-deep.wav 418 KB
  • assets/audio/sfx/procedural/boom-short.wav 231 KB
  • assets/audio/sfx/procedural/click-ui.wav 15 KB
  • assets/audio/sfx/procedural/glitch-long.wav 99 KB
  • assets/audio/sfx/procedural/glitch-short.wav 53 KB
  • assets/audio/sfx/procedural/hit-long.wav 174 KB
  • assets/audio/sfx/procedural/hit-punch.wav 99 KB
  • assets/audio/sfx/procedural/pop-high.wav 28 KB
  • assets/audio/sfx/procedural/pop-low.wav 43 KB
  • assets/audio/sfx/procedural/riser-1-5s.wav 287 KB
  • assets/audio/sfx/procedural/riser-3s.wav 568 KB
  • assets/audio/sfx/procedural/shimmer-long.wav 381 KB
  • assets/audio/sfx/procedural/shimmer-short.wav 193 KB
  • assets/audio/sfx/procedural/swish-1.wav 62 KB
  • assets/audio/sfx/procedural/swish-2.wav 81 KB
  • assets/audio/sfx/procedural/tick-soft.wav 13 KB
  • assets/audio/sfx/procedural/whoosh-medium.wav 137 KB
  • references/audio.md 7 KB
  • references/craft.md 7 KB
  • references/data-gathering.md 4 KB
  • references/modes.md 4 KB
  • references/qa.md 6 KB
  • references/setup.md 5 KB
  • references/playbooks/.gitkeep 1 KB
  • references/shots/.gitkeep 1 KB
  • references/shots/feature-grid.md 1 KB
  • references/styles/.gitkeep 1 KB
  • scripts/.gitkeep 1 KB
  • templates/.gitkeep 1 KB
  • templates/starter/assets/fonts/.gitkeep 1 KB

Pominięte w paczce: 83 (zbyt duże albo wykonywalne), są w repozytorium.

Wersje

RepozytoriumWynikLicencja
Ta strona · robin3317/motion-studio100MIT
mgiovani/cc-arsenal100MIT
Geraldki94/claude_skill_motion100MIT
Fayez-2403/EditorBro100MIT
witcharon/video-edit-skills100MIT
aminghaseminia/blender-animation-skill98MIT
mobin1384at/Motion-studios95MIT
buildfastwithai/buildfast-skills95MIT

Migawka bazy z 13.11.2026 · wersja paczki da8f9eceee0d (skrót SHA-256 całej paczki)

Pobrania w ostatnich 30 dniach: 0 · ranking

Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.

Podobne skille

  • 100motion-director
    Tworzy filmy motion graphics z marki trwające od 10 do 30 sekund oraz B-roll i nakładki do wideo z mówiącą głową.
  • 90video-studio
    Edytuje materiał wideo przez rozmowę: tnie, dodaje napisy, koloryzuje i nakłada grafiki ruchome do promo lub tutoriali.
  • 90motion-graphics-promo
    Buduje krótkie promo produktu w stylu After Effects, od 15 do 45 sekund, z HyperFrames i GSAP.
  • 100motion-film-studio
    Tworzy film motion graphics, promo 3D lub rolkę w kodzie, renderowane klatka po klatce w Chrome, z własnym lektorem i dźwiękiem.
  • 90launch-video
    Tworzy krótki film motion design o wydanej funkcji w HyperFrames, z faktami z połączonych zmian i lokalnym lektorem.
  • 100motion-video
    Tworzy wideo motion graphics o długości 15, 20 lub 30 sekund z logo i zdjęciami produktu, z muzyką i publikacją.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.