Skip to content
Back to search

motionmaxxing

SkillWorks with Claude CodeGitHub

Direct and build professional motion-graphics films (launch videos, product promos, feature reveals, brand stings, kinetic-type pieces, social ads, CTAs) that read as made by a senior motion designer, not AI slop. Finds an idea only this brand could own (from its logo, control, material, verb, voice), builds a world for it (ground, light, depth, texture, 3D or generated image surfaces for the hero moment), then builds with a seek-safe GSAP runtime driven by curves measured frame by frame from 53 professional recreations, and reviews its own render honestly with scripted gates (look.py, lint.mjs) instead of self-certifying. Given a website URL it captures the brand (colours, fonts, official logo, copy, screenshots); with an ElevenLabs key it writes and voices the script, scores it, and cuts picture and sound together. Use for any request to make, plan, storyboard, fix or critique a motion graphic or promo video, in HyperFrames or plain HTML.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/tejashmakwana--motionmaxxing/badge.svg)](/listings/tejashmakwana--motionmaxxing)

HTML

<a href="/listings/tejashmakwana--motionmaxxing"><img src="/api/listings/tejashmakwana--motionmaxxing/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

**Requirements:** Node 22+, Python 3, ffmpeg, Google Chrome. Optional: an ElevenLabs key for voice, sound effects and music (`ELEVENLABS_API_KEY`); the Codex CLI with image generation for surface plates.

**Claude Code**

git clone https://github.com/Tejashmakwana/motionmaxxing ~/.claude/skills/motionmaxxing
bash ~/.claude/skills/motionmaxxing/install.sh        # checks the requirements; installs nothing system-wide

Or, from a checkout somewhere else: `./install.sh` links it into `~/.claude/skills/motionmaxxing`. Flags: `--dir PATH` for another skills folder, `--copy` to copy instead of symlink, `--force` to replace an existing install.

**Other agents.** The skill is a folder with a `SKILL.md` at its root. Point any agent that reads `SKILL.md` at the folder (or copy it into that agent's skills directory) and give it shell access. The scripts only need `node`, `python3` and `ffmpeg` on the path.

**Check your machine**

bash scripts/providers.sh        # JSON: chrome, ffmpeg, key, codex, disk. Never installs anything.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 163 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. ---
  2. name: motionmaxxing
  3. description: Direct and build professional motion-graphics films (launch videos, product promos, feature reveals, brand stings, kinetic-type pieces, social ads, CTAs) that read as made by a senior motion designer, not AI slop. Finds an idea only this brand could own (from its logo, control, material, verb, voice), builds a world for it (ground, light, depth, texture, 3D or generated image surfaces for the hero moment), then builds with a seek-safe GSAP runtime driven by curves measured frame by frame from 53 professional recreations, and reviews its own render honestly with scripted gates (look.py, lint.mjs) instead of self-certifying. Given a website URL it captures the brand (colours, fonts, official logo, copy, screenshots); with an ElevenLabs key it writes and voices the script, scores it, and cuts picture and sound together. Use for any request to make, plan, storyboard, fix or critique a motion graphic or promo video, in HyperFrames or plain HTML.
  4. ---
  5. # motionmaxxing
  6. *Early release: this skill is still getting better, and more is on the way. If something feels off, say so in your note.*
  7. You are a motion director, designer and critic in one. Before you animate anything you decide **what idea only this brand could film**, **what world it happens in**, **what each beat makes the viewer believe**, and **how each shot becomes the next**. Then you build it with timing that comes from measurement, and you look at the render as a stranger would.
  8. Evidence: ~4,300 measured moves from 53 frame-accurate pro moments (cited as M913-M974), 52 ranked human films, ~30 AI films judged blind, and this user's own verdicts (`taste/verdicts.md`, `taste/edits.md`). `SKILL` = the folder containing this file. Work in your own project folder, e.g. `film/`.
  9. ## What good is (judgment decides WHAT; the craft laws decide HOW)
  10. 1. **The idea belongs to the brand.** Found in its logo shape, core control, material, verb, voice, success moment; cover the logo and the film must stop working for a competitor.
  11. 2. **The viewer watches something happen.** The claim is performed, not captioned: one specific case, real content, proof first, nothing invented and shown as real.
  12. 3. **Every beat causes the next.** Objects become the next scene; the interface's own gestures are the joints; a hard cut means something.
  13. 4. **Emphasis is spent, not spread.** Big needs small around it; one lit thing per frame; at most a quarter of the beats are "the point"; endings usually get smaller.
  14. 5. **Time breathes.** Hold before what matters, give reading exactly its time, stop before the idea is spent, end once and quieter than you began.
  15. 6. **It sounds like the brand.** Real phrasing, a writer's voice, phrase-level sound with silence before the hit.
  16. 7. **One world.** One logic of light, material, lens and drawing, in the brand's grammar; a ground is a surface, not a swatch.
  17. 8. **Made with ambition.** The hero moment gets the hardest work (3D, a shader, a dense comp, a generated surface) and is worth a screenshot. A small UI on a black void is the minimalist default, not taste.
  18. Read `references/judgment.md` before your first idea: each judgment with evidence, the default mind, and when "slop" things are right.
  19. ## Never ship (delete the element; do not justify it)
  20. A film containing any of these is not finished. These are what this user rejected hardest.
  21. 1. **Page chrome / HUD:** a brand or film name in a corner, tracked-caps labels, counters ("02 / SPEAK", "01 - THE WAIT"), timecodes, BPM/FPS/resolution text, progress bars, corner brackets, scene labels over a shot.
  22. 2. **Deck layouts:** a held left-aligned headline block, a kicker above a headline, a title then content, header/body/footer, "headline left, device right".
  23. 3. **Generic web-component UI:** a rounded card with a status dot and "label . $value", a "Good morning" card, "+ Aa :)" toolbars, zinc greys, skeleton bars; invented names or numbers where the capture has real ones.
  24. 4. **The name's first association** (stars for a star-like name, a glow for a fire-like name, a rocket for "launch").
  25. 5. **Invented stats and proof**, stat walls, a stat slide (centred big number over a small label), on-screen disclaimers.
  26. 6. **Slogan copy:** triplets of one-word sentences, "Not just X. It's Y.", "Unlock", "Meet X", "the future of".
  27. 7. **The house style as a default:** a giant cropped word, texture in the letters, a colour-flood or circle wipe between beats, a dot that becomes the logo, a transformation chain, a headline typed with a caret.
  28. 8. **Flat swatch grounds per scene, or a small card on a void** (a 0.3 W object for a 1 s read in empty ground).
  29. 9. **A pinned object** (a pill or dock at the same coordinates through three or more shots).
  30. 10. **A web-footer ending:** logo + tagline + button + URL, or any end card held past 1.4 s.
  31. 11. **Self-certification:** no `NOTE.md`, gates claimed without script numbers, "justified" fails.
  32. 12. **A film you did not look at.** A 15 s black render was once certified "VERIFIED / CLEAN".
  33. ## Craft laws (measured on the 53 moments; conditions in the references)
  34. 1. **Every beat buys one belief.** Write "Buys ___: the viewer ___." or cut the beat. Roles: hook, turn, proof, cta. `claim -> cut -> proof -> cta` is ONE film shape among several (`idea.md`); proof is the largest share of any film.
  35. 2. **Show it, then say it.** A claim in type is followed by a hard cut to the product performing it: real UI, real sentences, one specific case.
  36. 3. **Shots hand off.** The last frame of beat N is the first frame of beat N+1 (a carried object in the same pose, a transition that starts in A and finishes in B). A carrier is optional, travels at every seam, and the logo arrives by cause, never by default.
  37. 4. **Cut on motion, never on rest.** Cuts land mid-move; scenes start already moving; a film ends in motion or on a designed stillness, never a fade to black by default.
  38. 5. **Land decelerates, leave accelerates.** Entries put most travel in the first 4-8 frames and then a long tail, no bounce; exits ease in and leave or shrink. Median move about 10 f (0.33 s). One visible spring per film at most.
  39. 6. **Almost nothing fades** (a default, with conditions in `judgment.md` section 8: calm brands, privacy dissolves). Things arrive by moving, cutting, unmasking or focusing; fades are for single objects, not between scenes.
  40. 7. **Arrival decelerates from oversize.** Logos crash in from ~3.5x, cursors enter 2-5x oversize, a thesis word shrinks from 0.39 H; nothing grows from zero. Spend scale once: most type is small and the ending usually gets smaller. This is about arrival, not a giant cropped first word.
  41. 8. **No dead frame.** A held frame keeps one living thing (drift, caret, light, a breath of sound). A designed breath of 0.6-1.5 s before a payoff, with a stated reason, is good. look.py prints the human motion band (median 6.8, IQR 4.4-9.8) as a question: too busy to breathe, or too still to exist?
  42. 9. **One system per film.** One frame clock (smooth, or on-twos), one blur grammar (shutter OR smear OR crisp steps), the brand's own type (a serif if the brand uses one), one accent with one meaning ("AI acting", "on", "done"), an optional carrier that is never pinned.
  43. 10. **One focal point inside a built world.** 1 hero per frame, at most 3; every frame is ground + light + depth/texture; empty only when it does a job (under ~1.5 s). Statements are centred only when paced to speech, never as a page. UI is a magnified fragment with identity, readable by camera, not by inflated type.
  44. 11. **Words move at speech rhythm.** Whole words 2-12 f apart (2-4 fast hook, 4-6 headline, 8-12 voice-led). Letter-by-letter typing only in the product's own input, with an authored irregular rhythm.
  45. 12. **The demo is physical.** The cursor parks before it acts; the press squashes 2-3 f and the target answers a frame later; counters step per frame and land on a round hero value for a claim, an un-round value for data; cut before the release.
  46. ## Gates (computed on the render; a fail is fixed, never argued)
  47. Scripts print the numbers; you quote them in `NOTE.md`. Three gates are computed by a script (G0, G2, G3 by `look.py`; G5 by `lint.mjs`); G1 and G4 have no script: judge them on `look/sheet.jpg` and the stills and say so in `NOTE.md`; never estimate a number. Everything outside the gates is a "notice it" question. Same definitions in `references/slop.md` Part 2 and `references/tools.md`.
  48. | Gate | Pass | Computed by |
  49. |---|---|---|
  50. | G0 render exists | decodes, not blank, picture moves, length matches plan (`--expect S`), audio present and not silent if planned (`--expect-audio`) | `look.py` (GATES block, `look/gates.json`) |
  51. | G1 proof readable | UI text >= 0.04 H; fragment >= 0.55 W or cropped by the frame edge; lit, in a world; the camera magnified it | by eye on `look/sheet.jpg` and stills (no script number) |
  52. | G2 no empty frame | no run of more than 3 consecutive flat frames (a dip of <= 3 f inside a dive is ignored) | `look.py` |
  53. | G3 end card short | end hold (picture unchanged at the end) <= 1.4 s, so a resolved hold is 0.6-1.4 s (under 0.6 s is a note: it ends mid-motion); final shot (last cut to end) <= 25% of the film | `look.py` (`--hold-warn 1.4`) |
  54. | G4 one hero per frame | two panels only if one is >= 2x the other's area or the camera moves between them | by eye on the sheet (no script number) |
  55. | G5 no page chrome | no corner/edge labels (C1), tracked-caps labels (C2), counters (C3), timecodes/metadata (C4), header/footer bars (C6), kicker above headline (C7), left headline block at x < 0.2 with right-side media (D1), progress bars (P1), letter-typed headlines (T1), vibe-coded UI cards (V1-V5) | `lint.mjs` (`GATE G5: FAIL` = exit 1) |
  56. ## How you work
  57. ### 0. Toolbox and taste (a few seconds)
  58. ```bash
  59. bash "$SKILL/scripts/providers.sh" # Chrome, ffmpeg, HyperFrames, ElevenLabs, Codex image gen, disk
  60. ```
  61. Read `taste/verdicts.md` and `taste/edits.md` (what this user rejected and the before/after fixes) and skim `references/judgment.md`. `references/tools.md` lists every script and flag. No ElevenLabs means no voice or music: say so in a line and continue (`sound-sync.md` section 9).
  62. ### 1. Know the brand (`references/brand.md`)
  63. ```bash
  64. node "$SKILL/scripts/brand.mjs" <url> film/brand # board.png, screens/, media/, fonts/, BRAND.md
  65. node "$SKILL/scripts/fetch_logo.mjs" "<brand name>" <domain> --out film/assets/logo # official mark: SVGL, Simple Icons, then the site
  66. ```
  67. **Look** at `board.png`, `screens/` and `media/`; verify the ink and accent on the board (the headline ink is not the body grey; an accent is a colour from the imagery or primary button, not a link colour). Name only: find the official site, then capture it. Nothing: design an identity on purpose. Write the **inventory table** (`idea.md` section 2): logo shape, core control, material, key verb, success moment, voice, colour logic, type, what they would never do.
  68. ### 2. One true thing + PAGE (`references/idea.md` section 1)
  69. In four lines: Problem, Audience, Goal, Emotion; then the one real feature or moment the film is for (state your assumption; don't stop to ask routine questions).
  70. ### 3. Find the idea (`references/idea.md`)
  71. Write the first idea, then at least two more of different KINDS (a UI documentary of one task, a held shot, a joke with a character, a typographic essay, a real-output montage, stillness that breaks; a transformation chain is the most over-used). Cover-the-logo and competitor-swap tests; pick one and say why in one line. Optional rule-only precedent per beat: `python3 "$SKILL/scripts/precedent.py" query --beat proof --content ui --text "<what the beat shows>"` (also `show ID`, `rules`, `stats`).
  72. ### 4. Design the world and the time (`director.md`, `world.md`, `layout.md`)
  73. Write `film/STORYBOARD.md`: PAGE, inventory, the chosen idea and why, the film shape, **per act the ground, light, depth, texture and sound**, the picture source per act (which captured file or material is the ground), the **screenshot-worthy hero moment**, the importance of each beat (at most 25% are "the point"), script cues marked, then the beat table: role, **buys**, copy, what moves, entry and exit, the handoff, the carried object, frames, and **the picture with the text covered**. Check G1, G3 and G4 here, before building. Surfaces and 3D: `scripts/imagegen.py` for a missing surface plate, `runtime/hero3d` for a physical hero (`world.md`).
  74. ### 5. Script and voice (if VO fits; `references/sound-sync.md`)
  75. ```bash
  76. python3 "$SKILL/scripts/voice.py" vo --script film/vo.txt --out film/audio/vo.mp3
  77. node "$SKILL/scripts/sync.mjs" film/audio/vo.words.json --fps 30 --lead 2 --out film/beats.json
  78. ```
  79. Write VO in the brand's voice (real phrasing, then cut half) at ~2.5-3 words/s, one short line per beat; on-screen words land 0-2 f before they are heard. Without VO, the picture runs on a fixed cadence and the music grid.
  80. ### 6. Build the hardest beat first (`runtime/README.md`, `references/motion.md`)
  81. Copy `templates/film.html` to `film/index.html` and `runtime/` to `film/runtime/`. Build the hero moment and its hardest handoff, then `node "$SKILL/scripts/render.mjs" film/index.html --still 1.2,2.4,3.1 film/stills` and look. If it doesn't read at speed, fix the idea or the staging before building the rest. Every frame stays a pure function of time; run `Motion.selfTest(M)` for zero mismatches. Use `type.md`, `transitions.md`, `ui-demo.md`, `close.md` for mechanics.
  82. ### 7. Look honestly, then fix (`references/slop.md`)
  83. ```bash
  84. node "$SKILL/scripts/render.mjs" film/index.html film/draft.mp4 --scale 0.5
  85. python3 "$SKILL/scripts/look.py" film/draft.mp4 --out film/look --expect <planned s>
  86. node "$SKILL/scripts/lint.mjs" film/index.html
  87. ```
  88. `render.mjs` also writes `draft.mp4.events.json`, which `look.py` picks up (cuts are exact). Open `look/sheet.jpg`, `mid.jpg`, `strips.jpg`; read the report. **Describe every sheet frame in one sentence** (what is on screen, what is the hero, would it pass as a slide) before you judge. Run G0-G5 (`look/gates.json` and the printed GATES block, `GATE G5`; G1 and G4 by eye), then the thumbnail test, cover test, mid-change frames and real-speed reading. Fix in this order: broken or untrue; the idea; connections between beats; hierarchy and scale; timing; surface and finish; sound polish. Two or three honest passes.
  89. ### 8. Sound to picture (`references/sound-sync.md`)
  90. From `draft.mp4.events.json` pick the few events that deserve sound; write `film/sfx.json` (the `voice.py sfx --batch` list) and `film/events.json` (`mix.py --events` SFX placements `[{"t","sfx","gain"}]`; not render.mjs's `.events.json`).
  91. ```bash
  92. python3 "$SKILL/scripts/voice.py" split --vo film/audio/vo.mp3 --lines film/audio/vo.lines.json --out film/audio/vo
  93. python3 "$SKILL/scripts/voice.py" sfx --batch film/sfx.json --out film/audio/sfx
  94. python3 "$SKILL/scripts/voice.py" music --prompt "<style, bpm>" --ms <film ms> --out film/audio/music.mp3
  95. python3 "$SKILL/scripts/voice.py" music-fit --in film/audio/music.mp3 --out film/audio/music-fit.wav --duration <s> # if it reported a late start or short span
  96. python3 "$SKILL/scripts/mix.py" --duration <s> --vo film/audio/vo/line01.wav@<t - lead> --music film/audio/music-fit.wav --events film/events.json --out film/audio/mix.wav
  97. ```
  98. Silence before the hit; pull the music down while the viewer reads; 1-3 hand-placed hero hits. `look.py` audits the mix (LUFS, true peak, silent tail, hits against picture).
  99. ### 9. Final and deliver
  100. ```bash
  101. node "$SKILL/scripts/render.mjs" film/index.html film/final.mp4 --audio film/audio/mix.wav [--shutter 180 --subframes 8] [--grain 0.03]
  102. python3 "$SKILL/scripts/look.py" film/final.mp4 --out film/look-final --expect <planned s> --expect-audio
  103. ```
  104. `--shutter` and `--grain` are for smooth-clock films only (`world.md`). Deliver `final.mp4`, the editable `index.html` and **`NOTE.md` (mandatory)**: the one idea and why; each beat and what it buys; what is real (captured UI and copy) and what is illustrative or generated; the G0-G5 numbers quoted from `look.py` and `lint.mjs`; the one-sentence frame descriptions; what you would still improve. Describe what you checked; never certify the film as good. When there are versions to choose between, offer `python3 "$SKILL/scripts/blind_review.py" film/review --round "Versions" a.mp4 b.mp4` (opens `film/review/index.html`); append the user's reaction to `taste/verdicts.md`.
  105. ## Defaults when nothing is specified (starting values, not targets)
  106. - **Format:** 1920x1080, 30 fps, with sound. Length follows the idea: most launch films run 15-30 s; a single moment 3-7 s.
  107. - **Ground:** from the brand and built as a world (capture plate, material, light). Light grounds are as premium as dark; flat only if the brand is flat or for a flood under ~1 s.
  108. - **Type:** the brand's own face, else one neo-grotesk (Inter) at 400-550; sentence case. Headlines 7-11% H (9 typical), supporting 5-7%, hero numbers 22-29% H (spent once), wordmark ~0.22-0.38 W, statements centred at y ~0.47-0.53 only when paced to speech.
  109. - **Beats:** hooks 3-6 s, proofs 3-7 s of changing picture (no object unchanged beyond ~1 s unless a designed breath), CTA 2-3.5 s and <= 25% of the film, resolved end hold 0.6-1.4 s, cuts every 0.3-2.3 s. Know when to stop; if it already felt like the end, it is.
  110. ## Files
  111. | file | use it for |
  112. |---|---|
  113. | `references/judgment.md` | the 8 judgments with evidence, the default mind, the "slop is right when" table, importance budget, honest seeing |
  114. | `references/idea.md` | PAGE, brand inventory to idea, kinds of film, cover-the-logo, metaphor bank, script cues, copy voice, worked examples |
  115. | `references/world.md` | ground, light, depth, texture, real UI through a camera, hero3d, imagegen plates, shutter blur, grain, richness not clutter |
  116. | `references/director.md` | beat roles, "buys", film shapes and skeletons, film system, continuity, storyboard worksheet |
  117. | `references/brand.md` | URL to ground/ink/accent/type/carrier/copy decisions |
  118. | `references/layout.md` | coordinates, type size by role, emptiness, UI framing, lockups, layers |
  119. | `references/motion.md` | measured durations and eases, entry/exit grammar, smear, clocks, camera |
  120. | `references/transitions.md` | how beat A becomes beat B |
  121. | `references/type.md` | kinetic type, meaning on the word, counters, reading time |
  122. | `references/ui-demo.md` | cursor, click, typing, toggles, agent work, phones; recipes are mechanics, not a default film |
  123. | `references/close.md` | CTA and logo endings, picked by brand personality |
  124. | `references/sound-sync.md` | script, ElevenLabs VO/SFX/music, cutting to the voice, mixing, fallbacks |
  125. | `references/slop.md` | AI default vs pro decision table and the render review checklist |
  126. | `references/tools.md` | one page on every script and its flags |
  127. | `taste/verdicts.md`, `taste/edits.md` | this user's reactions; before/after lessons |
  128. | `studies/` | two calibration strips of invented frames (a default beside a decision, in different styles); look once, borrow reasoning, never the look; `PROVENANCE.md` says where they come from |
  129. | `runtime/` | seek-safe GSAP runtime: `motion.js` (`README.md` is the API), `hero3d/` (Three.js hero object, `M.hero3d`), `native-ui/` (iOS phone and notification builders, `NativeUI.*`), `fonts.css`, `vendor/` |
  130. | `templates/film.html` | the starting film (copy it to `film/index.html`) |
  131. | `examples/` | `demo/` (full beat set), `3d-hero/` (hero3d film), `phone/` (native-ui lock screen), `selftest/` (runtime regression page: must print PASS) |
  132. | `library/moments.jsonl` | 716 rule-only precedent rows read by `scripts/precedent.py` |
  133. | `scripts/` | `providers.sh`, `brand.mjs`, `fetch_logo.mjs`, `voice.py`, `sync.mjs`, `render.mjs`, `look.py`, `lint.mjs`, `mix.py`, `imagegen.py`, `blind_review.py`, `precedent.py` |
  134. ## Source and ethics
  135. The study corpus (other companies' published films, kept outside this skill) is private study material and is not part of it: learn mechanics and timing, never ship their brand assets, copy, layouts or frames as the user's. Principles here are in our own words; the strips are original images of invented brands. Generated images are for a missing surface only, never text, logos, UI or people shown as real, and are declared in `NOTE.md`. 3D is built in the film's own renderer (hero3d), not exported from other tools. When this skill and a HyperFrames house style disagree (decoratives, ghost text, hairlines, eyebrow or progress patterns, bold headings), this skill wins.
README.md · README · 307 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. <div align="center">
  2. <img src="docs/media/hero.gif" alt="motionmaxxing hero banner, a looksmaxxing face analysis applied to motion: on a dark glitchy ground the default grey serif word motion is measured with blue guide lines marked ?, dashed symmetry lines, a straight linear ease at 45.0 degrees and red, amber and green arrows reading default serif, tracking 0.0 and ease 0.0, with a tiny verdict: mid. A glitch cut, and the same overlays re-snap onto the brand type motionmaxxing, with maxxing in orange, and read measured values: weight 640, softLand 65.0 degrees, snapSettle, tracking minus 0.045 em, and the verdict maxxed, above the line looksmaxxing for motion graphics." width="100%">
  3. <sub>Your keyframes get a face rating: first the default (mid), then the measured version (maxxed). Typography and vector only, no photo of anyone, made with the skill's own runtime. <a href="docs/media/hero.mp4">Watch the MP4</a> &middot; <a href="docs/hero/NOTE.md">how it was made and checked</a></sub>
  4. <br>
  5. # motionmaxxing
  6. **Looksmaxxing for motion graphics.**
  7. An agent skill that turns AI slop videos into films that look made by a motion designer.
  8. <br>
  9. [![X](https://img.shields.io/badge/X-%40makwanatejas170-000000?style=flat-square&logo=x&logoColor=white)](https://x.com/makwanatejas170)
  10. [![Agent skill](https://img.shields.io/badge/agent%20skill-SKILL.md-111111?style=flat-square)](SKILL.md)
  11. [![Node 22+](https://img.shields.io/badge/node-%E2%89%A522-339933?style=flat-square&logo=nodedotjs&logoColor=white)](#install)
  12. [![Chrome + ffmpeg](https://img.shields.io/badge/renders%20with-Chrome%20%2B%20ffmpeg-4285F4?style=flat-square&logo=googlechrome&logoColor=white)](#install)
  13. [![Measured](https://img.shields.io/badge/eases-12%20measured-FF6A3D?style=flat-square)](#the-science-of-the-mog)
  14. [![Gates](https://img.shields.io/badge/mog%20check-G0%E2%80%93G5-2ea44f?style=flat-square)](#mog-check)
  15. [![Status](https://img.shields.io/badge/status-early%20release-orange?style=flat-square)](docs/KNOWN-LIMITS.md)
  16. [![License: Apache-2.0](https://img.shields.io/badge/license-Apache--2.0-blue?style=flat-square)](LICENSE)
  17. [![Lives inside FinalMotion](https://img.shields.io/badge/lives%20inside-FinalMotion-FF6A3D?style=flat-square)](https://finalmotion.video)
  18. **Follow [@makwanatejas170](https://x.com/makwanatejas170) on X** for the next glow-ups. Lives inside **[FinalMotion](https://finalmotion.video)**.
  19. [The glow-up](#the-glow-up) &nbsp;·&nbsp; [Why your AI video is mid](#why-your-ai-video-is-mid) &nbsp;·&nbsp; [The science of the mog](#the-science-of-the-mog) &nbsp;·&nbsp; [The protocol](#the-protocol) &nbsp;·&nbsp; [Mog check](#mog-check) &nbsp;·&nbsp; [Install](#install) &nbsp;·&nbsp; [Start maxxing](#start-maxxing) &nbsp;·&nbsp; [Toolbox](#toolbox)
  20. </div>
  21. <br>
  22. Your AI video is mid. Not broken, not ugly, just undecided: a label in the corner, a headline on the left, a card on a flat colour, a fade between every beat. **motionmaxxing** is an agent skill that makes launch films, product promos, brand stings and kinetic-type pieces, and makes them on purpose. Give it a website and it captures the brand, finds an idea only that brand could own, builds a world for it, and animates with timing measured frame by frame from 53 professional motion moments. Then it renders the film and looks at it with scripts, instead of telling you it came out well. It runs in Claude Code and in any agent that reads a `SKILL.md` folder. Everything renders locally from plain HTML.
  23. Motionmaxx it, then check the numbers.
  24. <sub>Early release. Still getting sharper, and more is coming soon.</sub>
  25. **The rating scale**
  26. | Rating | What it looks like on screen |
  27. |---|---|
  28. | **Slop** | A slide deck that moves: corner labels, a counter, a left headline beside a card, fades everywhere |
  29. | **Mid** | Clean and competent, and nobody decided anything: default eases, flat grounds, a held end card |
  30. | **Maxxed** | One idea, a world with light and texture, measured motion, cuts on motion, and gates that pass with the numbers shown |
  31. The skill's job is to move a film one notch right and to say honestly which notch it landed on.
  32. ## The glow-up
  33. Three before / after comparisons. In each, the same brand is filmed twice: once without the skill (mid) and once with it (maxxed). The brands belong to their owners and are used only to show the difference (see [Ethics and credits](#ethics-and-credits)).
  34. **Cluely, mid to maxxed. Same model, same brief: both films were generated with GPT-6.1.** Left: without the skill. Right: with it. You don't need the most expensive model to get motion that looks designed. The difference is the skill, not the price tag. [Watch the MP4](docs/media/cluely-comparison.mp4) (30 s with sound, 1920x600).
  35. [![Cluely, without the skill on the left and with it on the right, both generated with GPT-6.1](docs/media/cluely-comparison.gif)](docs/media/cluely-comparison.mp4)
  36. The left film is a slide deck: a brand label top-left, a "REAL-TIME ASSISTANCE" tag top-right, a held left headline in two colours, two cards, and a "02 / 05" counter. The right film is the product: the real assist overlay on a lit desktop, the ⌘ + ↵ shortcut as physical keys, then the overlay sitting on a live call.
  37. <br>
  38. **Spotify, mid to maxxed.** Left: without the skill. Right: with it. The without arm is a left headline beside a phone mock-up in every frame. The with arm carries one idea (the green full stop becomes a switch, floods the frame, and becomes the icon) through type, a wall of lines that collapses, and a full-bleed ending. [Watch the MP4](docs/media/spotify-comparison.mp4) (15 s, 2560x720).
  39. [![Spotify, without the skill on the left and with it on the right](docs/media/spotify-comparison.gif)](docs/media/spotify-comparison.mp4)
  40. <br>
  41. **Wispr Flow, mid to maxxed.** Top: without the skill. Bottom: with it. [Watch the MP4](docs/media/wispr-comparison.mp4) (24 s with sound, 1080x1440).
  42. <table>
  43. <tr>
  44. <td width="360" valign="top">
  45. [![Wispr Flow, without the skill above and with it below](docs/media/wispr-comparison.gif)](docs/media/wispr-comparison.mp4)
  46. </td>
  47. <td valign="top">
  48. **What to look for**
  49. - In the top film, small labels sit in the corners of most frames and a headline block is held on the left over a card. That is the pattern gate G5 exists to fail.
  50. - In the bottom film, the type is the picture, nothing sits in a corner, and the grounds change with the idea.
  51. - The comparison is 24 s with sound: the bottom film is cut to its voice and music.
  52. **What the skill is claiming.** Not that the bottom film is perfect. That it was decided (what to film, how it moves) and then checked, and the top one was not. Glow-up, not a miracle.
  53. </td>
  54. </tr>
  55. </table>
  56. ## Why your AI video is mid
  57. Most AI motion graphics fail in the same recognisable way: everything is a slide. A slop autopsy of rejected AI films ranked 25 default patterns. These are the loudest tells, and what catches each one.
  58. | The tell | What it looks like | Caught by |
  59. |---|---|---|
  60. | **Page chrome** | A brand name in a corner, a "02 / SPEAK" counter, a timecode, corner brackets, a progress bar | `lint.mjs` (G5), rules C1-C6, P1 |
  61. | **Deck layouts** | A held left-aligned headline, a kicker above it, "headline left, device right" | `lint.mjs` (G5), rules C7, D1 |
  62. | **Letter-typed headlines** | A headline typed out character by character with a caret | `lint.mjs` (G5), rule T1 |
  63. | **Vibe-coded UI cards** | A rounded card with a status dot and "label . value", a "Good morning" header, zinc greys | `lint.mjs` (G5), rules V1-V5 |
  64. | **Flat swatches** | A different flat colour per scene, or a small card floating on a void | Your eyes on the contact sheet ([`references/world.md`](references/world.md)) |
  65. | **Dead frames** | Runs of frames where nothing on screen moves | `look.py` (G2): more than 3 flat frames in a row fails |
  66. | **Fades everywhere** | Every element arrives by opacity and every beat ends in a dissolve | Measured instead: pros use 3,392 transform entries against 584 opacity fades, and cut on motion |
  67. Several of these were prescribed by earlier wording of the skill itself ("alternate flat grounds", "a recurring carrier"), so the fix was an edit to the skill, not a scolding for the agent. The lint is a lead, not a verdict: it cannot see text inside an image or a canvas.
  68. ## The science of the mog
  69. Most "motion design skills" are a list of opinions. This one started with a measurement.
  70. - **53 frame-accurate pro moments** were decoded by **6 parallel AI analyst agents** (reviewed by a human) into about **150 designer rules**, each as WHEN / DO / BECAUSE with evidence.
  71. - Every composition was **seeked frame by frame**. That gave about **4,300 distinct moves** (4,288) and **12 measured eases**.
  72. - A **slop autopsy** of rejected AI films ranked 25 default patterns, and traced which ones the skill's own wording had caused.
  73. - That was merged with **judgment from earlier generations**: the author's blind A/B rounds on an earlier generation of the skill (9 of 10 picks went to the with-skill film; one rater, so read it as a signal, not a benchmark), and a measured re-analysis of everything made so far.
  74. - The result is checked by **scripted gates** and a **page-chrome lint**.
  75. <br>
  76. <picture>
  77. <source media="(prefers-color-scheme: dark)" srcset="docs/media/eases-dark.svg">
  78. <source media="(prefers-color-scheme: light)" srcset="docs/media/eases-light.svg">
  79. <img src="docs/media/eases.svg" alt="The 12 measured eases of motionmaxxing plotted from 0 to 1, each with its typical length in frames, its share of measured moves and when professionals use it." width="100%">
  80. </picture>
  81. ### The numbers a designer can use
  82. | Finding | Measured |
  83. |---|---|
  84. | The median move | **10 frames** (0.33 s). Entries 11-12 f, opacity 6 f, blur 7-8 f |
  85. | Landing vs leaving | **Landing decelerates** (58% ease out); **leaving accelerates** (37% ease in, against 16% of landings) |
  86. | Overshoot | **Under 9%** of moves (8.8%), and late: a slow lean past the mark, not a fast spring |
  87. | Arrival | Decelerates **from oversize**. Logos crash in from about 3.5x. **Nothing grows from zero** |
  88. | Fades | Elements arrive by moving: **3,392 transform entries against 584 opacity fades** |
  89. | Cuts | Pros **cut on motion**, not on rest. 96% of moments use hard cuts, about 20 per moment |
  90. | Dead frames | In the 3-7 s pro moments, motion on **about 99% of frames** (median still run: 1 frame). Whole films breathe more; the skill treats that as a question, not a target |
  91. | The ending | A resolved end hold of 0.6-1.4 s; the ending usually gets smaller |
  92. ### The protocol is fundamentals
  93. Looksmaxxing is mostly boring fundamentals done consistently. So is motion design. The translation, with the measured finding behind each:
  94. - **Jawline = the hard cut.** Pros cut on motion. Scenes start already moving and clips end mid-move. A fade to black is a default, not a decision.
  95. - **Mewing for your frames = no dead frame.** In the 3-7 s pro moments, something moves on about 99% of frames. Gate G2 fails a run of more than 3 flat frames.
  96. - **Skincare = grain and light.** A ground is a surface, not a swatch: light, depth and texture on every frame. The hero moment gets the hardest work (real 3D, a generated plate, a dense composition).
  97. - **Posture = land decelerates, leave accelerates.** The same object on the wrong curve reads as floaty or as thrown. That single rule is most of the difference between "animated" and "directed".
  98. - **Nothing grows from zero.** Things arrive by decelerating from larger than life, or by moving in from the edge.
  99. ### Two more findings worth stealing
  100. - **Page chrome is the number one slop tell.** A brand name in a corner, a "02 / SPEAK" counter, a timecode, corner brackets: in the autopsy these were the strongest signal that no one decided anything. The lint catches them by geometry and typography.
  101. - **A fix is often an edit to the skill.** Several default tells had been prescribed by earlier wording ("alternate flat grounds", "a recurring carrier"). The skill was rewritten, not the agent scolded.
  102. The full account, with the method and the caveats, is in [docs/HOW-IT-WAS-BUILT.md](docs/HOW-IT-WAS-BUILT.md).
  103. ## The protocol
  104. Ten steps, the same order every time. The agent shows you the idea and the storyboard before it builds, then fixes in a fixed order when a gate fails.
  105. ```mermaid
  106. flowchart TD
  107. A["0 Toolbox and taste<br/>providers.sh, verdicts, edits"] --> B["1 Know the brand<br/>brand.mjs, fetch_logo.mjs"]
  108. B --> C["2 One true thing and PAGE<br/>problem, audience, goal, emotion"]
  109. C --> D["3 Find the idea<br/>3 kinds, cover the logo"]
  110. D --> E["4 World and time<br/>STORYBOARD.md, buys, handoffs"]
  111. E -.-> F["5 Script and voice<br/>voice.py, sync.mjs"]
  112. E --> G["6 Hardest beat first<br/>runtime, stills, selfTest"]
  113. F -.-> G
  114. G --> H["7 Look honestly<br/>render, look.py, lint.mjs"]
  115. H -->|a gate fails, fix it| G
  116. H --> I["8 Sound to picture<br/>sfx, music, mix.py"]
  117. I --> J["9 Deliver<br/>final.mp4, index.html, NOTE.md"]
  118. ```
  119. Dashed steps are optional: voice and sound need an ElevenLabs key, and some films are better silent.
  120. ## Mog check
  121. A gate that fails is fixed, not argued. Scripts print the numbers and the agent quotes them in `NOTE.md`. Two gates have no script, and the agent has to say so.
  122. | Gate | Pass condition | Computed by | The vibe |
  123. |---|---|---|---|
  124. | **G0** render exists | Decodes, is not blank, the picture moves, the length matches the plan, audio is present when planned | `look.py` | show up |
  125. | **G1** proof readable | UI text at least 0.04 H; the fragment at least 0.55 W or cropped by the frame; lit, in a world; magnified by the camera | by eye on the contact sheet | no squinting |
  126. | **G2** no empty frame | No run of more than 3 consecutive flat frames | `look.py` | mewing for your frames |
  127. | **G3** end card short | End hold at most 1.4 s; the final shot at most 25% of the film | `look.py` | do not overstay |
  128. | **G4** one hero per frame | Two panels only if one has twice the area of the other, or the camera moves between them | by eye on the contact sheet | one main character |
  129. | **G5** no page chrome | No corner labels, tracked-caps labels, counters, timecodes, header or footer bars, kicker over a headline, left headline block beside right-side media, progress bars, letter-typed headlines, or vibe-coded UI cards | `lint.mjs` | take the lanyard off |
  130. <p align="center">
  131. <img src="docs/media/gates.png" alt="The page-chrome lint catching a synthetic frame: a tracked-caps brand label, a counter, a timecode, a progress bar and a kicker over a left-aligned headline all fail gate G5." width="100%">
  132. </p>
  133. ## Install
  134. **Requirements:** Node 22+, Python 3, ffmpeg, Google Chrome. Optional: an ElevenLabs key for voice, sound effects and music (`ELEVENLABS_API_KEY`); the Codex CLI with image generation for surface plates.
  135. **Claude Code**
  136. ```bash
  137. git clone https://github.com/Tejashmakwana/motionmaxxing ~/.claude/skills/motionmaxxing
  138. bash ~/.claude/skills/motionmaxxing/install.sh # checks the requirements; installs nothing system-wide
  139. ```
  140. Or, from a checkout somewhere else: `./install.sh` links it into `~/.claude/skills/motionmaxxing`. Flags: `--dir PATH` for another skills folder, `--copy` to copy instead of symlink, `--force` to replace an existing install.
  141. **Other agents.** The skill is a folder with a `SKILL.md` at its root. Point any agent that reads `SKILL.md` at the folder (or copy it into that agent's skills directory) and give it shell access. The scripts only need `node`, `python3` and `ffmpeg` on the path.
  142. **Check your machine**
  143. ```bash
  144. bash scripts/providers.sh # JSON: chrome, ffmpeg, key, codex, disk. Never installs anything.
  145. ```
  146. ## Start maxxing
  147. In Claude Code, run `/motionmaxxing` or just ask in plain language. The skill triggers on any request to make, plan, storyboard, fix or critique a motion graphic.
  148. 1. **From a URL**
  149. > /motionmaxxing make a 20s launch film for https://example.com. I want to see the idea and the storyboard before you build.
  150. 2. **A single moment**
  151. > /motionmaxxing a 6 second kinetic-type sting for our brand: the one line "Ship it" with real 3D on the mark. Silent is fine.
  152. 3. **Rate and repair a mid video**
  153. > /motionmaxxing here is `draft.mp4` and its `index.html`. Rate it slop, mid or maxxed, run the gates, tell me what fails, and fix it.
  154. **What you get**
  155. | File | What it is |
  156. |---|---|
  157. | `film/final.mp4` | The film, with sound when a key is set |
  158. | `film/index.html` | The editable source: plain HTML plus the `Motion` runtime, reproducible frame by frame |
  159. | `film/STORYBOARD.md` | The plan: page, brand inventory, the chosen idea and why, per-act world, the beat table with what each beat buys |
  160. | `film/NOTE.md` | The honest account: the idea, what is real and what is illustrative or generated, the G0-G5 numbers from the scripts, what you would still improve. The agent never certifies the film as good |
  161. ## Toolbox
  162. Every script prints `--help`. Zero dependencies unless noted. Full flags in [`references/tools.md`](references/tools.md).
  163. | Script | What it does |
  164. |---|---|
  165. | `scripts/providers.sh` | Reports what the machine has: Chrome, ffmpeg, key, Codex image generation, free disk |
  166. | `scripts/brand.mjs` | Captures a site's brand: `brand.json`, `BRAND.md`, `board.png`, logo, fonts, screens, media |
  167. | `scripts/fetch_logo.mjs` | Fetches an official mark from SVGL, Simple Icons, or the site, and never redraws one |
  168. | `scripts/precedent.py` | Rule-only precedent per beat from `library/moments.jsonl` (716 rows) |
  169. | `scripts/imagegen.py` | Generated surface plates, with a guard that refuses text, logos, UI and people |
  170. | `scripts/voice.py` | ElevenLabs voice-over, sound effects, music, music fitting, and word alignment |
  171. | `scripts/sync.mjs` | Word timings to frames, for cutting picture to the voice |
  172. | `scripts/mix.py` | Voice, music and hits to one -14 LUFS / -1 dBTP mix, with predictable ducking |
  173. | `scripts/render.mjs` | Deterministic frame-by-frame render through Chrome and ffmpeg; stills, shutter blur, grain |
  174. | `scripts/look.py` | Contact sheet, cuts, end hold, audio audit and gates G0, G2, G3 |
  175. | `scripts/lint.mjs` | The page-chrome lint, gate G5 |
  176. | `scripts/blind_review.py` | A randomised, blind A/B page for choosing between versions |
  177. ### Runtime highlights
  178. `runtime/motion.js` is one plain `<script>` on top of GSAP, with no build step. The API is in [`runtime/README.md`](runtime/README.md).
  179. - **12 measured eases** by name (`M.ease.softLand`, `snapSettle`, `accelExit`, ...), also accepted anywhere an ease is.
  180. - **Seek-safe.** Every frame is a pure function of time; jumping to any frame equals playing to it. `Motion.selfTest` proves it, including canvas pixels.
  181. - **Words, typing, cursor, press, count, camera, snap, dive, whip, smear,** all authored in frames with measured defaults.
  182. - **`hero3d`.** A Three.js hero object (a logo extruded from SVG, a coin, a glass slab, a phone) with real light and depth, on the same clock. Three.js is vendored; no CDN, no build.
  183. - **`native-ui`.** True-proportion iOS furniture (phone, lock screen, notifications, tab bar) for when a product's real UI cannot be captured.
  184. ### Repo map
  185. ```text
  186. SKILL.md the skill: laws, "never ship" list, gates, the 10-step process
  187. references/ the craft, one topic per file (judgment, idea, world, motion, type, slop, tools, ...)
  188. runtime/ motion.js, hero3d/, native-ui/, fonts, vendored GSAP and Three.js
  189. scripts/ the toolbox above
  190. templates/film.html the starting film
  191. examples/ demo, 3d-hero, phone, selftest (must print PASS)
  192. library/ moments.jsonl: rule-only precedent rows
  193. taste/ verdicts and edits: what was rejected and the before/after fixes
  194. studies/ two calibration strips of invented brands, and where they come from
  195. docs/ hero source, showcase media, HOW-IT-WAS-BUILT, KNOWN-LIMITS
  196. install.sh requirement check, then link or copy into a skills folder
  197. LICENSE, NOTICE Apache-2.0, and the bundled third-party components with their own licenses
  198. ```
  199. ## Built by
  200. **Tejas Makwana**
  201. - X: [@makwanatejas170](https://x.com/makwanatejas170)
  202. - GitHub: [Tejashmakwana](https://github.com/Tejashmakwana)
  203. This skill lives inside **[FinalMotion](https://finalmotion.video)**.
  204. The full research and engineering story, with the method and the numbers, is in [**docs/HOW-IT-WAS-BUILT.md**](docs/HOW-IT-WAS-BUILT.md): the corpus, the six analysts, the frame-by-frame curves, the slop autopsy, the merge with earlier generations, and what each gate was born from.
  205. ## Ethics and credits
  206. - **Learned from private study of published films.** The films that were studied are not in this repo and are not redistributed. The skill ships principles in its own words and measured numbers. No third-party frame, clip or copy is included; the calibration strips in `studies/` are original images of invented brands.
  207. - **Showcase brands.** Cluely, Spotify and Wispr Flow belong to their owners. They appear only in the three before/after comparisons, to demonstrate the difference. These are unofficial demonstrations and are not affiliated with or endorsed by those companies.
  208. - **Generated images** are for a missing surface only, never text, logos, UI or people shown as real, and are declared in `NOTE.md`.
  209. - **Third-party code and fonts** are not covered by Apache-2.0 and keep their own licenses: [GSAP](https://gsap.com) 3.12.5 ([GreenSock Standard License](https://gsap.com/standard-license), which is not open source: you must comply with its terms), [Three.js](https://threejs.org) r186 (MIT), [opentype.js](https://opentype.js.org) (MIT), [Inter](https://rsms.me/inter/) (SIL OFL 1.1). See [NOTICE](NOTICE); license files sit beside the vendored code.
  210. - **License:** [Apache-2.0](LICENSE), the same license HyperFrames uses. Use it, fork it, ship it commercially. Keep the [NOTICE](NOTICE) with your copy.
  211. ## Status and known limits
  212. Early release, so a glow-up in progress. It works end to end on the author's machine and has been used on a handful of real briefs, but it has not had a wide test. Known limits, said plainly in [docs/KNOWN-LIMITS.md](docs/KNOWN-LIMITS.md): the lint cannot see text inside a canvas or an image; G1 and G4 are judged by eye; ElevenLabs music does not always hit tempo or length; WebGL renders run at about 7-11 fps; the sound-effect normaliser can corrupt very short clips; `look.py` treats show and hide events as cuts; and `fetch_logo.mjs` can return a header ribbon instead of the mark.
  213. ## License
  214. Open source under [Apache-2.0](LICENSE). Bundled third-party components (GSAP, three.js, opentype.js, Inter) keep their own licenses: see [NOTICE](NOTICE). Brand names in the showcase belong to their owners.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.