motiongraphics
Edit a LONG-FORM 16:9 talking-head video in the MOTIONGRAPHICS style — a dark violet/magenta glow look built from lit glass tiles, self-building node diagrams, slam headlines on wiping violet bars, re
SKILL.md.claude/skills/motiongraphics/SKILL.md
GitHubVideo Studio (read first; overrides the rest of this file)
This skill is installed in Video Studio (Edits/.claude/skills/), from the NextPhase / creator packs
Bryan downloaded on 2026-10-31. Video Studio's CLAUDE.md wins wherever the two disagree.
- Start from the studio job, never an intake. Jobs come from the studio page (
jobs/<id>/job.json). This skill runs when the job'sstyleisglow-motion. Never ask "short or long?" or "which style?": the page already asked. (nextphase-studio's intake is not used here.) - One transcript:
projects/<id>/transcript/words.json(WhisperX). Don't run this skill's own transcriber. - Brand:
styles/brand.json(purple#8B5CF6/#C4B5FD/#4C1D95, Inter + Playfair Display) replaces this skill's default colours. Generate its theme/tokens from it. - Show the motion graphics before rendering (Bryan's rule): design first, show stills of each graphic over
the real frame in batches (the
design→graphicsstages,graphics.json), wait for approval, then render. - Sounds and music: the studio's library (
styles/sfx.json,node studio/sfx.mjs place) on the locked picture; music only from tracks Bryan owns or licensed. - Windows Git Bash, not macOS/WSL:
pythonnotpython3; a venv's Python isScripts/python.exe; long paths fail, so map a short drive (cmd //c "subst R: <winpath>"). Run this skill's setup/doctor script only the first time the style is used, and ask before any download over ~500 MB. - Where work goes: editing work in
projects/<id>/, deliverables injobs/<id>/.
MOTIONGRAPHICS — long-form editing style (16:9)
A talking-head + screen recording goes in; a chaptered edit comes out where the
motion graphics are the show. Where nextphase-long-editing dresses footage with
coral cards on a dotted grid, this style builds objects on a lit stage: everything
glows, everything is a surface with a rim, and the camera never fully stops moving.
⚠️ Before editing ANY video, ask the user which style:
motiongraphics (this skill) or nextphase (nextphase-long-editing). Never assume.
Where this look comes from
It was reverse-engineered frame-by-frame from two reference edits. Every colour, size and duration in
assets/tokens.cssis measured, not invented:
- REF-A —
youtube.com/watch?v=vwlK6MbdAto, 0:00–3:10 — violet + coral, diagram-led, no captions, long clean face beats.- REF-B —
youtube.com/watch?v=xkj74ykB9Fw, 0:00–0:50 — magenta, glow/3D-led, captions always on, faster cadence.They are the same language at two temperatures. This skill is that language.
reference/reference-log.mdhas the shot-by-shot breakdown with timecodes — read it when you need to know exactly how a beat was done.
⛔ Five hard rules
- Never hardcode a colour, font or duration. Import
assets/tokens.css; use its variables only. The measured cadence is what makes an edit read as this style. - Animate transforms only —
opacity, x, y, scale, scaleX/Y, rotation— plusfilter: blur(). Bars, rings and rails grow withscaleX+transform-origin, neverwidth. Animatingtop/left/width/heightstutters and fails lint. - Nothing has a hard edge without a halo. A shape with no glow is not in this
style. Every lit edge is
--mg-stroke(2px) and carries a--mg-glow-*. - Every word on screen is derived from THIS video. The copy in
components/is placeholder — it exists so the bricks render. Never let it survive into a segment. Runbash scripts/check-placeholders.sh videos/<name>before every render; it exits non-zero if kit copy is still there. - Nothing publishes without explicit approval. Deliver rendered + scored, open it for review, stop there.
Read reference/ before building — look.md (the visual language) ·
motion-grammar.md (measured motion, read before animating anything) ·
layouts.md (the ten scenes and the rotation) · motion-library.md (17 motion
devices that fill those scenes) · catalog-map.md (which HyperFrames catalog block to
reach for) · sound-design.md · workflow.md (the build pipeline) · reference-log.md
(the frame-by-frame source study).
0 · The frame
1920×1080, 30fps. Margin --mg-pad 96px × 54px title-safe. Same safe zones as
nextphase-long-editing, so a recording can be cut either way.
| Zone | Box | Rule |
|---|---|---|
| Title-safe | inset 96 × 54 | No text or mark outside this, ever. |
| Camera bubble | 1364,738 → 1864,1020 |
Reserved on screen-stage layouts. |
| Circle bubble | Ø420, free-placed | ⌁ REF-A 162s. The persona variant. |
| Point stack | 96,300 → 800,860 |
Left third. The speaker keeps the right. |
| Caption band | centred y 880–990 |
Nothing else lives here. |
1 · The look in one paragraph
Near-black tinted violet (--mg-base), never flat — every scene has a light
source: a centre bloom, a bottom-edge glow, a soft S-curve ribbon, or two edge rails.
Objects are lit surfaces: a violet glass tile with a white rim brightest along its
bottom edge; a near-black pill with a 2px lavender stroke glowing in and out; a panel
outlined in --mg-violet. Type is heavy italic caps when it slams and a clean
grotesque when it is read. One accent per frame — violet or magenta — with coral
reserved for brand marks. Full detail: reference/look.md.
2 · The ten scenes, and the rotation
Full spec in reference/layouts.md.
| # | Scene | Face? | For |
|---|---|---|---|
| 1 | Clean Face | full | the rest beat — and it is 30–40% of the runtime |
| 2 | Face + Floating Panels | full | labelled semi-transparent UI drifting over footage |
| 3 | Face + Point Stack | full | a violet headline over white bullets, left third |
| 4 | Face + Stat Card | full | a glass stat card with a count-up, near the hand |
| 5 | Face + Brand Lockup | full | co-brand lower third, held long |
| 6 | Diagram Takeover | none | the system builds itself on the gear field |
| 7 | Slam Card | none | one headline + one hero media on the bloom |
| 8 | Screen Stage | bubble | real recording as a rimmed card, camera follows the cursor |
| 9 | Object Stage | none | glass tiles / 3D objects with real camera moves |
| 10 | Persona Circle | circle | circular bubble + typed quote + typed name |
⭐ THE ROTATION — the retention rule
Never the same scene twice in a row. Never hold one past ~25s. Every scene change carries a transition AND a sound.
⌁ But note what the references actually do: Clean Face is used generously — REF-A
holds it for 8–12s at a stretch between graphic beats. Wall-to-wall graphics is the
amateur read. The graphics land, make their point, and leave fast (--mg-dur-exit
is 3 frames). The frame breathing between beats is what makes the next one hit.
⭐ REAL MEDIA, NOT TYPESET TEXT
Every beat that supports a point shows a real resource — a screenshot, a real
product UI, a repo card, real footage, a docs or pricing page — presented as a
.mg-media card on a stage, never full-bleed and never letterboxed. The Point Stack
(scene 3) is the one sanctioned text-only support beat.
- ≥1 real media per 25s of finished video, minimum 10.
- Check every screenshot actually rendered content before using it. Login-walled and JS-heavy pages come back blank and ship as empty panels.
- No random humans as b-roll. Only cut to a person when that person is the subject of the line.
- Never use emojis. Build designed marks — a coral-asterisk doc icon, a CSS glyph,
a small SVG pictogram in the
--mg-magentatile style.
⭐ COPY IS DERIVED, NEVER INHERITED
The bricks ship with realistic-looking text so they look like something in a browser. None of it describes the video being edited. A brick is a shape and a motion; the words in it are scaffolding to be thrown away.
Every string that reaches the screen comes from one of exactly four places:
| On screen | Comes from |
|---|---|
| A caption, headline or slam | The speaker's own words, at their onset in transcript-tight.json. Trim to fit; never paraphrase into something they didn't say. |
| A label on a diagram, pill, tile or list | The thing actually being described in that beat — the real tool, step, cost or objection named in the audio. |
| A stat, name, handle or quote | The real, verified figure or the real person. Sourced and fact-checked, never invented and never carried over. |
| Media inside a card | The real screenshot or clip from assets/broll/. A grey placeholder panel is an unfinished shot, not a design choice. |
Write the actual copy into PLAN.md, per beat, before authoring. A beat whose copy
is still "the brick's words" is not ready to build. This is also where a factual error
gets caught cheaply — on a plan line, not in a render.
⚠️ Some kit copy was lifted from the reference edits and belongs to other creators.
Those names have been neutralised in components/, and check-placeholders.sh flags
them as BORROWED if any copy resurfaces. Another creator's name in a monetized
client-facing video is not a cosmetic mistake.
CAPTIONS — a per-video switch, decided in PLAN.md
The references disagree, deliberately: REF-A has none, REF-B has them constantly. Both work. Pick one per video and hold it:
captions: off(REF-A) — the graphics carry the meaning. Denser diagrams, longer clean-face holds. The more premium read.captions: on(REF-B) — the accumulating style: the phrase builds word by word, already-said words bold white, the newest word lighter, and key words get a--mg-magenta-hihighlight box or violet colour on their own onset. Position per the safe-zone table; centred mid-frame on takeover scenes, bottom band over the face.
Either way, ship a separate .srt for the whole video. That's the accessibility
layer and it is not the same thing as burned captions.
SILENCE-TIGHTENING — do this FIRST
Reuse ../nextphase-long-editing/scripts/tighten-cuts.py. After it runs,
assets/transcript-tight.json is the only timing source. Timing against the
original transcript is the single most common way this edit goes wrong — it shows up
as "the graphics are all slightly early", never as an error.
3 · Motion — the measured cadence
Read reference/motion-grammar.md before animating anything. The five that define the
style:
| Move | Timing | What it is |
|---|---|---|
| Word build | ⌁ word fades 0→1 in --mg-dur-word (8f), onsets --mg-dur-cadence (7f) apart |
A title assembles one word at a time while the whole line continuously re-centres. The re-centring is the tell. |
| Bar wipe | ⌁ --mg-dur-wipe (12f), scaleX 0→1 from left |
The violet highlight bar opens behind a slam word, leading edge brightest. |
| Light leak | ⌁ --mg-dur-leak (12f): bloom 5f → white-out 3f → reveal 4f |
The signature transition. A violet-white streak washes the frame and the next scene is already in place behind it. |
| Ring draw | ⌁ --mg-dur-ring (18f), starting ~0.5s after the lockup settles |
The conic progress ring sweeps clockwise. Late arrival is deliberate. |
| Camera pull-back | ⌁ continuous, ease:"none", across the whole scene |
The title lands large and everything shrinks as content arrives beneath it. One unbroken zoom-out — not discrete re-fit steps. |
Three standing rules:
- Nothing on screen is ever perfectly still. A held scene carries a continuous slow
drift — scale
1.00→1.04over its whole life,--mg-ease-drift(linear). This is the difference between "a motion graphic" and "a slide". - Enter slow, leave fast. Entrances are
--mg-ease-outover ~0.3–0.6s. Exits are--mg-dur-exit— 3 frames. Elements are gone, not dissolved. - Overlap the handoff by
--mg-overlap(2f). ⌁ In REF-B the incoming headline is already on screen while the outgoing tiles are still sliding off. Never leave a gap.
Timing anchors: every graphic enters on, or a hair after, the tight-transcript word onset it illustrates — never before. Round-number timings are banned. Exit at the OUT word + 0.3s; cap a support graphic at 10s, then evolve it rather than let it sit.
4 · Building — HyperFrames + the catalog
All the HyperFrames skills are installed, and the 372-item catalog is available.
reference/catalog-map.md maps every pattern in this style to the catalog blocks that
already implement it — reach there first, restyle with tokens.css, and only hand-build
what the catalog misses.
export PATH="$HOME/.nvm/versions/node/v24.18.0/bin:$HOME/.local/bin:$PATH" # every step
npx --yes hyperframes catalog --search "<pattern>" # find it
npx --yes hyperframes add <name> # install into the segment
components/mg-system.html is the whole style on one page — all ten scenes, the light
leak, all seventeen devices, the palette, the materials and the cadence table, every one of
them live. Open that first; it is the fastest way to see what the style is. The two kits
below are the same material split by purpose:
components/mg-blocks.html— the ten scene layouts (layouts.md).components/mg-motion-library.html— 17 motion devices (motion-library.md): cursor cards, tag clouds, blur swaps, wave charts, ladder climbs, alert flashes, orb count-ups, skeleton fills, chip marquees, orbit hubs, notification pile-ups, 3D numbered grids and more. Scenes are where a beat is staged; devices are what moves in it. One device per beat — two is a collision, not a composition.
Bricks are starting points, not a cage.
5 · Sound
reference/sound-design.md. Reuse the palette and mixer from the sibling skill —
call the script by its path inside that skill folder, never copy it:
bash ../nextphase-long-editing/scripts/mix-sfx.sh out/visual.mp4 out/sfx.mp4 sfx-cues.txt
bash ../nextphase-long-editing/scripts/mix-bgm.sh out/sfx.mp4 out/final.mp4 assets/music/bed.mp3
- Density ~12–20 cues per minute. Every layout change carries a sound. No silent cuts.
- The light-leak always gets a sound —
swooshon the bloom with arisertail cresting on the white-out. It is the loudest transition in the style; treat it as one. - SFX sit ~13–15 dB under voice peaks; the music bed ~22–26 dB under. Calibrate with
ffmpeg -af volumedetect, don't guess.
6 · The workflow
Full detail in reference/workflow.md. The skeleton:
- Ask which style. Then locate
videos/<name>/,VIDEO.md,assets/recording.mp4. - Transcribe → gap-check minute by minute (Parakeet silently drops audio).
- Tighten →
transcript-tight.jsonis now the only timing source. PLAN.md— chapters with word onsets, the scene assigned to each beat, the caption switch, the re-hook marks, the media shopping list. Get a "go" before authoring.- Source real media (≥1/25s, min 10). Verify facts. Normalise every sourced clip
to 30fps CFR and check with
ffprobe— sourced clips come back at 24/25/50/60 or VFR. - Author per-chapter compositions —
seg-01-…/, one per chapter. Symlink assets into each segment folder; HyperFrames rejects../assets/. - Per segment:
lint(0 errors) →check-placeholders.sh(must exit 0) →snapshot --at <times>→ read the contact sheet and fix →render -q high -f 30. - Concat + chapters via
../nextphase-long-editing/scripts/concat-segments.sh— chapter timestamps come from real rendered durations, never fromPLAN.md. - Sound: SFX pass, then the ducked bed.
- Deliver:
out/final.mp4·out/chapters.txt·out/<name>.srt. Report rendered + scored, open it for review. Publish nothing without approval.
Files in this skill
assets/tokens.css— the measured design tokens. Import into every composition. (CSS comments do not nest — never write a*/inside one of its header blocks, or the whole sheet silently stops applying and every scene renders unstyled.)components/mg-system.html— everything, one page. Start here.components/mg-blocks.html— the brick kit: all ten scenes on a real 1920×1080 stage, animated at the measured durations. Open it in a browser. Append?still(or hit freeze) to hold every scene at its settled state for comparison.reference/look.md·motion-grammar.md·layouts.md·motion-library.md·catalog-map.md·sound-design.md·workflow.md·reference-log.md.reference/stills/— the source evidence: four frame-by-frame transition strips at 30fps plus nine key stills, including 3× zooms on a node pill and a glass tile. Open the matching still besidemg-blocks.html?stillwhen tuning a scene.scripts/check-placeholders.sh— the render gate. Fails if kit copy or a borrowed name survives into a segment. Run it before every render.- Other scripts are shared with
nextphase-long-editing— tighten, mix-sfx, mix-bgm, concat-segments, the RVM background remover. Call them by their path in that folder.