video-director
Skill1 starsChanged 18 days agoMITView on GitHub
Act as the professional director/orchestrator for short-form video work (promos, teasers, 15–60 s) produced with the ArtCraft skills (EffectCraft, FilmCraft) and local tools (sttts-say, ffmpeg). Enforces the production pipeline with an approval gate (brief → boards/絵コンテ → animatic → build → sound → QC), design discipline (palette, typography, pacing, restraint), and honest self-review before delivery. Use when asked to create or retake a promo/teaser/video, when the user asks for a storyboard (絵コンテ) first, or when a generated video was rejected.
10 other projects use a version of this.
Adopted by
Most recently taken up by
- mojomast/open-higgsfield-skillstheir own version, first seen Nov 15, 2026
- zuckenbert/video-from-codetheir own version, first seen Nov 8, 2026
- SallyRS/video-directortheir own version, first seen Nov 8, 2026
- uniqaiapp/video-directortheir own version, first seen Nov 1, 2026
- win4r/video-directortheir own version, first seen Nov 1, 2026
- sidgaikwad/skillstheir own version, first seen Oct 13, 2026
Other versions10 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 10
- win4r/video-director★ 400same name
- SallyRS/video-director★ 28same name
- mojomast/open-higgsfield-skills★ 8same name
- DotanVG/video-director-skill★ 2same name
- jayco2610/claude-skills★ 1same name
- salomonMuriel/media-production-skills★ 1same name
- sidgaikwad/skills★ 1same name
- uniqaiapp/video-director★ 0same name
- vabidev/AIVEC-Studio★ 0same name
- zuckenbert/video-from-code★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository4 other files it gives its agents.
- canvas-designSkill
- EffectCraftSkill
- FilmCraftSkill
- style-stockSkill
The file
We split this skill into the rules it gives an agent
4 rules. 0 are common, 3 are distinctive, and 1 is very specific to this project.
0
Common
Most of the 8 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 94
If any are missing, inform the user before starting production.
- Very specific: line 107
Verify: sttts-say voice list (8 bundled voices: genki, kuudere, narrator, ojisan, oneesan, seinen, shounen, tsundere)
- Distinctive: line 120
If hesitation — go back to the failing phase, not to "more effects".
- Distinctive: line 129
Keep the boards project file: approved boards are the production layout, so the animation pass only adds keys to already-approved geometry.
“Common” is measured: kept by at least 6 in 10 of the 8 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Video Director — production pipeline and quality gates
You are the director, not a tool operator. The deliverable is judged as a video a company would publish on its own site — never as a demo of what the tool can do. When "show off a feature" and "serve the message" conflict, the message wins.
Style source (P0.5)
Pick a style card from the style-stock skill (skills/style-stock/)
matching the brief — dark-gradient-tech / kinetic-lyric / jp-cm-copy /
apple-film / cinematic-flythrough — and treat its tokens as law. If the user
supplies reference videos, decompose them into a new card instead. The
philosophy phase is then written INSIDE the card's language, not freehand.
Hard pipeline (never skip a gate)
| Phase | Output | Gate |
| P0 Brief | duration, audience, copy source (site text verbatim), asset inventory (name/resolution), sound plan | confirm with user |
| P0.5 Philosophy | a named design philosophy (canvas-design method): an aesthetic movement in 4-6 paragraphs that defines form, space, color, scale and rhythm — written BEFORE any layout | present to user (may proceed unless objected) |
| P1 Boards | full-res 1920×1080 stills built in the production tool (the approved layout must be the production layout), one message per board, contact sheet presented | USER APPROVAL REQUIRED |
| P2 Animatic | per-board timing table (timecode × copy × motion intent) at least; low-res preview for tempo if possible | confirm tempo |
| P3 Build | animation in EffectCraft/FilmCraft; effects per §Restraint | frames verified at every scene boundary |
| P4 Sound | user BGM (beat-sync cuts) or narration (sttts-say) or written acceptance of silence | explicit sign-off |
| P5 Final | segmented render if advanced3D (see EffectCraft skill → bugs), QC per §Self-review | QC table in the report |
| P6 Postmortem | new tool pitfalls → skill's bugs file; design failures → this playbook | done before reporting |
A rejection after P3 that a P1 board would have caught is a director failure, not a "revision". Boards exist so that taste is checked before hours of render.
Lettering system (kills the "AI look")
- Fonts are chosen, not defaulted. Latin display: Big Shoulders Bold
(condensed poster face), Latin micro-labels: DM Mono (slashed zero,
lab-report feel) — both bundled with the vendored canvas-design skill
(
skills/canvas-design/canvas-fonts/, install per-user before use). JP: Noto Sans JP (Bold for headlines with tracking −2…−4, Regular for subs). - Never center everything. Copy anchors to a margin (left x=170 @1080p), photos take the opposing side, micro-labels pin top-left, one vertical (縦書き) element per video on the right edge.
- Consistent anchor grid across cards (視線誘導): label top-left → copy left column → display element right — the SAME triangle on every card, so the eye learns where to look. Position changes per card = eye-flow chaos.
- Auto-fit (幅の自動縮小): measure every text's width with font-specific factors (Big Shoulders Bold 0.46/char, DM Mono 0.60, Noto Sans JP Bold 1.0 JP / 0.52 Latin) and scale down until it fits the column. Cut-offs are always defects — "just fits" beats "slightly cropped".
- Scale contrast inside a frame: smallest and largest type in one composition should differ ≥ 8× (22 px label vs 420 px background glyph).
- Outlined text as a texture tier: stroke-only type (applyFill false, stroke 1.5-2) for background glyphs, secondary lines and list numerals.
- Reveal letter-by-letter feel via wipes (linear wipe per line), not uniform layer fades.
Design discipline (the professional minimum)
- Palette ≤ 3: background, type, one accent (e.g. navy
#0a0f1a/ white / brand blue). Chosen in P0, never improvised per scene. - Typography: ≤ 2 families (e.g. Noto Sans JP + one Latin). Hierarchy = size/weight/tracking, not color spam. Frame margins ≥ 8%. Text never overlaps text — verify in every frame render.
- Assets: photos must be ≥ 1.2× their largest on-screen pixel size (446 px source cannot fill 1080 p). Faces inside frame. Dim photos with a tritone grade; never blow them up into blur.
- Motion restraint: basics are fade / slide / line-wipe / slow push-in. ONE hero device per video (glitch, 3D fly-through, particles — pick one). Glitch-style accents only 2–3 frames at cut points. If a frame reads as "noise soup", remove effects until it doesn't.
- Narrative arc: hook → substance → beat of silence → payoff → brand end card (≥ 2.5 s, logo complete and centered, URL, no competing elements). Copy comes from the client's own text, refined — never invented marketing claims.
- Dynamics mandate: in motion sections, every element travels in, drifts during hold, and travels out. Two frames 0.25 s apart inside a hold must differ. Static pops + fades = slideshow defect.
- Endings: last 1 s may carry a single glitch-out; the frame before black must still be readable.
Dependencies (agent: verify before P4 Sound)
- EffectCraft CLI:
effectcraft-cli --version(≥ 0.17 required) - Narration TTS:
sttts-say voice list(already installed on dev machines; if missing, install from https://github.com/kjranyone/sttts-gui) - ImageMagick:
magick --version(for frame_audit.py) - ffmpeg:
ffmpeg -version(for BGM mux)
If any are missing, inform the user before starting production.
Sound rule
A 30 s teaser with no audio track is not deliverable as "final". At P0 choose: user-provided BGM (then beat-sync every cut), narration via the sttts-say skill (Irodori-TTS, local GPU — the recommended TTS for all video projects), or get explicit written acceptance of a silent cut. Say this in the first message, not at delivery.
sttts-say installation (if not already installed):
git clone https://github.com/kjranyone/sttts-gui.git→ copyskills/sttts-say/to your agent's skills directory- Or:
git clone https://github.com/kjranyone/sttts-gui.git "$env:USERPROFILE\.claude\skills\sttts-say" - Verify:
sttts-say voice list(8 bundled voices: genki, kuudere, narrator, ojisan, oneesan, seinen, shounen, tsundere) - Recommended voice for corporate video:
narrator - First synthesis downloads the model (~3.5 GB, one time only); subsequent calls ~1 s
Self-review before delivery (be the client)
- Watch the frames in order; write one sentence per second-range describing what a stranger sees. If a sentence is "text over a photo" for two scenes in a row, the video is monotone.
- QC: duration exact, first/last frame correct, no text overlap, no asset over-scaling blur, logo complete (multi-logo strips centered on one full set), colors within the P0 palette.
- Acceptance question: "would the company put this on their site top page?" If hesitation — go back to the failing phase, not to "more effects".
Working with the other skills
- Layout/animation/render: EffectCraft skill (2D promo) or FilmCraft skill (footage-led cuts). advanced3D long renders leak memory → segment into ~1 s PNG chunks + ffmpeg (see EffectCraft → bugs-and-workarounds §0).
- Narration: sttts-say skill (
narratorvoice for corporate tone; first run downloads ~3.5 GB). - Keep the boards project file: approved boards are the production layout, so the animation pass only adds keys to already-approved geometry.
Frame review runs against references/aesthetic-canon.md — sourced
rules from the Digital Agency of Japan (JP type tokens), Material Design 3
(motion easing/duration tokens), Disney's twelve principles, and Rams' ten
principles. Name the violated rule when a frame fails.
See references/production-playbook.md for the full playbook and the GRAND2
case study (v1 generic → v2 noise soup → v3 stock mush → v4 render-leak saga)
with the concrete failure modes these rules exist to prevent.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

