- Home
- Skills
- Image & Video
- Motion Director — Code-Built Promos, Explainers & Showreels with a Critic Loop
Motion Director — Code-Built Promos, Explainers & Showreels with a Critic Loop
Turns a brief into a finished motion film made entirely in code — fact-locked, beat-synced, brand-checked, measured, and reviewed by independent critics until it ships.
$25
Works with the AI Agents you already use
About this skill
Brief in. Finished film out.
Motion Director turns your agent into a motion director and creative technologist. Give it a brief, a product URL, an article or a set of references, and it designs, animates, scores and checks a short film entirely in code — Remotion, HTML + GSAP or HyperFrames, rendered frame by frame and encoded with ffmpeg.
Most "make me a video" skills stop when the render finishes. This one treats a bug-free render as the floor: a film is done when it holds up next to the best work in its genre — and it has the measurements and independent critics to prove it.
What makes it different
1. Facts are locked before anything moves.
Every claim that appears on screen or in voice lives in facts.json with the exact source sentence and the URL where that sentence actually appears. No invented prices, ratings, testimonials, results or "used by" logos. Vendor or estimated numbers carry their qualifier on the same frame — including frame one and the thumbnail.
2. Pictures can't imply what the sources don't say. A dedicated rule (and a dedicated critic) catches shots that are accurate word-by-word but misleading picture-by-picture — e.g. a product's visuals flowing into an unrelated warning headline.
3. The builder never grades its own work. After the storyboard, each scene and each full render, a fresh critic subagent that has seen none of the building pulls its own frames and measures. Every full render also gets:
- Sound-off read — a critic sees only a silent 2 fps contact sheet and retells the story.
- Sync check — every cut and hit within 1 frame of the beat.
- Fact check — every word traced to
facts.jsonand its URL. - Director's review — what feels assembled, slow or generic beside your references.
It fixes the biggest problem first, re-renders, and a new critic rechecks everything, with before/after numbers logged in a ledger. Critique has a round budget so it doesn't spin forever.
4. Measured, not eyeballed. Bundled scripts check:
- exact duration, size and constant frame rate
- frame one is a finished picture (no fade-from-black thumbnails)
- frozen frames and dead holds (slow drift counts as motion; true freezes don't)
- cuts within ±1 frame of the beat
- loudness (~−14 LUFS) and true peak (≤ −1 dBTP)
- WCAG text contrast, including text over moving picture
- brand colour in the encoded file (ΔE2000)
In testing, video_qa.py caught every planted fault in a test film — a 1.85 s freeze, a cut 2 frames off the beat, −16.8 LUFS, a +0.1 dBTP peak — and a colour shift of ΔE 3.47 from ffmpeg's default BT.601 conversion, which the skill's encode settings fix (ΔE 0.36).
5. Audio first; the picture is cut to it.
beats.py (numpy only) detects beats, downbeats and the drop and writes timing.json. Every animation reads its timing from that file, so changing the music re-times the film. On a test track it placed every beat within 1 frame at 60 fps and found the drop to the frame. It also writes a per-frame loudness envelope for lip-sync.
6. Brand guidelines outrank sampled colours. The skill looks for the brand's published guidelines before sampling a website, uses only approved colours, lock-ups and clear space, never redraws a logo, and records the approval route before anything is published.
7. Craft rules that separate authored films from templates. Eleven rules, each with a "why" and a measurable check: frame one is finished; one carrier object lands on the same pixels across every handoff; transitions are made of content; no linear easing; type is motion, not captions (with read-time math); holds breathe; never repeat a layout; every effect earns its place.
The workflow (gated or unattended)
- Lock facts →
facts.json - Study your references (cut maps, pacing, contact sheets) and how the audience talks
- Concept — 2–3 directions, three signature moments described without effect names
- Storyboard — table + rough frames rendered in code for every format ◆
- Style bible + three style frames rendered by the real pipeline ◆
- Audio and timing map — music, voice, SFX,
timing.json; audio-only mix first ◆ - Build + critic loop until the quality bar passes
- Deliver ◆
Choose gated to approve at each ◆, or unattended and every decision is logged for you to audit.
Four modes
- Product — real logo, real UI rebuilt as live components, every action has a visible result, ends on name + URL.
- Explainer — hook → familiar world → disruption → mechanism → discovery → consequence → recap; narration 125–145 wpm.
- Launch story — characters that act, from 2–3 written treatments.
- Showreel — one idea that keeps transforming; no fake clients, logos or metrics.
What's in the package
SKILL.md— the director: brief template, modes, claims rules, workflow, craft rules, critic loop, quality bar, deliverablesreferences/— modes, craft (with critic checks), audio (music prompts, ducking, two-pass loudness), build (determinism, toolchains, BT.709 encode, 9:16 safe zones, production traps), critic prompts, quality barscripts/—beats.py,video_qa.py,contact_sheet.py,boards.py,colour.py(Python 3.9+, numpy, Pillow, ffmpeg)assets/— templates for brief, facts, storyboard, style bible, ledger and timing map
Requirements
- An agent with terminal and file access (built and tested in Claude Code). Subagents are strongly recommended for independent critics; it falls back to headless
claude -psessions or clearly-labelled cold passes. - Python 3.9+ with
numpyandPillow;ffmpeg/ffprobe. - A renderer: Remotion (Node) or headless Chromium/Playwright for HTML + GSAP — or a HyperFrames kit.
- Optional: a music or voice generation tool, or your own licensed audio.
Honest limits
- It won't trim or edit existing footage, and it isn't a one-click AI video generator — it builds films in code.
- Quality depends on your inputs: real logos, brand files and facts make the difference.
- Measurements can't judge whether a joke lands or a voice sounds right; every delivery ends with a short list of what a human should still check by eye and ear.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Video Skills.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 28 days ago
- Software
- remotion
Node.js