# Motion Video Kit: a Claude Code skill for AI-assisted business commercials > Motion Video Kit is a Claude Code skill that packages motion principles from 28 professional launch films, critic prompts and measurement scripts for building AI-generated business commercials. Rendering is the caller's responsibility, and the scripts require ffmpeg with the ebur128 filter. **echris6/motion-video-kit** — Claude Code skill kit for premium AI-assisted business videos: independent critic loop, motion principles from 28 launch films, quality bar, sound design, business offers, Three.js patterns, scripts - Repository: https://github.com/echris6/motion-video-kit - Stars: 1,091 · Forks: 100 - Language: Python - License: MIT - Published: 2026-10-11 · Updated: 2026-10-11 · Language: en - Canonical page: https://hysenlabs.com/projects/echris6-motion-video-kit ## Prompts and analysis files, not footage, are what this repository ships Business-motion-film/ holds SKILL.md, eleven reference Markdown files, a scripts/ folder and two template files. Nothing in the kit is rendered video or generative model weights. Work product from adopting it is whatever an LLM produces when given the skill as context. Source footage is not included; the repository holds only original written analysis and links to each of the 28 films. GitHub lists Python as the primary language for this repository. That designation comes entirely from the scripts/ subdirectory. Primary content of business-motion-film/ is Markdown, and a developer who has never installed Python can adopt the skill and use every reference file and prompt without touching a script. Knowledge inside the kit came from two activities the README names. Studying 28 professional SaaS launch films was the first. Building two complete sample commercials through dozens of rounds of independent critique was the second. One was a calm service film, the other a 40-second Three.js product specification ad for a foldable phone in which every frame is code. Both builds are preserved as case studies inside references/. A detail worth noting: GitHub labels this a Python project, but none of the four root-level entries are Python files. Every Python script is nested inside business-motion-film/scripts/, which means a team that wants only the reference material and prompts can clone the repository and ignore that subdirectory entirely. ## Gauntlet loop: builder and critic run as separate passes Builder and judge must not be the same call. That constraint is the core of what the kit calls the Gauntlet loop. Four steps define the process: a builder constructs the video, fresh critics evaluate the actual render, each critique item is verified against that render, and every finding goes into a ledger. Fresh critics receiving the rendered output, not a description of it, is what breaks a specific failure mode. When the person or agent that built something also checks it, the check tends to miss the same gaps the construction missed. Separating the roles does not guarantee quality. Breaking the feedback loop is what it achieves: a critic who did not build the thing has no stake in its flaws going unnoticed. Two files support the loop. gauntlet.md documents the process and includes real findings from the two sample commercial builds, so readers can see what the loop actually caught rather than what it theoretically catches. critic-prompts.md carries four separate prompt templates: component review, full-film review, verification and storyboard review. Each mode carries a different framing because questions appropriate for a single visual component differ from those a final-pass critic asks across an entire film. Ledger tracking is what keeps the loop from collapsing into a list of unresolved notes. Each finding is tracked against the render that produced it and verified in a subsequent pass. Managing that ledger is the team's responsibility; the kit supplies the prompts and the structure, and tracking findings across passes remains entirely outside its scope. ## Installing the skill in Claude Code takes two shell commands Installing for all projects requires cloning the repository and copying business-motion-film/ into Claude Code's global skills location. ```sh git clone https://github.com/echris6/motion-video-kit.git cp -r motion-video-kit/business-motion-film ~/.claude/skills/ ``` For teams that want the skill available within a single project rather than globally, a per-project alternative exists. ```sh cp -r motion-video-kit/business-motion-film /.claude/skills/ ``` After the copy, the skill loads automatically when a user asks Claude Code for a business commercial, a sample reel, an explainer, or a review of one. No API key, configuration file or environment variable is required for the integration itself. For LLMs other than Claude, pasting the contents of SKILL.md plus whichever reference files apply into the context window is the documented approach. Described as "also usable as plain context for any LLM," the Claude Code path is a convenience layer rather than a hard dependency. HyperFrames is the recommended renderer. Its README explicitly states that "the principles, prompts and checks are renderer-agnostic," which means the motion rules, quality checks and critic prompts apply regardless of which tool does the actual rendering. No rendering API is called; that work is entirely outside the kit's scope. ## Six motion rules and 16 reusable mechanisms extracted from the reference films motion-grammar.md is organized in three layers. Six rules form the base and function as non-negotiable constraints distilled from the 28 reference films. A catalog of 16 reusable mechanisms forms the middle layer: specific moves that appeared across multiple films and can be applied in new productions. Pacing numbers, timing figures extracted from the same films, form the third layer. A team working with this kit is working from analysis of 28 films, then 16 named mechanisms, then 6 rules. Those are three different levels of abstraction, and knowing which level a constraint comes from matters for deciding when to override it. A rule is more binding than a catalog entry, which is more binding than a pacing number. launch-film-notes.md covers all 28 reference films with links to the originals and no reproduced footage. A caveat in the README applies to any performance figures drawn from those notes: "Reach numbers quoted from studio portfolios are studio claims." Those figures originated with the studios that produced the films, not from independent measurement, and the README says so directly. Using those reach numbers as audience benchmarks means relying on unverified claims. Two case study files add a different kind of evidence. case-study-alder.md documents the calm service film built with the kit. case-study-duo.md documents the Three.js product film round by round, preserving the decisions made at each critique pass alongside the final result. That round-by-round format is more instructive than a finished-film summary because it shows what the Gauntlet loop actually changed between passes. ## Measurable quality checks and the audio rules that attach numbers to them quality-bar.md divides ship criteria into three categories: measured checks, visual checks and business checks. Four values occupy the measured category: frozen time, loudness, contrast and brand colour. Each of those is something a script or calibrated eye can verify against a specific number, which is the point of separating them from visual judgements. Audio has its own file. audio.md covers three topics: matching music to the buyer's customer, sparse clean sound effects, and mix targets. Mix targets are numbers and are what the ffmpeg loudness script measures. Music matching to a buyer's customer is a judgement the kit encodes as a rule rather than a formula, because selection depends on who the buyer is trying to reach. Business criteria carry their own practical concern. business-offers.md includes price anchors for production verticals, a pilot-offer template and what the README calls honesty rules. One of those honesty rules appears in the README itself: "Never present AI-generated imagery as a real customer, job or result." SKILL.md enforces this as a non-negotiable, meaning the skill carries the constraint into every LLM session that loads it. Price anchors in business-offers.md carry a re-verify advisory: the README labels them as a point-in-time survey of public listings and asks callers to confirm before quoting. Any team citing those numbers to clients needs to check against current listings first. That advisory is in the README, not inside business-offers.md, so a team that quotes prices without reading the README could pass stale figures to clients. ## Three.js patterns built to prevent non-deterministic renders three-js-patterns.md addresses a failure mode specific to code-generated 3D video: scenes that behave differently on each render, making it impossible to seek to a precise frame. Coverage includes a render contract that prevents this, specific patterns and a realism checklist. templates/ includes projected-overlays.js, which references describe as the projected-overlay module for the techniques documented in three-js-patterns.md. product-hero-realism.md covers one category of shot in detail: making a 3D device model move to match the real physical object frame by frame. Techniques named in the README are measured angle keys, a monotone cubic easing function, a locked camera, screen continuity management, frost handling, deterministic accumulation motion blur, and lighting that reveals the device angle without causing a flare. That last item deserves careful reading. Lights that reveal the angle without flashing is not a generic lighting principle. It refers to a specific constraint that appears when a product shot shows a device rotating: the specular reflection sweep produces a visible flash artifact when the camera position intersects the reflection angle at a particular frame. product-hero-realism.md documents the kit's solution to that constraint. A developer building a product shot without that file would rediscover the constraint through iteration, at a cost the file removes. ## Scripts need ffmpeg with the ebur128 filter and cover seven measurement tasks scripts/ contains 7 utilities. Named in the README: frozen-time, loudness, contact-sheet, soften-sfx, sfx-candidates, solve-sfx-gains and offline-mix. All seven depend on ffmpeg and ffprobe with the ebur128 filter. Node 22 or newer is listed as optional and is required only for the lab template in templates/component-lab.html. ebur128 is absent from some ffmpeg builds. Minimal distributions and older binaries sometimes ship without it. Verify your build includes the filter before running the loudness or offline-mix scripts; a build without ebur128 will fail those scripts at the filter step. templates/ holds two files: component-lab.html, described as an isolated component lab, and projected-overlays.js. Neither is a standalone application; both are designed to be integrated into a larger HTML or Three.js project. For versioning: the repository was last pushed on 2026-09-28 and is not archived. It carries no GitHub releases, so there is no tagged version to pin. A team adopting the kit works from the default branch, and any revision to the skill files or reference documents requires re-pulling the repository and re-copying business-motion-film/. ## Conclusion Motion Video Kit suits developers building AI video production pipelines who want a critique-driven workflow backed by motion analysis from 28 professional SaaS launch films and two documented sample builds. It is not the right tool when you need rendered output, prebuilt video templates, or a workflow that runs without an LLM at the centre. Skip it also if your ffmpeg build lacks the ebur128 filter, since the measurement scripts depend on it. Before taking it into a client project, verify your HyperFrames integration and re-verify the price anchors in business-offers.md against current listings before quoting them to clients. ## FAQ ### What is Motion Video Kit? Motion Video Kit is a Claude Code skill that packages motion principles, critic prompts and measurement scripts for building AI-assisted business commercials. It ships Markdown reference files and shell scripts; no footage or rendered video is included. Loading happens automatically when a user asks Claude Code for a business commercial, sample reel or explainer. ### Does Motion Video Kit work with LLMs other than Claude? The README says to paste SKILL.md plus the needed reference files into context for other LLMs. Described as also usable as plain context for any LLM, the Claude Code integration is a convenience layer rather than a hard dependency. ### What does the Gauntlet loop in Motion Video Kit do? It separates builder from critic: a fresh critic evaluates the actual render rather than a description of it, each finding is verified against the render, and findings go into a ledger. gauntlet.md ships with real findings from two sample builds, and critic-prompts.md carries prompts for four evaluation modes: component, full-film, verification and storyboard. ### What scripts are included in Motion Video Kit? scripts/ contains seven utilities named in the README: frozen-time, loudness, contact-sheet, soften-sfx, sfx-candidates, solve-sfx-gains and offline-mix. All require ffmpeg and ffprobe with the ebur128 filter. ### What renderer does Motion Video Kit require? HyperFrames is the recommended renderer but the README states that the principles, prompts and checks are renderer-agnostic. No rendering API is called; rendering is the caller's responsibility. ## Sources - [echris6/motion-video-kit on GitHub](https://github.com/echris6/motion-video-kit) - [Issues](https://github.com/echris6/motion-video-kit/issues) - [License: MIT](https://github.com/echris6/motion-video-kit/blob/main/LICENSE) - [README](https://github.com/echris6/motion-video-kit/blob/main/README.md) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/echris6-motion-video-kit