motion-tools-and-skills

Tools, agent skills, prompts and starters for making motion graphics with AI agents (Claude Code). The focus is code-rendered motion (Three.js + GSAP, HyperFrames, Remotion, Canvas, WebGL/WebGPU, p5.js) and agent-driven DCC tools (Blender, Unreal Engine, ComfyUI).

The first collection was gathered from ten open-source Opus 5.5 video projects on 6 October 2026, a second pass on 21 October added six more and refreshed three, a third on 22 October added five more, a fourth the same day added material from fifteen of twenty motion-skill repos, and a fifth on 4 November added the Anatomy isometric-explainer skill and a guide to React Bits, and a sixth on 14 November added dream-loop, Hairline, three of Meng To's skill sets and Emil Kowalski's choreography guides (SOURCES.md). Everything copied here carries its upstream licence, and every folder has a NOTES.md that says what it does, when to use it, what it needs, and what was changed.

New here? Read docs/techniques.md first. It distils the shared method behind all of these projects into working rules for our stack.

What's inside

skills/ — Claude Code skills

prompts/ — prompts and briefs

• film-styles/: 51 film-style prompts (satellite briefing, archive light table, Swiss motion, data storytelling, isometric infographic, dark keynote, spy titles, glass product, ukiyo-e…) in one reusable 11-section schema.
• starters/mg-styles-15/prompts/: 15 motion-design style prompts with the source of every film they made, a template for new styles and an AI-jury review rubric.
• video-prompts/: 64 + 40 curated motion-design prompts and cases from two large collections of Opus 5.5 videos.

tools/ — render, check and finish

• headless-render/: review sheets, frame strips, tracked crops and parallel rendering for any time-pure page; GPU probe for headless Chrome.
• lemo-core/: parallel video render, reading-time check, loudness master with grain, procedural SFX, TTS with a Whisper check, three.js post and CRT passes.
• song-analysis/: song → beat grid, downbeats, sections, per-instrument onsets from separated stems, loudness envelopes, and word-level lyric timing.
• tidewater-asset-pipeline/: CC0 sound sourcing and −23 LUFS normalisation; Blender batch retargeting to GLB and decimation.
• sync-check/: sound against picture in a rendered MP4 (off-beat hits, startling jumps).
• mixamo-retarget/: human joints → animated Mixamo character GLB, in numpy, without Blender.
• skills/3d-product/references/dsh-blender-modeling/: Blender modelling pitfalls, numeric gates, Blender 5.2 API breaks, three-view calibration, verifier checklist.
• Plus beat detection, cut detection, green-screen keying, SFX mixing and a synth score inside the skills (listed in tools/README.md).

starters/ — templates to build on

• tidewater-webgpu-fx/: FFT ocean, physical sky, caustics, lens flare, motion blur and a full post chain in raw WebGPU/WGSL, with headless capture.
• pdoom-video-engine/: a three.js engine for audio-driven films with adaptive per-tile motion blur, true 4K from a 1080p layout and an HDR post chain.
• mg-styles-15/: 15 motion-design films with full source, a renderAt(t) render harness and the mgaudio soundtrack toolkit.
• papermotion/: a paper-cutout film engine with rigs, physics and simulation-driven sound, plus its skill.
• dioramas/: cinematic 3D pages with volumetric light, reflections and an AgX post stack; image → 3D assets through fal (Nano Banana 2 → Meshy 7.1); deterministic video capture of any live-clock page (dioramas, Anatomy figures, React Bits components), with --scale and --clip.
• Pointers to the kinetic-reel, painted-cartoon, web-intro and p5-overlay templates inside the skills.

docs/ — guides and research

• techniques.md: the shared method, applied to our tools.
• reference-only.md: useful material that could not be copied (no licence, or third-party content), with where to find it.
• guides/lemo-opuscar/: directing and pipeline guides for code-made films, and how to make a style your own.
• guides/: motion fundamentals (iart.ai), field rules for code films (Howseen), effect recipes, news-insert rules, motion tokens, Disney principles in frames.
• guides/emil-kowalski/: Emil Kowalski's choreography rules (what leads, what follows, what stays still), a build-in-order animation guide with exact curves, and a glossary that names any motion effect for a prompt.
• guides/react-bits/: using React Bits backgrounds and text effects in films (loop export, frame-exact capture, picks, licence limits); nothing copied.
• research/athemeroy/: a study of 1,000+ agent-made videos, with a production brief, production paths, workflow analysis and data.

Using a skill

Copy a skill folder into %USERPROFILE%\.claude\skills\ (Windows) or symlink it into ~/.claude/skills/ (macOS, Linux, WSL), then describe the video you want in Claude Code. Details in skills/README.md.

What stays out of git

Renders (*.mp4, *.mov, ProRes masters), Unreal caches (DerivedDataCache/, Intermediate/, Saved/, Binaries/), node_modules/, model weights, .env files and large media. See .gitignore.

Licences

• Our own writing (this README, the README.md and NOTES.md files, SOURCES.md, docs/techniques.md, docs/reference-only.md, prompts/video-prompts/yihui-prompt-index.md, and the scripts added to copied folders: starters/dioramas/scripts/render-video.mjs, starters/dioramas/scripts/capture/virtual-time.js, skills/video-shotcraft/assets/scripts/build-card-index.py): MIT © 2026 Ahmed Albadri. English translations of copied material stay under that material's upstream licence.
• Copied folders keep their upstream licence: each has its own LICENSE file, and SOURCES.md lists them all. MIT for the code from lemo-opuscar, ClaudeAnimationBase, opus-video-skills, motion-graphics-music-video-skill, tidewater, pdoom-video, mg-styles-15, papermotion, 3d-product, video-production-skills, dioramas and the dsh-blender-plugin modelling notes; Apache-2.0 for video-shotcraft, img2threejs and the Motion2MixamoRig retarget core; MIT for the fourth-survey skills, tools and guides (LottieFiles, iart.ai, Lob0Garou, kaventro, whaleyxbt, Sunwood AI Labs, Howseen, charlie947, Promptible, Dylan Tarre) for Anatomy (© Ryan); MIT for the sixth-survey skills and guides (dream-loop © Anshu Chimala, Hairline © Lucas Marques, Emil Kowalski, Meng To, and Leon Lin's sound effects); CC0 for the motion-token files; CC BY 4.0 for the athemeroy research and the adapted case list.
• Third-party assets fetched by these tools at install or render time (fonts, sample libraries, models) keep their own licences.