motion-anything

agent
Security Audit
Pass
Health Pass
  • License — License: Apache-2.0
  • Description — Repository has a description
  • Active repo — Last push 15 days ago
  • Community trust — 516 GitHub stars
Code Pass
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

✨ The agentic motion layer — an open-source, chat-native motion engine. Describe the feeling; your AI ships the animation.

README.md

motion-anything

A project in the nexu.io · Open Design family — the same team's take on motion. If motion-anything clicks for you, Open Design is where the full agent-era design studio lives.

Describe the feeling — your AI ships the animation. The agentic motion layer: a local-first, chat-native motion engine. Generate animated pages and launch videos from one sentence, then edit motion on the running page, component by component — 4 triggers, 13 motion verbs, spring easing, a full keyframe editor. Driven by 8 coding-agent engines + BYOK (Claude Code · Codex · Cursor · OpenCode · Grok Build · Hermes · Gemini · Open Design Cloud), armed with 403 curated motion recipes, wired into the Open Design ecosystem (59 DESIGN.md brand packs · 58 HyperFrames video templates · 112 HTML prototype templates · 2,680 icons), and exported to anything: JSON · CSS · React · Lottie · MP4 · GIF · portable skills. Zero npm dependencies. No watermark. No per-render fees.

motion-anything — Anything becomes motion, on your laptop. Editorial dark banner with light-mode app windows: the component-motion workbench and the launch-video compositor, plus a stats row — 403 motion recipes, 230 portable skills, 2,680 icons, 59 design systems.

License Engines Recipes Export Quickstart Zero deps

Design systems Video templates HTML templates Icons Portable skills

Family Follow on X

English · 简体中文


Showcase

Every tile below is a live, dependency-free recipe from the library — real GPU shaders, canvas engines and kinetic text, faithfully ported so they run with two <script> tags on any page (no React, no three.js, no build step). All tiles are animated GIFs recorded from the actual previews.

Typewriter multi — terminal-style lines typing in sequence Shiny text — a highlight band sweeps across the label
typewriter-multi · kinetic text
Terminal-style multi-line typing with live cursors — agent-boot energy.
shiny-text · kinetic text
A sheen sweeps across text-clipped gradient type. One class to use.
Star border — a slow light circles the button border Rotating text — words cycle in place
star-border · ambient CTA
A slow light orbits the border of your call-to-action. Pure CSS.
rotating-text · kinetic text
One line, many claims — words cycle in place with a soft travel.
Magnet lines — a grid of lines tracks the pointer Elastic slider — a springy gradient fill follows the drag
magnet-lines · interactive
A field of compass needles that swivel toward your pointer.
elastic-slider · feedback
A slider whose fill overshoots with a spring — tactile, GPU-safe.
Bounce cards — a stack fans out with a bounce Strands — flowing luminous fibers
bounce-cards · entrance
A card stack fans out with a staggered bounce as it enters the view.
strands · GPU shader
Luminous fibers drifting like hair in water — a quiet hero backdrop.
Silk — flowing silky fabric waves Waves — perlin-warped line field with a pointer wake
silk · GPU shader
Flowing fabric, ported from a three.js scene to a single fragment shader.
waves · canvas 2D
A perlin-warped line field with a pointer wake — organic, editorial, calm.
Faulty terminal — CRT glitch grid Pixel blast — drifting pixel pattern with click ripples
faulty-terminal · GPU shader
A glitching CRT grid for cyber / dev-tool moods.
pixel-blast · GPU shader · interactive
A drifting pixel pattern that ripples outward from every click.

Kinetic text

Motion is not only backgrounds — text is a first-class citizen. These run on plain DOM, dependency-free:

Decrypted text — glyphs resolve into ACCESS GRANTED Text scramble — a techy decode reveal between phrases
decrypted-text · kinetic text
Random glyphs resolve left-to-right into the real line — hacker-movie energy, one attribute to use.
text-scramble · kinetic text
Decode-reveal between any phrases, programmable via scrambleTo(el, text).
Count up — stats rolling from zero to their values True focus — a focus frame hops word to word, blurring the rest
count-up · kinetic text
Numbers roll up to their value on view — the stat-block moment, eased properly.
true-focus · kinetic text
A focus frame hops word to word while the rest stays blurred — spotlight your claim.

35 of the library's distinctive effects are faithful dependency-free ports — see ATTRIBUTION.md for sources and permissions. The full library holds 403 recipes.


Product tour

Home — describe the motion you want in one sentence
Home — one sentence in, animated artifact out. Pick a design system (59 brand packs) and a motion profile (Subtle → Cinematic), or just type.
Workbench — component-level motion editing on the running page, with the agent's plan, a keyframes inspector and the timeline
Workbench — the heart. The agent plans, generates and self-checks on the left; click any component on the running page and give it motion — triggers, presets or a 6-track keyframe editor with scrub + auto-keyframe. The timeline below warns when a view exceeds the restraint budget.
Video editor — layers, five scenes with transitions, film background and reference upload
Video editor — a canvas compositor for launch videos: text / shape / image / video layers, five scenes with real transitions, a shared film background, kinetic typography, and in-browser WebCodecs MP4 export. No watermark, nothing uploaded.
Motion path — draw a curve on the canvas and the selected text travels it
Motion paths — draw a line on the canvas, and the selected layer travels it: as drawn, or snapped to a clean line / circle / ellipse, with eased speed and per-character text motion on top.
Motion presets — entrance, emphasis, attention and exit presets stack onto any layer
Motion presets — entrance / emphasis / attention / exit presets stack onto any layer (text, image, video, shapes), plus the full 200+ recipe library one click away.
Library — recipe cards with live animated previews
Library — 403 curated recipes with live card previews, searchable by intent. Every recipe carries avoid_when and a restraint budget: taste is enforced, not hoped for.
Engines — 8 coding-agent CLIs plus BYOK
Engines — bring the agent you already pay for. 8 CLIs auto-detected on your PATH, plus BYOK for direct Anthropic / OpenAI / Google API calls. Keys never leave your machine.
Dark mode
Dark mode — the whole app, both ways. Follows your OS in system mode.

Why motion-anything

Motion is one of the highest-leverage, lowest-understood parts of digital craft. Four problems keep great motion out of reach:

  1. AI-generated pages are dead ends — you can generate a landing page, but tweaking its motion means re-rolling the whole page or hand-writing CSS. There is a gap between generating and refining. motion-anything edits motion on the running page, per component, and writes it back to the file.
  2. AI has no taste — default output is either everything-fades-in-at-once or a fireworks show. Here, taste is a feature: every recipe declares avoid_when and a restraint budget; the editor warns when a view exceeds it; prefers-reduced-motion is always honored; GPU-safe properties only.
  3. The ecosystem is scattered — GSAP, Framer Motion, anime.js, Lottie… each a learning curve. You express intent ("a liquid-metal background"), the router picks from curated recipes. You never need to know a library name.
  4. Tools lock you in — proprietary models, per-render fees, watermarks, handoff-only artifacts. This is Apache-2.0, local-first, engine-agnostic, and everything exports.

vs. Figma Motion — their artifact is a handoff; ours is the running page itself. Their timeline has no interaction triggers; hover and click are first-class here. Their motion lives in their app; ours exports to JSON / CSS / React / Lottie or a portable skill any agent can use.


Open Design ecosystem

motion-anything is not a standalone island — it is the motion layer of the Open Design family, and the interop is real, not a logo wall:

  • Every recipe is an Open-Design-compatible SKILL.md — export any of the 230 portable skills and drop it straight into Open Design's skills/ (or any agent that reads skills).
  • 59 brand packs in Open Design's DESIGN.md format — generation starts on-brand; the same contract your team already uses in OD drives motion here.
  • 58 HyperFrames video templates + 112 OD HTML prototype templates shipped in the library — briefs get matched against them so output starts from proven structure, not a blank canvas.
  • Open Design Cloud is a first-class engine — the same ACP runtime (vela) that powers OD drives motion-anything, with one recharge across GPT / Claude / Gemini / DeepSeek.
  • Shared craft assets — agent engine icons, the icon library (2,680 icons · 38 categories), and the design language all come from the same family.

The loop: pick a beautiful design system (OD's superpower) → generate a brand-grade page → give it motion, per component (our superpower) → export it back as a skill anyone's agent can reuse.


Quickstart

git clone https://github.com/nexu-io/motion-anything.git
cd motion-anything
node cli/bin/motion.js serve 4399
# open http://localhost:4399

That is the whole install. No npm install — the project has zero dependencies. You need Node 18+ and at least one agent engine: any supported CLI on your PATH (it will be auto-detected), or an API key for BYOK (Settings → Execution mode).


Engines

Your prompt runs on your agent — the CLI session you already pay for, or your own API key. Nothing is proxied through anyone's server.

Engine Vendor Transport
Claude Code Anthropic headless -p + stream-json
Codex CLI OpenAI exec --json event stream
Cursor Agent Cursor stream-json
OpenCode OpenCode run --format json event stream
Grok Build xAI plain text via prompt file
Hermes xAI ACP (JSON-RPC over stdio)
Gemini CLI Google plain text
Open Design Cloud nexu.io ACP via the vela CLI
BYOK Anthropic / OpenAI / Google direct API with your key

The library

  • 403 motion recipes — 77 real effects with live previews, plus curated reference cards. Categories: ambient backgrounds, feedback & delight, interaction, text, transitions.
  • Standardized manifests — every recipe is a folder with recipe.motion.yaml, a self-contained preview.html, the implementation, and a SKILL.md. Three fields you will not find elsewhere: intent_keywords, avoid_when, restraint.
  • 59 design-system brand packs + 58 video templates + 112 HTML prototype templates + 2,680 icons — generation starts brand-grade, not blank.
  • 230 portable skills — export any recipe as a SKILL.md and drop it into Open Design or any agent that reads skills.

The taste contract lives in MOTION-SPEC.md. The library golden path (how to add a recipe) lives in AGENTS.md.


Export anything

From To
Any component's motion JSON (portable) · CSS · React · Lottie
Any animated page MP4 (freezable-timeline frame capture) · GIF · single-file HTML
Any launch video MP4 via in-browser WebCodecs — no watermark, no upload, no fees
Any recipe a portable SKILL.md for your agent

Architecture

Plain files, no build step, zero npm dependencies — the whole app is deliberately boring to run:

  • app/index.html — the entire client (workbench, editors, library, i18n ×4) in one file.
  • cli/bin/motion.js — server + CLI in one file: static serving, project store, and the engine dispatch (stream-json / JSONL events / plain / ACP / BYOK behind one interface).
  • app/video/ — the canvas compositor: engine, WebCodecs MP4 export, HTML frame capture, vendored mp4-muxer + gifenc (MIT).
  • recipes/ — the library. recipes/web/_fx/shaderbg.js (58 lines) is the dependency-free WebGL runner behind all shader recipes.
  • skills/ — the router skill + bundled companion skills.

Agents are first-class citizens of the codebase itself: AGENTS.md is a working agreement any coding agent can follow, and PROGRESS.md keeps the state.


Roadmap

  • v0.1 — recipe library + router skill + workbench + video line + 8 engines + BYOK ← we are here
  • v0.x — richer video motion (recipe reuse inside video), deeper Figma import, streaming BYOK
  • v1 — auto-add motion to existing Open Design artifacts; design-component → motion auto-assign

License & credits

Apache-2.0. A project in the nexu.io / Open Design family.
Third-party sources and permissions are recorded in ATTRIBUTION.md.


📖 中文文档见 README.zh-CN.md

Reviews (0)

No results found