motion-react

motion-react — Motion for React

Overview

Implements UI animation in React with Motion — the library previously published as framer-motion, independent since late 2024 and published as the npm package motion (import from "motion/react"). The skill covers the DOM product-UI animation surface: declarative enter/exit, variant orchestration, layout and shared-element transitions, motion values and number tickers, and gesture states — with three non-negotiables woven through: reduced-motion accessibility, compositor-friendly performance, and restraint (purposeful, short, consistent motion — especially in data-dense tools). It also owns the "do I even need Motion?" decision: many jobs are better served by plain CSS or a one-line AutoAnimate call.

When to activate

• ✅ Adding enter/exit/list animations, page or view transitions, shared-element moves, animated counters, or gesture feedback to a React SPA.
• ✅ Migrating or writing code that references framer-motion (the old package) — this skill teaches the current motion idioms.
• ✅ Wiring reduced-motion support, or auditing existing animation for accessibility/performance.
• ✅ Choosing between Motion, CSS/Tailwind transitions, and AutoAnimate for a given interaction.

Do NOT activate when:

• The task is animated illustrations / Lottie files — that is imagery, not UI motion (a dedicated illustrations skill owns it).
• The task is video rendering (Remotion) or scroll-storytelling/parallax marketing pages — different genre; only whileInView basics are covered here.
• The product's motion design language (which durations/easings a specific app adopts) is being decided — that is a design-system decision; this skill supplies the vocabulary and constraints.

Workflow

Step 1: Choose the right tool — most animations don't need Motion

Job	Reach for	Why
Hover/focus color, opacity, small scale	Plain CSS / Tailwind transition-*	Zero JS, zero bundle, GPU-friendly
Open/close styling on Radix-based components (accordions, popovers, dialogs)	CSS keyed off Radix data-state attributes	The primitive already exposes state; CSS animates it
List add / remove / reorder in an internal tool	AutoAnimate — one useAutoAnimate() ref on the parent	Zero-config, animates child add/remove/move, respects prefers-reduced-motion automatically
Exit animations on unmount, orchestrated sequences, shared-element moves, springs, number tickers, drag / drag-to-reorder (Reorder components)	Motion	Needs JS lifecycle control CSS cannot express
Browser-native view morphs (same-document, or cross-document/MPA)	View Transitions API	Named for awareness; for view/route swaps in a React SPA, AnimatePresence covers the equivalent (whole-document morphs like a theme crossfade remain VTA-only)

Escalate down the table only when the simpler tier genuinely cannot do the job.

Step 2: Install and set the app-level posture

npm install motion

import { motion, MotionConfig } from "motion/react"

// App root — respect the user's OS reduced-motion setting everywhere:
<MotionConfig reducedMotion="user">
  <App />
</MotionConfig>

• Package is motion; React entry is "motion/react". React 18+ (docs state 18.2; the peer range is ^18.0.0 || ^19.0.0). Anything importing framer-motion is the pre-rename package — same lineage, but new projects use motion and current docs/idioms live under that name.
• MotionConfig reducedMotion accepts "user" (respect the OS setting), "always" (force-reduce; debugging), "never" (default). Under reduction, Motion disables transform and layout animations while keeping opacity/color animations — usually exactly the right degradation. Set "user" at the root as the default posture; see references/a11y-and-restraint.md for the full contract.
• The reduced-bundle wiring is one wrapper: <LazyMotion features={domAnimation} strict><m.div animate={{ opacity: 1 }} /></LazyMotion>.
• React Server Component frameworks: motion components are client components — add a "use client" boundary (or import from motion/react-client); a plain Vite SPA needs neither.
• Bundle-sensitive apps: LazyMotion + the m component (import * as m from "motion/react-m") renders for ~4.6 kB initially (vs ~34 kB for full motion), with features (domAnimation) loaded sync or — for the full saving — async after hydration; strict on LazyMotion throws if a full motion component sneaks in.

Step 3: Core animation

<motion.div
  initial={{ opacity: 0, y: 8 }}
  animate={{ opacity: 1, y: 0 }}
  exit={{ opacity: 0 }}
  transition={{ duration: 0.2, ease: "easeOut" }}
/>

• initial → animate runs on mount; exit runs on unmount only inside AnimatePresence (Step 5).
• transition picks the model: type: "tween" (duration + easing — UI default) or type: "spring" (physical; use for gestures/drag, sparingly for entrances). Keyframes: pass arrays (animate={{ x: [0, 100, 0] }}); null as the first keyframe means "current value".
• Gesture states: whileHover, whileTap, whileFocus, whileDrag, whileInView — values revert when the gesture ends. Prefer CSS for hover-only styling (Step 1); use these when the gesture must coordinate with other animated values.
• Product-UI durations: 150–300 ms for micro-interactions, ≤ 400 ms for larger transitions. Longer reads as lag, not polish.

Step 4: Variants — orchestrate parent and children

import { motion, stagger } from "motion/react"

const list = {
  hidden: { opacity: 0 },
  visible: {
    opacity: 1,
    transition: { when: "beforeChildren", delayChildren: stagger(0.05) },
  },
}
const item = { hidden: { opacity: 0, y: 4 }, visible: { opacity: 1, y: 0 } }

<motion.ul variants={list} initial="hidden" animate="visible">
  {rows.map(r => <motion.li key={r.id} variants={item} />)}
</motion.ul>

• Variant names flow down: children with variants animate when the parent's state changes — no prop drilling.
• Orchestration lives in the parent's transition: when: "beforeChildren" | "afterChildren", and delayChildren: stagger(interval) — the current idiom, available since motion 12.22.0 (staggerChildren/staggerDirection were deprecated in 12.21.0). On older installs (framer-motion, early 12.x) staggerChildren: 0.05 still works and stagger()-in-delayChildren is not supported (the option was a plain number there) — bump the package before migrating the idiom.
• Dynamic variants: a variant can be a function of custom (e.g. per-index delay).

Step 5: AnimatePresence — exit animations

import { AnimatePresence, motion } from "motion/react"

<AnimatePresence mode="popLayout">
  {rows.map(r => (
    <motion.div key={r.id} exit={{ opacity: 0, x: -8 }} layout />
  ))}
</AnimatePresence>

• Works by watching its direct children; every child needs a stable, unique key (array indices break exit animations — the classic failure).
• mode: "sync" (default — in/out simultaneously), "wait" (out fully before in; one child at a time — right for view/route swaps), "popLayout" (exiting elements pop out of layout flow so siblings reflow — right for lists; the parent needs non-static position).
• In-place content swaps (skeleton→content, tab panels): with "wait", match the two children's dimensions or the gap flashes; when dimensions differ, crossfade in place via "popLayout" or a grid-stack/absolute overlay — and pair the container with layout if the size change itself should animate rather than snap.
• initial={false} skips mount animation for children present at first render. onExitComplete fires when all exits finish.
• Exiting components cannot receive prop updates — for direction-aware exits (carousels, paginated views), pass the direction via AnimatePresence's custom prop and read it with usePresenceData (or variant custom) inside the child.
• AnimatePresence must wrap the conditional — never be inside it.

Step 6: Beyond the core — load the reference that matches the job

• Layout animations, layoutId shared elements, LayoutGroup, distortion fixes → references/layout-and-shared-elements.md
• Motion values, useTransform/useSpring, imperative animate()/useAnimate, animated number tickers for stat tiles → references/values-and-imperative.md
• The full reduced-motion contract, WCAG mapping, reduce-vs-keep table, motion-restraint tokens → references/a11y-and-restraint.md
• Radix/shadcn integration (forceMount), Tailwind boundaries, router page transitions, the AutoAnimate worked example, and testing (MotionGlobalConfig.skipAnimations, Playwright reduced-motion) → references/integration-and-testing.md

Rules

Hard rules (never violate):

• Respect reduced motion. Every Motion-using app sets MotionConfig reducedMotion="user" (or an equivalent explicit strategy) at the root. Never ship large translations, zooms, parallax, or auto-playing motion that ignores the OS preference (WCAG 2.3.3; autoplaying

  5 s motion needs a pause control per WCAG 2.2.2).

• Animate transform and opacity, not layout-inducing properties. Never animate width/height/top/left/margins directly — use the layout prop (which animates via transforms) or scale; cheap non-layout properties (color/background, SVG stroke values) are fine. Scale-based fills (progress bars, meters) need an explicit transform origin (originX: 0 for a left-anchored fill) — the default is center. One sanctioned exception: expand/collapse reveals (accordion, disclosure) animate height: 0 ↔ "auto" with overflow: hidden — scale distorts text and layout can't clip-reveal; keep it short and user-initiated. Note reduced motion does NOT auto-reduce it (only transform/layout animations are disabled) — if the reveal displaces substantial content, gate it with useReducedMotion() to snap open instead.
• Stable keys inside AnimatePresence. Array indices as keys silently break exit animations.
• Import from "motion/react". Do not add framer-motion to a new project; do not mix both packages.
• Motion must be purposeful. Animation communicates state change, spatial relationship, or feedback — decorative-only animation in a product UI is a defect, not polish.

Preferences (override-able):

• Micro-interactions 150–300 ms; route/page swaps ≤ 200 ms; other view-level transitions ≤ 400 ms (the outer bound, not the target).
• Tweens for UI state changes; springs for gesture-driven/physical interactions.
• Centralize durations/easings as design tokens rather than scattering literals.
• Prefer the simplest sufficient tier (Step 1) before reaching for Motion.

Gotchas

• Exit animation never plays: the element isn't a direct child of AnimatePresence, its key is unstable, or the AnimatePresence itself unmounts with the element. Fix the tree, not the animation.
• Old-package drift: tutorials and LLM completions frequently emit framer-motion imports, staggerChildren, or motion(Component)-era idioms. Current: motion pkg, delayChildren: stagger(), and motion.create(Component) to wrap a custom component (it must expose its ref — forwardRef on React 18, the plain ref prop on React 19); check the live docs when an API looks dated.
• popLayout needs positioning: exiting elements are absolutely positioned; a static-positioned parent misplaces them.
• Layout animation distorts rounded corners/shadows: set borderRadius/boxShadow via style (or animate them) so Motion can scale-correct; give stretching children their own layout prop.
• Springs on mount feel bouncy-slow: physics-configured springs (stiffness/damping) derive their own settle time and ignore duration expectations; use a duration-based spring (type: "spring", duration, bounce) or a 200 ms tween for entrances.
• Motion values don't re-render React: rendering {x.get()} in JSX goes stale — render the motion value as a motion element child or subscribe via useMotionValueEvent.
• Tests hang or flake on animations: set MotionGlobalConfig.skipAnimations = true in the test setup and assert end-states (see the integration reference).
• jsdom has no layout: layout animations and whileInView need real measurements; unit-test the state logic, e2e-test the visuals.

Anti-patterns

• "Animate everything — it looks premium." Data-dense tools earn polish through a few consistent, short, purposeful animations; motion noise slows comprehension.
• "Skip reduced-motion for now, add it later." It's a root-level one-liner when done first and an audit when done later. There is no "later".
• "Just animate the height." Layout properties jank on the main thread; layout prop or a different design.
• "AutoAnimate is beneath us — rewrite the list in Motion." If add/remove/move is all the list needs, one ref beats thirty lines of variants.
• "Copy the framer-motion snippet from that 2023 blog post." Verify against current docs; the package, imports, and stagger idiom changed.

Output

Working animation code in a React codebase: components using motion/react primitives (or deliberately simpler tiers per the Step-1 table), an app-level reduced-motion posture, and test setup that keeps animated components deterministic. Consumers are the codebase's reviewers and test suite — animation that ships with accessibility, performance, and restraint already satisfied.

Related

• A UI/UX design-guidelines skill (e.g. one carrying animation duration/meaning rules) is the METHOD source for when/what to animate; this skill is the implementation counterpart.
• A component-styling skill (Radix/Tailwind) owns the data-state CSS tier of the Step-1 table.
• A router skill (e.g. TanStack Router) owns route structure; this skill only wraps route views for transitions.
• An illustrations skill owns Lottie/animated imagery — adjacent, not this skill's surface.
• A component-testing skill owns the RTL/vitest harness this skill's testing guidance plugs into.

Progressive disclosure

• references/layout-and-shared-elements.md — load when animating layout changes, shared elements (layoutId), or reordering lists.
• references/values-and-imperative.md — load for motion values, transforms/springs, imperative sequences, or animated number tickers.
• references/a11y-and-restraint.md — load when wiring reduced-motion, auditing accessibility, or defining motion tokens/restraint rules.
• references/integration-and-testing.md — load when combining Motion with Radix/shadcn, Tailwind, or a router, when using AutoAnimate, or when tests involve animated components.
• references/sources.md — research provenance; load only when auditing claims.

Body budget

• description ≤ 1,024 chars (agentskills.io cap).
• Body ≤ ~500 lines / 5,000 tokens soft target; overflow lives in references/.