ui-animation

UI Animation

• IS: designing, implementing, reviewing, debugging UI motion (springs, gestures, drag, easing, CSS transitions, keyframes, framer-motion), sweeping an interface for the moments that would genuinely benefit from motion, measuring motion from a recording (extract frames, track, fit curves) to emit code plus a handoff spec, and naming a described motion effect (reverse-lookup vocabulary).
• IS NOT: choosing overall visual direction, palettes, or typography (use ui-design Direction mode), auditing a whole page's UI quality (use ui-design Audit mode), or named text-effect specs (use the external animate-text skill where installed).

product-design, ui-design, or ui-animation?

An interface is a set of states and the passages between them. That decomposition assigns the work.

• Subject beats artifact. When motion is what the request is about, it is this skill whether or not code exists yet.
• Artifact is the opening presumption, not the verdict. Code, a diff, or a running UI in hand presumes ui-design; a brief, spec, mockup, or intent with no code presumes product-design. Either can be overturned by the two tests below.
• Capability beats presentation. With code in hand, ask whether the change alters what a user can do, which objects an action affects, whether it is reversible, or whether a state exists at all. That is a capability, so product-design decides. If it only changes how the same capability looks or reads, ui-design owns it.
• A gesture that replaces a control is a capability decision. Swipe-to-delete, hold-to-confirm, and drag-to-reorder change what the user can do and how recoverable it is, so product-design settles the interaction and this skill builds its physics.
• Motion incidental to a build stays in ui-design. A hover transition or a fade added while building a component is a property of that component. It arrives here when motion is the subject or its craft is in question.

Worked: "Delete should be undoable" is product-design. "The undo toast is ugly" is ui-design. "The undo toast should slide, not pop" is this skill.

Where the choice is between product-design and ui-design and motion is not the subject, those two carry two further tiebreaks (control patterns with different reachability, and undebatable missing states). Neither changes an answer here.

Canonical home for reverse-engineering motion from a recording: route "reverse engineer this animation" and "match this easing" here, not to a separate skill. If the input is a screen recording or video, you are MEASURING motion: follow the Reverse-engineer workflow. Otherwise (designing, implementing, reviewing) use the rules and Workflow below.

Reference files

Core rules

• Animate for feedback, orientation, continuity, or deliberate delight. If it's just "it looks cool" and the user sees it often, don't.
• Never animate keyboard-initiated actions (shortcuts, arrow navigation, tab/focus); they repeat constantly and animation makes them feel slow.
• Prefer CSS transitions for interruptible UI: keyframes restart from zero on interruption, transitions retarget. Use keyframes only for predetermined sequences.
• Implementation priority: CSS transitions > WAAPI > CSS keyframes > JS (requestAnimationFrame); under load CSS stays smooth while JS drops frames.
• Asymmetric timing: occasional interactions can enter slightly slower, exit fast. High-frequency ephemeral UI (hover highlights, popovers, panel toggles) inverts this: enter instantly (0ms), exit with a brief fade (100-150ms) so the action feels immediate.
• Tappable controls press on :active at 0ms and set touch-action: manipulation.
• Use @starting-style for DOM entry; fall back to a data-mounted attribute where unsupported.
• A small filter: blur(2px) hides rough crossfades between swapped content.

Motion design principles

• Continuity over teleportation. Elements visible in both states transition in place; expand from where elements sit rather than fading in a new instance. Never duplicate a persistent element or hard-cut between views that share components; hard cuts lose spatial context.
• Directional motion matches position. Tab and carousel transitions animate in the direction matching spatial layout (left-to-right forward, right-to-left back).
• Emerge from the trigger. Overlays, trays, and panels animate outward from the element that opened them; generic centre-screen entrances break spatial orientation. Better still where the shapes allow: let the trigger become the surface (see the container-morph recipe).
• Confirm in place, not in a corner. An action's result belongs on the control that caused it: the button becomes "Copied", holds, and reverts. A toast in the far corner makes the user's eye leave the thing they just touched to find out whether it worked. Reserve corner toasts for results with no on-screen origin (a background job finishing, an incoming message).
• Animate paired states together. If open animates, close animates. If hover has motion, focus and pressed states get equivalent feedback. Do not polish only one half of a repeated interaction.
• Delight scales inversely with frequency. Rarer interactions get more personality; high-frequency actions must be invisible.
• Motion enhances perceived speed. Smooth transitions feel faster than hard cuts, even at identical load times.

What to animate

• Movement: transform and opacity only; they skip layout and paint.
• State feedback: color, background-color, and opacity are acceptable.
• Never animate layout properties (width, height, top, left); they trigger layout recalc every frame. (Exception: a deliberate container tween, see the card-resize and container-morph recipes.)
• Never use transition: all; it animates unintended properties and silently adopts future ones. List them explicitly.
• Avoid filter animation for core interactions; if unavoidable keep blur ≤ 20px (heavy blur is expensive, especially in Safari).
• SVG: apply transforms on a <g> wrapper with transform-box: fill-box; transform-origin: center; without it they rotate/scale around the canvas origin.
• transform: scale() also scales children (icons, text, borders scale proportionally), unlike width/height: a feature for press feedback, but account for it when an inner element must stay fixed-size.
• Disable transitions during theme switches ([data-theme-switching] * { transition: none !important }), or every themed property animates at once.

Easing defaults

Keep routine UI under 300ms; scale duration with distance (a full-screen slide can exceed 300ms, a 6px tooltip shift stays under 150ms).

Named curves

• Enter: cubic-bezier(0.22, 1, 0.36, 1) for entrances and transform-based hover
• Move: cubic-bezier(0.25, 1, 0.5, 1) for slides, drawers, panels
• Drawer (iOS-like): cubic-bezier(0.32, 0.72, 0, 1)

Avoid ease-in for UI: it starts slow, so the element lags the user's action and feels sluggish. Prefer custom curves from easing.dev over built-in ease/ease-out, whose gentle acceleration reads soft, not decisive.

Transition decision rules

Match the UI element first, then pick the recipe from references/transition-recipes.md:

Prefer lower-overhead transitions (CSS-only) unless the design requires JS orchestration.

Spatial and sequencing

• Popover transform-origin at the trigger (modals stay center), dialog/menu entrances from scale(0.85-0.9) not scale(0), and 30-50ms staggers (total under 300ms, most important element leading). Full rules and code in references/component-patterns.md and references/contextual-animations.md.
• Paired elements rule: elements that animate together (modal + overlay, tooltip + arrow, FAB + label) must share easing and duration. Mismatched timing is the usual cause of "something feels off".

Accessibility

• Every animation needs a prefers-reduced-motion: reduce path: disable transform/keyframe motion, keep instant state changes or opacity-only fades. All recipes include the guard.
• Gate hover (motion and paint) behind @media (hover: hover) and (pointer: fine), or touch devices replay hover on tap. Tailwind hover: is not gated unless the project set hoverOnlyWhenSupported or a custom variant.
• During direct manipulation, keep the element locked to the pointer with no easing; add easing only after release.

Performance

• Pause looping animations off-screen with IntersectionObserver; they burn GPU even when invisible.
• Toggle will-change only during heavy motion and only for transform/opacity; remove it after. Each promotion costs compositor memory; permanent promotion across many elements is worse than none.
• Do not animate drag via CSS variables on a container; every update recalculates styles for all children. Set transform directly on the moving element.
• Motion x/y values are the default for axis movement and drag (they bypass React re-renders). Use a full transform string only when one owner must combine multiple transform functions or interop with non-Motion code.
• See references/performance-deep-dive.md for WAAPI, compositing layers, and the CSS vs JS comparison table.

Anti-patterns

High-signal failures not covered above:

• Animating on mount without a user trigger: unexpected motion disorients; the user did nothing to cause it.
• Hard stops on drag boundaries feel broken; apply friction/damping so movement diminishes past it (see gesture-drag reference).
• Animating both a container and staggering its children: pick one entrance per container. If the panel slides in, its content should already be visible on arrival.
• Tooltip animation after the first is open: subsequent tooltips in the group open instantly, or the toolbar feels laggy.

Workflow

Copy and track:

Animation progress:
- [ ] Step 1: Decide whether the interaction should animate
- [ ] Step 2: Choose purpose, easing, and duration
- [ ] Step 3: Pick the implementation style
- [ ] Step 4: Load the relevant component or technique reference
- [ ] Step 5: Validate timing, interruption, and device behavior

1. Answer the four questions in references/decision-framework.md: animate? purpose? easing? speed?
2. Pick duration from the easing defaults table above.
3. Choose implementation: CSS transition > WAAPI > spring > keyframe > JS.
4. Load the reference for your component or technique.
5. When reviewing, apply the strict posture in references/review-format.md: measure against the ten standards, output the Before/After/Why table, then a tiered verdict ending in a Block/Approve decision.

Validation

Produce evidence for each check (DevTools observations, not "looks fine"):

• Grep the diff for layout property transitions (width, height, top, left) and transition: all.
• Retoggle components rapidly; confirm transitions retarget instead of restarting from zero.
• Slow to 10% in the DevTools Animations panel to catch timing and transform-origin issues invisible at full speed.
• Emulate prefers-reduced-motion: reduce (DevTools Rendering panel) and confirm every animation has a reduced path.
• Confirm will-change is toggled around animations, not permanently set, and looping animations pause off-screen.
• Test touch interactions on real devices; simulators under-report gesture and hover-on-tap issues.
• Review again with fresh eyes the next day; imperfections missed during development stand out.

Discovery workflow

Use this branch when the request is "where should this animate", not "animate this". Every other mode starts from motion that exists; this one starts from its absence. It reports and never implements: hand a surviving suggestion back to the main workflow above to build it.

Discovery progress:
- [ ] Step 1: Recon the stack, existing motion tokens, and product personality
- [ ] Step 2: Sweep every seam class
- [ ] Step 3: Gate each candidate
- [ ] Step 4: Report survivors and rejections

1. Recon. Identify the motion library (if any), the easing and duration tokens already in use, and how often each surface is visited. Suggestions extend the existing vocabulary rather than introducing a parallel one, and a dense dashboard earns fewer and subtler suggestions than a playful consumer app.
2. Sweep. Walk the seam table in references/decision-framework.md, which carries the grep signature for each. Clear a seam explicitly rather than skipping it silently.
3. Gate. Run each candidate through questions 1 and 2 of the same file: frequency, then purpose. "It looks cool" is not a purpose. Most candidates die here, which is the point.
4. Report. Cap at five to seven suggestions ordered by leverage, each with file:line, what happens today, the named purpose, the frequency tier, and exact values (property, duration, curve) drawn from the tables above. Then list two to five rejected candidates, each naming the question that killed it. Close with which single suggestion has the highest leverage.

Where the interface already carries the right amount of motion, say so. That is the correct result for a well-built UI, not an empty report.

Reverse-engineer workflow

Use this branch to measure an existing animation from a screen recording, then emit code and a handoff spec that reproduce it. The scripts under scripts/ are the canonical, deterministic path; run them rather than reconstructing their logic.

Dependencies: ffmpeg for frame extraction (brew install ffmpeg); Python with pip install opencv-python numpy scipy for tracking and curve fitting. Degrades gracefully: with only ffmpeg you can extract frames and reason visually; tracking and fitting need the Python packages.

Reverse-engineer progress:
- [ ] Step 1: Extract frames + contact sheet (per direction if open differs from close)
- [ ] Step 2: Vision pass: identify element, effects, phases
- [ ] Step 3: Decide precision (eye-only vs scripted)
- [ ] Step 4: Track motion and fit curves (if escalating)
- [ ] Step 5: Annotate choreography (delays, asymmetry)
- [ ] Step 6: Emit code for the target(s)
- [ ] Step 7: Validate against the recording

1. Extract. Run python3 scripts/extract_frames.py <video> <outdir>. Trim to just the transition with --start/--duration; if the interaction has both an open and a close, trim two windows and run the pipeline once per direction (they are almost never mirror images). Match --fps to the source (probe with ffprobe), never sampling above the source rate. Open contact_sheet.png first.
2. Vision pass. Name the element(s) that move, every effect (translate, scale often anisotropic, opacity, blur, corner radius, shadow, color), and the phases, noting which property leads and lags. Use the checklist in references/measurement-guide.md.
3. Decide precision. Simple fade or linear slide: read timing off the contact sheet, skip to step 5. Elastic, springy, or multi-property motion: escalate to step 4 (eyeballing a spring is unreliable).
4. Track and fit. Run python3 scripts/track_motion.py <outdir> for metrics.json (pass --bbox X,Y,W,H to isolate one element), then python3 scripts/fit_curves.py <outdir>/metrics.json for spring params, cubic-bezier, and per-property fit error. Pass the same --fps you extracted with. Read references/curve-fitting.md to pick the model; high error on both means multi-phase motion (split and fit each segment).
5. Annotate. Load references/choreography.md. Build the timing-offset table (when each property starts and settles); lead/lag gaps and over-stretch carry more feel than any single curve.
6. Emit. Substitute fitted parameters into the templates in references/code-output.md for the target. Keep movement on transform/opacity. Emit two transitions when open and close differ, plus the consolidated handoff spec so it can be implemented without the video.
7. Validate. Re-derive: play the emitted animation, screen-record it, run it back through extract_frames.py, and compare contact sheets side by side. Slow to 0.1x to confirm phase order and over-stretch survive. Confirm the code only animates transform, opacity, and filter.

Reverse-engineer gotchas:

• fit_curves.py defaults to --fps 30: extract at 60 but fit at the default and every duration_ms doubles while fitted stiffness drops to a quarter. Always pass the extraction fps to the fit.
• Sampling above the source rate duplicates frames: a 24 fps GIF extracted at 60 inflates fit error with plateaued runs in metrics.json. Probe and match the source rate.
• Screen recordings drop frames and iOS/QuickTime captures are variable-frame-rate; consecutive identical rows are duplicated frames, not a pause. Re-record at a steadier rate if plateaus dominate.
• Measure open and close as separate clips and report two curves; never fit one and reuse it reversed (see references/choreography.md). Treat a fit error above 0.08 as suspect.

Maintenance only: when changing Discovery routing or the gate, run the scenarios in evaluations/ as a regression rubric. They never load during a user task.

Related skills

• product-design: which states exist, what an action affects, and whether it is reversible. Route here first when a gesture replaces a control, since swipe-to-delete and hold-to-confirm change what the user can do before they change how it moves.
• ui-design Direction mode: visual direction, palettes, typography; settle the visual system before tuning motion.
• ui-design Audit mode: page/feature-level UI quality audit. Its motion- rules are the shallow presence check (animated layout properties, missing reduced-motion); the craft and the fix belong here.
• Optional external animate-text skill where installed: curated named text effects (typewriter, line reveal, stagger builds) with exact JSON specs.
• Taste Training (blode.co/taste-training): trains the eye these rules encode, across type, copy, craft, interaction, and motion.