animation-judgement
UI animation rules from Emil Kowalski's "Train Your Judgement" (emilkowal.ski/ui/train-your-judgement). Use when writing, reviewing, or critiquing motion — transitions, entries, exits, hovers, popover
SKILL.mdskills/animation-judgement/SKILL.md
GitHubAnimation Judgement
11 rules for making UI motion feel right instead of just "working." Source: Emil Kowalski, Train Your Judgement.
When to use this skill
Apply proactively when:
- Adding or tweaking any
transition-*,@keyframes,animate-*,duration-*,ease-*class. - Writing
motion.*orAnimatePresencefromframer-motion/motion. - Reviewing hover, press, entry, exit, or popover behavior.
- User says a motion "feels off," "sluggish," "janky," "too much," or "delightful but annoying."
Don't invoke for non-motion work.
Framing
"AI can write animation code. What it can't do is know what feels right. It produces motion that works, but feels mediocre, and if you can't tell the difference, you'll ship it."
The rules below are the named, articulable reasons behind "that feels off."
The 11 Rules
1. Duration scales with distance
- Default: keep UI animations under 300ms (tooltips, dropdowns, buttons).
- Exception: large elements (full-screen menus, panels, sheets) need longer durations — a 150ms full-screen slide feels jarring.
- Mental model: a truck takes longer to stop than a bicycle. Heavier-looking things should animate slower.
2. Entry animations use ease-out
ease-instarts slow → feels sluggish. Never use for entries.ease-outstarts fast (instant feedback) and decelerates gently → feels elegant and faster.- Exits can use
ease-in(element accelerating away).
3. Never scale(0) — start from 0.90–0.95
scale(0)means "from nowhere" — unnatural.- Real objects always have a shape (a deflated balloon still has form).
scale(0.9)toscale(0.95)preserves the illusion the element was "always almost there."
4. Bounce is brand personality, not default behavior
- Bounce = playful. Zero bounce = professional.
- Learning/consumer app: bounce can add energy.
- Finance/enterprise app: bounce feels unserious, wrong for the context.
- Default: zero bounce. Add intentionally.
5. High-frequency interactions = no animation
- Animation on something used hundreds of times/day becomes annoyance, not delight.
- Hover-highlight fade looks smoother in a demo, but trails the cursor in real rapid use — the highlight is always one step behind.
- Instant highlights track the mouse perfectly. Direct connection > smoothness.
- Example: Raycast doesn't animate its open because frequent users would hate it.
- Rule of thumb: if the user will trigger this >10×/minute, strip the animation.
6. Button press = scale(0.97), not 0.9
0.9(10% shrink) is visibly too aggressive in most product UIs.0.97is barely visible but still felt — subtlety reads as quality.- Go lower (0.92–0.95) only for intentionally playful brands.
- Red flag: if you can clearly see the button shrinking in a recording, it's too much.
7. Use mode="popLayout" when removing list items
- Default
AnimatePresence: exit animation finishes → then siblings reflow into the gap. Causes a visible pause. mode="popLayout": exiting element leaves document flow on frame 1; siblings reflow in parallel with the exit. Feels tight.- Applies to chips, tags, list rows, toast stacks — anything where removal triggers sibling reflow.
8. CSS transitions > CSS keyframes for toggleable UI
- Keyframes jump to the new end state when re-triggered mid-animation.
- Transitions smoothly animate to the new target from wherever they currently are.
- Applies to menus, modals, accordions, drawers — anything a user can toggle rapidly.
- If a user rapid-clicks and sees a jump, you used keyframes where you wanted a transition.
9. Popovers: transform-origin at the trigger
- Default
transform-origin: centermakes popovers appear "from nowhere." - Origin at trigger location creates cause-and-effect: "I clicked this, this came from it."
- Radix UI and Base UI expose CSS variables for this (
--radix-popover-content-transform-origin, etc.). Use them.
10. Stagger by hierarchy, never uniform
- Uniform stagger (same delay, duration, distance per element) = the motion equivalent of
lineareasing. Feels artificial. - Vary timing: the most important element appears first and gets screen time alone before supporting elements follow.
- Decorative elements (badges, pills) can fade-only — no slide — because they matter less.
- Questions to ask before staggering: Which element matters most? Which is supporting? Which is decorative?
11. Don't stack entry animations
- Panel slides in → then menu items stagger in = two layers; user waits twice.
- Panel slides in → content is just there = faster, cleaner, readable immediately.
- "Sometimes the best animation is no animation."
Meta-principles
- Demo-polish ≠ product-quality. (Rule 5) What looks better in a 5-second demo can be worse in actual repeated use.
- Subtlety reads as quality. (Rule 6) If the effect is obvious, it's probably overdone.
- Absence is a valid choice. (Rules 5, 11) "No animation" is often the right answer.
- Context determines correctness. (Rule 4) Bounce, stagger, duration all depend on brand + frequency, not universal "best."
How to apply in review
When reviewing motion code, check in this order:
- Does the element need animation at all? (Rule 5, 11)
- Is duration appropriate for the element's size? (Rule 1)
- Entry uses
ease-out? (Rule 2) - Scale starts from 0.9+, not 0? (Rule 3)
- Button press ≤
scale(0.97)? (Rule 6) - Removals use
popLayout? (Rule 7) - Toggleable UI uses CSS transitions? (Rule 8)
- Popover
transform-originanchored to trigger? (Rule 9) - Stagger has varied timing by importance? (Rule 10)
- Bounce intentional for the brand, not accidental? (Rule 4)
Source
https://emilkowal.ski/ui/train-your-judgement — captured 2026-04-19. Interactive page with 11 A/B exercises; each rule's "Vocabulary" line above is Emil's verbatim compression.