Motion is where good designs go to stall — keyframing is tedious, so transitions ship as defaults or not at all. This playbook drafts keyframes and easing with AI in Figma Motion or Jitter, keeps timing and intent in your hands, checks the reduced-motion variant, and hands off through Dev Mode so engineering builds what you tuned.
Workflow area: Figma AI & Prototyping · Tier: Core · Hands-on time: ~30 minutes
Micro-interactions carry a disproportionate share of perceived quality: the save confirmation that settles instead of popping, the list item that slides out with weight, the loading state that breathes. Most teams under-invest not because they undervalue motion but because keyframe work is slow and the tooling gap between design and engineering swallows whatever gets made. AI drafting changes the first economics — a described transition becomes editable keyframes in seconds — but the draft always arrives generically pleasant: default springs, uniform timing, no opinion. Timing and intent are the design. That part stays yours.
Write the intent line before touching the tool. One sentence per interaction: what changed, and what the motion should communicate. "The task slides off and the list closes the gap — completing feels like progress, not deletion." No intent line, no animation; motion without a communicative job is decoration.
Draft with a prompt that specifies intent, not just geometry. In Figma Motion on your frames, or Jitter for standalone comps:
Animate the transition between these two frames (before/after
attached): a task item being marked complete.
Intent: completion should feel satisfying and final, but quick —
this happens dozens of times per session.
- Checkmark: quick confirm on tap, ~150ms, slight overshoot.
- Task row: fades and slides out, list below closes the gap
smoothly. Total sequence under 400ms.
- Stagger: checkmark first, row exit begins ~100ms later.
- Easing: decisive start, gentle settle — no bounce on the row.
Generate keyframes I can edit; keep layers separate per element.
Tune the draft — this is the actual design work. Play it at real speed, then at quarter speed. Adjust the three levers the AI cannot judge for you: duration (drafts run long; frequent actions want the fast end of 150–400ms), easing (replace default curves with your product's standard set), and choreography (what leads, what follows, what overlaps — sequence is meaning). Iterate until the motion says the intent line without you explaining it.
Produce the reduced-motion variant — it is a requirement, not a courtesy. Vestibular disorders make large movement, parallax and zoom physically unpleasant; WCAG expects motion triggered by interaction to be disableable. For each animation, define the prefers-reduced-motion behavior: usually a simple cross-fade or instant state change that preserves the feedback without the movement. Draft it the same way ("same transition, no translation or scale — opacity-only, 150ms") and check that the completion signal still reads.
Hand off timing as data, not adjectives. Export Lottie files where the format fits (icon animations, self-contained comps), and for interface transitions document the spec engineering actually needs — element, property, duration, delay, easing curve, reduced-motion behavior — inspectable in Dev Mode alongside the frames:
ELEMENT PROPERTY DURATION DELAY EASING REDUCED
Checkmark scale 0→1 150ms 0 cubic(.2,.8,.2,1) opacity only
Task row x, opacity 250ms 100ms cubic(.4,0,1,1) opacity only
List items y (gap close) 250ms 150ms cubic(.2,0,0,1) instant
motion-quick, ease-exit) and reference them in every draft prompt — pb-24's documentation pass then covers motion too.Uploading competitor recordings or brand assets as motion references is fine for internal drafting, but generated animations that visibly imitate another product's signature motion invite the same trouble as copied visuals — reference for timing and feel, not for identity.
AI collapses the cost of drafting keyframes, which moves all the value to what it cannot do — choosing what motion means, tuning timing to frequency, honoring reduced-motion users, and shipping the numbers so the build matches the tune.