12 KiB
name, description, metadata
| name | description | metadata | ||
|---|---|---|---|---|
| camera-cursor-tracking | Two-phase virtual camera that locks viewport to a moving focal point with configurable initial positioning. |
|
Two-Phase Camera Cursor Tracking
Keeps a horizontally-growing element (e.g. a search bar with typing text, a long URL animating in) visible by switching between two camera modes.
How It Works
Separate World Space (the full target element with all content) from Screen Space (the viewport). Two phases:
- Phase 1 (Static) — The world container sits at a fixed initial offset. Camera doesn't move. This anchors the viewer's eye to the composition before tracking begins.
- Phase 2 (Tracking) — Activates when the focal point (cursor, highlight, last typed glyph) exceeds a target screen position (e.g. a configurable fraction
CURSOR_TARGET_FRACTIONof viewport width from the left). The world container translates leftward (x: -<delta>) keeping the focal point pinned at that screen position.
The offset math is mathematically continuous at the phase boundary — at the instant tracking starts, the world position equals what the static phase had. So the transition is seamless.
The piecewise form used in code is:
finalWorldX = Math.min(INITIAL_OFFSET, trackingOffset)
INITIAL_OFFSET is the static-phase value; trackingOffset is whatever shift keeps the focal point at CURSOR_TARGET_FRACTION × viewportWidth. While the focal point hasn't grown past the target screen X, trackingOffset exceeds INITIAL_OFFSET (it's a less-negative number) and Math.min returns the static value. Once the focal point would cross the target, trackingOffset overtakes INITIAL_OFFSET and tracking takes over.
HTML
<div
class="scene"
id="tracking-scene"
data-composition-id="tracking-scene"
data-start="0"
data-duration="5"
data-track-index="0"
>
<div class="viewport">
<div class="world">
<div class="search-bar">
<span class="text" id="reveal-text">{phrase}</span><span class="cursor">|</span>
</div>
</div>
</div>
</div>
CSS (hero-frame layout)
.scene {
position: relative;
width: 100%;
height: 100%;
}
.viewport {
position: absolute;
inset: 0;
overflow: hidden; /* clip the world content */
display: flex;
align-items: center;
justify-content: flex-start;
padding-left: VIEWPORT_PAD_LEFT; /* "left margin" — variation: left-aligned init */
}
.world {
display: flex;
align-items: center;
white-space: nowrap; /* keep text on one line for camera-tracking */
transform: translateX(0); /* GSAP will animate this */
}
.search-bar {
font-family: {font};
font-size: BAR_FONT_SIZE;
font-weight: BAR_FONT_WEIGHT;
color: {textColor};
letter-spacing: BAR_LETTER_SPACING;
}
.search-bar .text {
/* Width grows as more characters reveal */
display: inline-block;
overflow: hidden;
vertical-align: bottom;
}
.search-bar .cursor {
display: inline-block;
width: CURSOR_WIDTH;
margin-left: CURSOR_GAP;
background: {accentColor};
height: CURSOR_HEIGHT_EM;
vertical-align: bottom;
/* No `animation: blink` CSS keyframe here — HF renders by seeking a paused
timeline, and CSS animation clocks are NOT synced to that seek. A CSS
blink will flicker non-deterministically. Drive cursor blink as a finite
yoyo tween on the GSAP timeline instead — see GSAP Timeline section. */
}
GSAP Timeline
<script src="https://cdn.jsdelivr.net/npm/gsap@3.14.2/dist/gsap.min.js"></script>
<script>
window.__timelines = window.__timelines || {};
const tl = gsap.timeline({ paused: true });
// Pre-measure target text width to compute tracking distance.
// Build the timeline SYNCHRONOUSLY — see Critical Constraints for why
// a fonts.ready gate causes worker-race flicker.
const textEl = document.getElementById("reveal-text");
const fullText = textEl.textContent;
const targetCursorScreenX = CURSOR_TARGET_FRACTION * VIEWPORT_WIDTH;
const fullWidth = textEl.scrollWidth; // total text width after full reveal
const trackingDelta = Math.max(0, VIEWPORT_PAD_LEFT + fullWidth - targetCursorScreenX);
// Phase 1 — text reveals progressively; camera holds.
// Reveal via max-width tween (no layout-property tweens on width/left/top).
tl.fromTo(
".search-bar .text",
{ maxWidth: 0 },
{
maxWidth: fullWidth,
duration: REVEAL_DUR,
ease: "none", // linear typing rate
},
REVEAL_START,
);
// Phase 2 — camera tracks. Begin BEFORE full reveal so the boundary feels
// continuous (text is still revealing as camera starts moving). The
// Math.min(INITIAL_OFFSET, trackingOffset) formulation makes the handoff
// mathematically continuous; see How It Works.
tl.to(
".world",
{
x: -trackingDelta,
duration: TRACK_DUR,
ease: "power2.inOut",
},
TRACK_START,
);
// Cursor blink — GSAP-driven (NEVER CSS @keyframes infinite, which doesn't
// sync with HF's seek-by-frame). Finite yoyo, repeats computed from scene
// length so blinks land deterministically across frames.
const blinkRepeats = Math.ceil(SCENE_DURATION / BLINK_HALF_PERIOD) - 1;
tl.to(
".search-bar .cursor",
{
opacity: 0,
duration: BLINK_HALF_PERIOD,
ease: "steps(1)", // hard on/off, no fade
yoyo: true,
repeat: blinkRepeats,
},
0,
);
window.__timelines["tracking-scene"] = tl;
</script>
Variations
- Centered → Center-Tracked: set
.viewport { justify-content: center; padding: 0; }. Camera tracks once the focal point crosses the midline (CURSOR_TARGET_FRACTION = 0.5). - Left-Aligned → Right-Tracked: as written above. Best when content exceeds viewport width from the start.
- Continuous typing driver: replace the
maxWidthtween with anonUpdatetyping clock (charsTyped = Math.floor(progress)) plus per-framemeasureNodeWidthto drive the cursor screen X. Required when the typed text is consumed elsewhere in the scene (e.g. by a parent strip's camera offset).
How to Choose Values
- VIEWPORT_PAD_LEFT — left-edge padding of the world inside the viewport (Phase 1 anchor X).
- Range: 0 → ~10% of viewport width
- Effects: 0 hugs the left edge; larger inset feels more like a centered hero frame
- Constraints: must match the CSS
padding-lefton.viewportor the camera math drifts
- VIEWPORT_WIDTH — the composition's
data-widthin CSS pixels.- Constraints: must equal the scene root's
data-width; never tweened
- Constraints: must equal the scene root's
- CURSOR_TARGET_FRACTION — fraction of viewport width where the focal point locks during Phase 2.
- Range: 0.5 (center-tracked) → 0.75 (right-leaning, more text visible behind cursor)
- Effects: lower values leave less revealed text in frame; higher values delay tracking
- BAR_FONT_SIZE — hero element font size.
- Range: ~8-12% of viewport height; below ~6% reads as a UI widget rather than a cinematic element
- BAR_FONT_WEIGHT — weight of the search-bar text.
- Range: discrete; 400 for neutral demo text, 700 for hero / headline framing
- BAR_LETTER_SPACING —
letter-spacingfor the bar text.- Range: slight negative (tighter, more cinematic) → 0 (default)
- CURSOR_WIDTH — visual cursor stroke width in CSS pixels.
- Range: ~4-10 px at 1080p; thinner reads as typed text, thicker reads as a block caret
- CURSOR_GAP —
margin-leftbetween text and cursor.- Range: a few px of breathing room; do not exceed the cursor width or it visually detaches
- CURSOR_HEIGHT_EM — cursor height as an em multiple of the font.
- Range: 0.85-1.0; matches the visual height of the typed glyphs
- REVEAL_START — when Phase 1 typing begins.
- Constraints: typically 0; if preceded by another phase, ≥ that phase's end + small buffer
- REVEAL_DUR — duration of the Phase 1 reveal tween.
- Range: scale with character count (target an average per-character cadence in the 0.05-0.15s range)
- Constraints: must end before
SCENE_DURATIONand ideally overlap slightly with the tracking phase
- TRACK_START — when Phase 2 camera motion begins.
- Range: usually before reveal completes so the handoff feels continuous; can equal
REVEAL_STARTif the focal point is already past the target at t=0 - Constraints:
TRACK_START < REVEAL_START + REVEAL_DURfor a smooth crossfade
- Range: usually before reveal completes so the handoff feels continuous; can equal
- TRACK_DUR — duration of the camera pan.
- Range: 0.8-2.0s; under 0.5s reads as a snap, over 2.5s drags
- SCENE_DURATION — must match the scene root's
data-duration.- Constraints: feeds the blink repeat count; mismatch causes blinks to truncate or run past the end
- BLINK_HALF_PERIOD — half-period of the cursor blink (one on-state OR one off-state).
- Range: 0.2-0.4s; 0.3s reads as a natural caret blink
- Constraints: derived value
Math.ceil(SCENE_DURATION / BLINK_HALF_PERIOD) - 1must be ≥ 0
- Ease choices — discrete:
- Camera pan:
power2.inOutorpower3.inOutfor cinematic settle; avoidback.out(overshoot reads as UI bounce, not camera) - Reveal:
"none"for linear typing; any easing distorts the per-keystroke cadence - Blink:
"steps(1)"for hard on/off; any easing fades the cursor and breaks the caret feel
- Camera pan:
Key Principles
- Measure with
getBoundingClientRect()/ probe nodes, not by character count × font-size. Proportional fonts have variable glyph widths. white-space: nowrapon the world — text must stay on one line for camera math to work- Pre-allocate the world width by setting
maxWidthat full target width — prevents layout shift mid-tween - Eased camera (
power2.inOut/power3.inOut), not linear — natural pan feel - Spring-like via easing, not via stiffness/damping params — GSAP doesn't have a built-in spring, but
back.out(${BOUNCE_FACTOR})orpower4.outapproximate the settling feel
Critical Constraints
- Build the timeline SYNCHRONOUSLY, no fonts.ready gate — HF renders frames in parallel workers, each a fresh browser. If you wrap the timeline build in
document.fonts.ready.then(...), some workers will seek frames BEFORE the Promise resolves and find no timeline registered → those frames render at CSS initial state (e.g.max-width: 0⇒ empty text), other workers render correctly → visible flicker between empty and filled. Registerwindow.__timelines[id] = tlat script-parse time, even if fonts haven't loaded yet — the camera math can tolerate a few percent width error from fallback-font measurement, but worker-race flicker is unacceptable. - If precise post-font measurement matters, re-measure inside the tween's
onUpdate(still deterministic per-frame seek), not via a Promise gate. Or setfont-display: blockon the @font-face to force the browser to wait for the font before painting any text. - Timeline must be paused:
gsap.timeline({ paused: true }). Nevertl.play() - Registry key =
data-composition-id:window.__timelines["tracking-scene"]must match scene root - Continuous math at phase boundary: the world's
xat the moment tracking starts must equal the static-phase offset. TheMath.min(INITIAL_OFFSET, trackingOffset)formulation guarantees this; do NOT switch to a hardif (typingProgress > threshold)branch or the camera will visibly jump. - Inline cursor, not absolutely positioned: cursor should be a sibling of the text (inline-block) so it follows text flow naturally — absolute positioning misaligns with the camera math
overflow: hiddenon.viewport: clip the world's left edge as it pans off-screen- Cursor blink via GSAP, NOT CSS
@keyframes ... infinite— HF renders by seeking the paused timeline; CSS animation clocks are NOT synchronized with that seek, so any CSS-driven blink will flicker non-deterministically across frames. Always drive blink as a finite yoyo tween on the paused GSAP timeline (repeat count computed from scene length).
Combinations
- context-sensitive-cursor.md — change cursor color/style per text segment during typing
- discrete-text-sequence.md — non-linear text reveals that pair with this camera
Pairs with HF skills
/hyperframes-animation— timeline + tween API/hyperframes-core— composition wiring +data-*attributes/hyperframes-cli—hyperframes lintto validate the registry key + duration