Effect #016 in our rendered series
Camera-follow captions
The view drifts to follow the active word through a multi-line caption block, so the eye stays on the word being spoken.
Rendered by us at 1920x1080 over a shot from Tears of Steel (Blender Foundation, CC-BY 3.0). Expand plays the full 1080p file with the film's sound.
Edit in Premiere, Resolve or After Effects? Jump to editor templates. Making many versions of the same video? Use the code route.
Make this effect with
Built and tested by us
Other ways, not tested by us
Use it in a real edit
The demo above already plays the composition over real footage: the clip is a muted <video> layer under the graphics (the file plate.mp4 in the code below), and the film's sound is added to the finished file. Replace that file with your own clip and render again. To use the graphics in an editor instead, switch the background off and render a transparent MOV (an H.264 MP4 cannot carry transparency):
npx hyperframes render . -o overlay.mov --format mov --fps 24 --variables '{"plate":false}'The routes in full
With Claude Code: the prompt we used
Rendered and tested by us in HyperFrames 0.8.10 on Nov 7, 2026.
Make a 5 second caption demo where the camera drifts and follows the active word across a three-line caption block, so the eye is always on the spoken word, with a transparent background option.
What changed since this prompt: the demo on this page is our second version, 6 seconds long, rebuilt on a shot from Tears of Steel with the same animation idea. The full composition below is the one we rendered; its build log lists every change.
Built by a Claude Code subagent (Sonnet) from the HyperFrames registry block caption-camera-follow, HyperFrames 0.8.10. Claude Code then checked the composition with hyperframes lint and hyperframes check before we rendered it.
What we had to correct (4 notes from the build log):
- Lint run 1 (2026-11-05): 0 errors, 0 warnings.
- What changed from the registry block: Registry block measures nothing but uses a bespoke camera; rebuilt with a monospace face so word positions are computed from character counts, not measured, so the camera targets are right even before the font loads. Camera follow strength and zoom are capped so no word leaves the 6% margin.
- Decided against: using Math.random, Date, CDN GSAP, and any live measuring of text for layout.
- v2 (7 Nov 2026): rebuilt on Tears of Steel footage (plate dlg_space, 6.0 s, the film's pink on-screen-text cutaway cut out) with the film's sound. The caption block moved from mid-frame to the bottom (one line per page, bottom of the line 105 px above the bottom edge; two lines covered Thom's mouth in the close-up) and now works in three pages; the camera pushes onto each word as it is spoken in 'You have your robotics... and I just want to be awesome in space.' Text 72 px (76 to 94 px at the camera zoom of 1.05 to 1.3); the transform origin sits on the caption block so the zoom never pushes text off the frame.
With HyperFrames code
Rendered and tested by us in HyperFrames 0.8.10 on Nov 7, 2026.
npx hyperframes add caption-camera-follow
npx hyperframes render . -o out.mp4 --fps 24Settings you can change without rebuilding (pass them with --variables):
| Setting | Name | Type | Default |
|---|---|---|---|
| Active word colour | accent | color | #ffd84d |
| Text size (px) | size | number | 72 |
| Camera zoom | zoom | number | 1.2 |
| Show background | plate | boolean | true |
Our full composition (index.html, 7 KB)
<!doctype html>
<html lang="en" data-composition-variables='[{"id":"accent","type":"color","label":"Active word colour","default":"#ffd84d"},{"id":"size","type":"number","label":"Text size (px)","default":72,"min":72,"max":76},{"id":"zoom","type":"number","label":"Camera zoom","default":1.2,"min":1.05,"max":1.3,"step":0.05},{"id":"plate","type":"boolean","label":"Show background","default":true}]'>
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=1920, height=1080" />
<link href="https://fonts.googleapis.com/css2?family=JetBrains+Mono:wght@800&display=block" rel="stylesheet" />
<script src="gsap.min.js"></script>
<style>
*, *::before, *::after { box-sizing: border-box; margin: 0; padding: 0; }
html, body { width: 1920px; height: 1080px; overflow: hidden; background: transparent; }
#root { position: relative; width: 100%; height: 100%; overflow: hidden; background: transparent; }
#plate { position: absolute; inset: 0; background: radial-gradient(120% 90% at 30% 40%, #1b2335 0%, #0b0d12 70%); overflow: hidden; }
#world { position: absolute; left: -200px; top: -120px; width: 2400px; height: 1320px; }
.blob { position: absolute; border-radius: 50%; filter: blur(60px); }
.b1 { width: 760px; height: 760px; left: 180px; top: 260px; background: #7b4dff; opacity: 0.35; }
.b2 { width: 900px; height: 900px; left: 1200px; top: 80px; background: #0fa3b1; opacity: 0.45; }
.grid { position: absolute; inset: 0; background-image: linear-gradient(rgba(255,255,255,0.05) 1px, transparent 1px), linear-gradient(90deg, rgba(255,255,255,0.05) 1px, transparent 1px); background-size: 120px 120px; }
#drift { position: absolute; inset: 0; }
#cam { position: absolute; inset: 0; transform-origin: 50% 50%; }
.line { position: absolute; left: 0; width: 100%; text-align: center; white-space: pre; font-family: "JetBrains Mono", monospace; font-weight: 800; letter-spacing: 0; }
.w { display: inline-block; color: #fff; opacity: 0.28; text-shadow: 0 6px 22px rgba(0,0,0,0.45); }
/* footage plate (v2): real footage, pinned, no generated blobs */
#plate { background: #000 !important; }
#world { left: 0 !important; top: 0 !important; width: 1920px !important; height: 1080px !important; transform: none !important; }
#world .blob, #world .grid { display: none !important; }
#pv { position: absolute; left: 0; top: 0; width: 1920px; height: 1080px; object-fit: cover; }
#scrim { position: absolute; inset: 0; background: linear-gradient(180deg, rgba(0,0,0,0) 45%, rgba(0,0,0,0.55) 100%); }
</style>
</head>
<body>
<div id="root" data-composition-id="main" data-start="0" data-duration="6.0" data-fps="24" data-width="1920" data-height="1080">
<div id="plate"><div id="world"><video id="pv" src="plate.mp4" muted playsinline data-start="0" data-duration="6.0"></video><div id="scrim"></div><div class="blob b1" id="b1"></div><div class="blob b2" id="b2"></div><div class="grid"></div></div></div>
<div id="drift"><div id="cam"></div></div>
</div>
<script>
window.__timelines = window.__timelines || {};
(function () {
const v = Object.assign({}, {"accent": "#ffd84d", "size": 72, "zoom": 1.2, "plate": true}, window.__hyperframes.getVariables());
const tl = gsap.timeline({ paused: true });
if (!v.plate) document.getElementById("plate").style.display = "none";
tl.fromTo("#world", { x: 0, y: 0 }, { x: -160, y: -50, duration: 5, ease: "none" }, 0);
tl.fromTo("#b1", { x: 0, y: 0 }, { x: 140, y: -80, duration: 5, ease: "sine.inOut" }, 0);
tl.fromTo("#b2", { x: 0, y: 0 }, { x: -120, y: 100, duration: 5, ease: "sine.inOut" }, 0);
const SIZE = v.size, CW = SIZE * 0.6, LH = Math.round(SIZE * 1.45), Z = v.zoom, FOLLOW = 0.45, BASE = 975, ORY = 900;
const PAGES = [{"lines": [[["You", 0.24], ["have", 0.8], ["your", 0.98], ["robotics...", 1.22]]], "end": 2.32}, {"lines": [[["and", 2.49], ["I", 2.71], ["just", 3.43], ["want", 3.61]]], "end": 3.8}, {"lines": [[["to", 3.83], ["be", 3.93], ["awesome", 4.09], ["in", 4.45], ["space.", 4.63]]], "end": 5.85}];
gsap.set("#cam", { x: 0, y: 0, scaleX: 1, scaleY: 1, transformOrigin: "960px " + ORY + "px" });
const cam = document.getElementById("cam");
PAGES.forEach((pg, pi) => {
const page = document.createElement("div"); page.id = "p" + pi;
page.style.cssText = "position:absolute;inset:0;opacity:0;visibility:hidden";
const n = pg.lines.length;
const flat = [];
pg.lines.forEach((ln, li) => {
const el = document.createElement("div"); el.className = "line";
el.style.fontSize = SIZE + "px"; el.style.height = LH + "px"; el.style.lineHeight = LH + "px";
const top = BASE - (n - li) * LH;
el.style.top = top + "px";
const total = ln.reduce((s, w) => s + w[0].length, 0) + (ln.length - 1);
const left = 960 - (total * CW) / 2;
let idx = 0;
ln.forEach(([text, ts], wi) => {
const w = document.createElement("span"); w.className = "w"; w.id = "w" + pi + "_" + li + "_" + wi; w.textContent = text;
el.appendChild(w);
if (wi < ln.length - 1) el.appendChild(document.createTextNode(" "));
const cx = left + (idx + text.length / 2) * CW;
flat.push({ id: "#" + w.id, ts, dx: cx - 960, dy: top + LH / 2 - ORY });
idx += text.length + 1;
});
page.appendChild(el);
});
cam.appendChild(page);
const first = flat[0].ts;
tl.set(page, { visibility: "visible" }, first - 0.12);
tl.fromTo(page, { opacity: 0 }, { opacity: 1, duration: 0.12, ease: "power1.out" }, first - 0.12);
flat.forEach((f, k) => {
const next = flat[k + 1];
const gap = next ? next.ts - f.ts : 0.6;
const d = Math.min(0.34, gap * 0.9);
tl.to("#cam", { x: -f.dx * Z * FOLLOW, y: -f.dy * Z * FOLLOW, scaleX: Z, scaleY: Z, duration: d, ease: "power2.inOut" }, f.ts);
tl.to(f.id, { opacity: 1, color: v.accent, scaleX: 1.07, scaleY: 1.07, duration: 0.14, ease: "power2.out" }, f.ts);
const done = next ? Math.min(next.ts, f.ts + 0.45) : f.ts + 0.45;
tl.to(f.id, { opacity: 0.95, color: "#ffffff", scaleX: 1, scaleY: 1, duration: 0.2, ease: "power1.out" }, done);
});
tl.to("#cam", { x: 0, y: 0, scaleX: 1, scaleY: 1, duration: 0.3, ease: "power2.inOut" }, pg.end - 0.3);
tl.to(page, { opacity: 0, duration: 0.15, ease: "power1.in" }, pg.end - 0.15);
tl.set(page, { visibility: "hidden" }, pg.end);
});
// slow constant drift so the camera never feels locked off
tl.fromTo("#drift", { x: -16, y: 6 }, { x: 16, y: -6, duration: 3.0, ease: "sine.inOut" }, 0);
tl.to("#drift", { x: -16, y: 6, duration: 3.0, ease: "sine.inOut" }, 3.0);
window.__timelines["main"] = tl;
})();
</script>
</body>
</html>
With Remotion
We have not built this effect in Remotion. If you use it, read its licence first: Individuals (personal or commercial use), organisations or teams with a total headcount of 3 or less, non-profits, and anyone evaluating Remotion before commercial use. A company does not qualify merely because it is unprofitable. Everyone not eligible for the Free License needs a Company License. Remotion licence FAQ, checked Nov 4, 2026.
In your editor: Premiere, Resolve, After Effects, CapCut
We have not reviewed an editor template for this effect yet.
Variations
The same composition rendered again with different settings. Nothing was rebuilt.
{"accent":"#9fd0ff","size":72,"zoom":1.05}{"accent":"#ff4fa0","size":72,"zoom":1.3}{"size":76,"zoom":1.15}Spec and test details
- Engine
- HyperFrames 0.8.10, GSAP 3.14.2 (local copy)
- Length and frame rate
- 6 seconds at 24 fps, rendered at 1920x1080
- Video files on this page
- H.264: 1920x1080 for Expand (5704 KB), 1280x720 in the page (1553.9 KB), 640x360 in galleries (277.4 KB); the 1080p and 720p files carry the film's sound
- Picture check
- The 1080p file scores VMAF 97.78 against our render master and keeps 98% of its edge detail (measured by us with FFmpeg libvmaf and OpenCV)
- Footage
- Celia close-up, then Thom close-up (the film's pink text cutaway removed with a matched 1.87 s cut in picture and sound); 'You have your robotics... and I just want to be awesome in space.' (Tears of Steel, 5:20)
- Tested in editors
- None yet
- Flash check
- Passed: at most 1 general and 0 red flashes in any second. automated approximation of WCAG 2.3.1 on a 96x54 grid, whole loop including the restart; not a certification.
- Licence
- Composition adapted by us from the HyperFrames registry (Apache-2.0); text fonts from Google Fonts under the SIL Open Font License.
- Last tested
- Nov 7, 2026
- Machine
- Laptop: AMD Ryzen 7 7735U (8 cores, 16 threads), 31 GB RAM, Windows 11 Home; CPU rendering
Technical details: render times and file sizes
- Render of the 6-second master: 67.1 s (2 render workers, on the laptop above).
- full.mp4: 5704 KB
- hero.mp4: 1553.9 KB
- hero.webp: 33.2 KB
- tile.mp4: 277.4 KB
- tile.webp: 11.1 KB
- Variation "subtle": rendered in 83.3 s, 267.1 KB
- Variation "bold": rendered in 41.4 s, 283.8 KB
- Variation "large": rendered in 49.4 s, 280.9 KB
Credits
- Built from the HyperFrames registry component caption-camera-follow by HeyGen (Apache-2.0), adapted by us.
- Fonts from Google Fonts, under the SIL Open Font License.
- Footage: Tears of Steel, (c) Blender Foundation, mango.blender.org, licensed CC-BY 3.0. We cropped the shots to 16:9 and laid our graphics over them.
- Editor templates and other routes on this page belong to their creators, who are credited on each card.
Related effects and jobs
Page checked on Nov 4, 2026.


