Motion & 3D · Intermediate
A blank browser tab types <div>, and over eight bars the page builds itself in blueprint lines. Every piece is labelled with the real code that makes it: the grid, the orb's grid-column, the headline's font, a cubic-bezier the hills rise on, a palette that fills one plate per beat, and a GSAP tween the sun rises with. The finished poster plays, then a scrubber rewinds it all to the blank tab, and that's the loop. The whole scene is drawn from one number, the build progress, which is why the loop is exact. You get the exact prompts, a check for every step, and the complete source.
The recipe
Every step gives you the action, the exact input (a prompt, settings or a file) and a check to compare your result against the original.
Claude Code: Claude Opus 5.5 (claude-opus-5-5). Wrote every file. Uses your Claude plan. HyperFrames: 0.8.74 (npx hyperframes@0.8.74). Preview, check, snapshot, render (HTML → MP4). Free, runs locally. GSAP: 3.14.2 (jsDelivr CDN). The one paused timeline and its eases. Free. Node.js: 24.x. Runs the HyperFrames CLI. Free. FFmpeg: any recent build. The Instagram encode and the loop check. Free. Fonts: League Gothic, JetBrains Mono. The poster headline, and everything that is code. Free (OFL). Codex or any agent that can run shell commands works too. HyperFrames downloads its own headless Chrome on first run, and on the first check or render it fetches the two fonts from Google Fonts and caches them. Provenance: Recorded September 24, 2026: Claude Code with Claude Opus 5.5 (claude-opus-5-5); HyperFrames 0.8.74 (headless Chrome 152.0.7977.30, beginframe capture, software GPU); GSAP 3.14.2; Node.js 24.13.1. The build session ran past midnight, and the final render was made at 00:10 on September 25. Master rendered locally at 1080×1920, 30 fps, 24.0 s, H.264 High (HyperFrames "looks" quality, CRF 16). Instagram copy: libx264 preset slow, CRF 23, maxrate 16M, yuv420p, faststart, no audio. Loop check: SSIM 0.9994 (master) and 0.9987 (Instagram) between the first and last frames after a sigma-4 blur, and pixel-identical before encoding. No external images, footage or generative-media models were used.