Colophon · showcase
How the ascent was built
Firn is one idea: the whole site is a single, unbroken climb, and scrolling is the train.
The film
The ascent is a cinematic AI film generated with Higgsfield (Kling 3.0 Turbo): one continuous ten-second vertical rise through cold alpine fog that thins and warms until it breaks above a sea of cloud onto a sunlit summit. Two takes were generated and the stronger climb was curated on its mid-frames. This is disclosed as AI-generated under the EU AI Act, and no real place is shown.
Scroll becomes the train
The film is extracted to a sequence of WebP frames (100 for desktop, a lighter 80-frame set for phones) and drawn to a canvas. Instead of mapping scroll linearly, a dwell engine builds an inverse cumulative curve with a Gaussian bump at each of the five stations, so the climb slows almost to a stop while you read a station, then accelerates between them. Frames load progressively, with a nearest-frame fallback so there is never a white flash.
Readable over a moving picture
Each station's text sits in its own soft dark scrim, so contrast stays AA no matter how bright the cloud behind it becomes. The stations are real document blocks flowing over the pinned film, which means the page is fully readable even when the film is not running.
Light and accessible
The whole frame set is barely over a megabyte. Text paints first (the summit is a tiny poster image behind the canvas); the canvas takes over once frames arrive and pauses when the tab is hidden. With prefers-reduced-motion the film is skipped entirely and you get the summit still with every word in place.
Tools
- Astro + Tailwind on Cloudflare Pages.
- Higgsfield / Kling 3.0 for the film; ffmpeg for the WebP frame extraction.
- Hand-written canvas dwell-scrub engine; Archivo variable (condensed) throughout.