Fontaine, at night
Three stops in one take: the aquabus terminal, the arcade, the opera roof.
Describe a place in one line and scroll moves a real camera through it: generated scenes, a dive into each, fly-outs you never see. No Three.js, no GSAP.




Powered by the Scroll World Video skill.









Three stops in one take: the aquabus terminal, the arcade, the opera roof.
Neta Studio generates each scene, holds one grade across all of them, and flies the camera in and out between them. The page you publish only scrubs that footage, so scroll position is the camera clock — and the words stay editable in the same place the page lives.
It is pre-rendered. Each scene is a generated still, the camera move is a generated clip, and scroll position scrubs that clip. You get real depth and real motion at a few megabytes per scene — but you cannot orbit the model or walk off the path.
No. You write one sentence. No library, no timeline, no scene graph, and no shoot budget either.
Three scenes, roughly 23–25 MB, behind a progress gate before the page opens. Each extra scene adds about 9 MB to that wait.
Yes. Pass them in as references and each scene is derived from them, so a real place keeps its look and light. For a world that does not exist yet, a sentence is enough.
Each scene’s headline, one line of body and up to three tags, plus the scene order. The camera work stays as rendered.
The page opens and scrolls with lighter encodes and a centre crop. Scrubbing is a desktop gesture; on a slow phone the first load is what you feel.
A template animates elements you already have. Here the scene, the depth and the camera path are generated, so the page can fly through a place that never existed.
It publishes as a work with its own public link. Send it, embed it, or fork it and swap the scenes.
Three scenes, a camera that moves through them, and a link you can send.