Take One turns the player into a film set. The video waits behind a full-frame clapperboard, and pressing play slams it shut.

The set stays on during the film. Chapters become numbered scenes, and each new scene gets its own small clap. Time runs as a camera timecode, volume reads like a sound meter, and the settings menu is a camera report with your choice circled in red pen.

The controls themselves stay familiar, so nobody has to learn them. Dark shading keeps text and icons readable over bright scenes, buttons are 44 px tap targets, focus is a bold turquoise ring, and menus close with Esc. With reduced motion on, nothing slams or flashes: the slate simply fades.

Demo film: MARÉE, my own concept film for a fictional perfume, made with AI and streamed from Mux.

What happens to a logo the moment you ask an AI to move it? 🌊

I gave ChatGPT my Datum type system, five weights partly filled with water, and asked for a looping GIF where only the tide moves.

First try: it dropped a row, redrew the letters, turned my flat teal into glossy cyan and cycled the whole stack instead of the water.

So I steered it. Only the water line moves, everything else stays locked, and the last frame matches the first. The second try kept all five rows and the type intact, with a seamless 5 second loop built with ffmpeg.

Still to fix: thin seams in the teal at the tide’s peak, and the rows rise in sync instead of rolling like a swell.

Original on the left, loop on the right.

Which of your stills would you animate first, and what are you afraid it would break? 👇