Text is a normal object — it can react to the fluid, be hidden but still reactive (fluid reveals the hidden word), or float on top non-reactive.

Select an object to change how it meets the fluid.

Slip is global — how freely fluid slides along every edge.

A4 stroke paths in millimetres — pen plotter and riso separation ready. Nothing rasterised.

Sequences render offline from the seed, faster than real time — an export is a regeneration, not a recording.

SVG shapes import as vector objects. Raster images are thresholded into a silhouette stencil so the fluid can actually collide with them — adjust threshold for busy photos.

Link must point directly at an audio file with CORS enabled — not a YouTube/Spotify/SoundCloud page. Browsers can't decode audio out of those pages client-side.

Bass drives splat strength and beat hits; treble thins the fluid so it feels brighter on hi-hats. A hidden reactive object (if you've placed one) flashes into view on strong beats. Auto-render captures live, since audio timing isn't seed-reproducible — plain seed exports still are.

The track is decoded once and its beat grid extracted as data, so the render is driven by the music but stays reproducible.

Silent renders offline from the beat data — reproducible, faster than realtime, no sound. With audio records in real time with the track muxed in, which is what you hand a client to post. Poster frame gives you a matching still for static placements.