AI narrator
Eleven variants of the same shape, each with a different technique underneath. They run live side by side here, because which approach feels most like a talking AI only shows on the same sound.
Without an edge
Two nested domain warps drive both the shape and the colour, so what you get is a graded ribbon instead of flat noise.
Spheres melt together through an associative smooth minimum, rendered as an emissive density that fades out towards the edges.
A raymarch accumulates light through a three-dimensional density field, which gives the softest edges and the most depth of them all.
A real fluid simulation carries the colour along, where we only decide where to inject and the flow makes the shape.
A ring of springs under internal pressure is a physical body, so this blob keeps wobbling and reacts instead of following a path.
Millions of particles drift on a divergence-free flow field, where the grain is the shape rather than a layer on top of it.
Dark slats that catch a lamp travelling across them, turning open and shut as it passes, so the shape is never lit but always read off something else.
With an edge
A real mesh is carved out of a density grid every frame, the only variant where the shape can split and flow back together.
A sphere is pushed outwards per vertex by four-dimensional noise, so the shape has a real normal and therefore a real rim light.
The same distance field as Molten Core, but lit as a lacquered surface, with thin-film interference that shifts the colour of the edge.
A vector path from noise that is correlated along the outline, filled with a gradient, and so the only one with a truly sharp edge.
Digitas Convention 26
The five marks are analytic distance fields rather than textures, so they stay sharp at any size, carry their own edge and glow, and the sixth face runs one straight into the other.
Up close the cube is thousands of loose blocks. On the hit it really falls over, with real gravity and real collisions, and on the phrase it pulls itself back into shape.
The block as a hundred thousand gaussian splats rather than a surface: sharp from across the room, a cloud of soft smears up close, and every splat breathes on its own.
The block turns out to be made of itself. A fractal fold repeats it three scales deep and the five marks are not drawn but are the holes cut out of it, so falling in never arrives anywhere.
Every mark is a hole rather than a shape, and behind each one lies a different room that is larger than the box it sits in. Turn the box and the five views follow five different logics.
The block is hollow and the marks are cut-outs in its wall, so what you see is not the block but its light in real fog, shaped like the marks, sweeping the floor as it turns.
The block is made of particles rather than covered in them. A million of them hold the shape, a warp field deforms the skin, and on the hit they scatter and come back on the phrase.
Five blades shaped like the marks cut into a solid block, so it opens and you see the inside, and one slider takes the edge from knife to wax.
You fall endlessly into the block. A complex logarithm ties one full turn to exactly one zoom level, so the repetition folds into itself without a seam, and the zoom clicks on the bar.
Two line grids sliding over each other, where the mark is never drawn at all: it shifts the phase of the second grid, and the interference fringes are its contour lines.
A hundred thousand instanced marks with the camera dollying through them, wrapping on z in the vertex shader, so the tunnel is endless without endless geometry.
A block of glass with the marks cut hollow into the body rather than printed on it, where dispersion breaks the light into colour along every edge.
Everything in two colours through an ordered dither, with the marks in halftone. Flat, hard, Bauhaus, and the pattern falls per device pixel so it never crawls.
The previous frame returns with a little zoom and rotation, and that same buffer is then read as a height map, so the trails become real relief instead of a smear.
One mark filling the screen that flows into the next on the bar. Only possible because they are distance fields, and the morph function underneath is the one every other scene leans on.
The cube turns out to be the shadow of a four-dimensional one. It rotates in a plane you cannot see, so it turns itself inside out without anything passing through anything, and the marks sit on faces that keep changing depth.
A white square that flashes on the beat, and on the phrase in its second mode. The only honest way to tell whether the music clock is actually in time.
A human figure made of particles held on a skeleton, which walks, dances and falls apart into a cloud on its own clock. The only scene where the shape is a body rather than a field.
Imagineering
Six worlds on the same voice, where the narration puts you inside the image instead of in front of it.

A humanoid built from fourteen tapered bones melted together by a smooth minimum, with the particles projected onto the surface so the figure is a skin and not a cloud.

Particles are pulled onto the glyphs of the word being spoken at that exact moment, using the per-character timeline from forced alignment, so the type is formed before the sound arrives.

Images from the archive are read pixel by pixel on the GPU, where brightness becomes depth, so a photograph turns to grain and re-forms as the next one.

A logarithmic spiral disc generated in closed form and rotated with a differential angular velocity, so the arms wind up over time without a single frame of simulation.

An infinite city derived from a hash of the cell index, so the buildings exist as a function rather than as data and the flight forward costs no memory at all.

A ridged heightfield whose profile is literally the spectrogram of the narration, drawn twice at a half-period offset on an exponential scale loop so the descent never ends.