Strike a Chord! Modal Kinetic Typography

Strike a Chord! Modal Kinetic Typography. By Maham Tanveer and three others, from Simon Fraser University and Adobe. Published on the twenty-ninth of September. No code repository is linked to it.

Kinetic typography animates a letter so that it expresses a concept, while the letter stays readable. The authors point to two weaknesses in earlier ways of doing this with a vector glyph.

One approach optimizes the outline's points freely, under video score distillation, known as SDS. Every point moves on its own, and so does every frame, each following noisy gradients with nothing tying it to its neighbours. The result is an outline that tears and motion that jitters, which is hard on a letter that is supposed to stay legible.

The other approach is more structured. It relies on skeletons or keypoints, and those come from priors specific to a category of shape. The structure has to be brought in from outside the glyph, and it is only as available as the category prior that supplies it.

Modal kinetic typography builds motion from the glyph's own natural vibration modes, the ways the shape would most naturally flex if it were struck. Motion is then expressed in terms of those modes rather than in terms of individual points. That one choice answers both weaknesses together. Modes are smooth, so the outline does not tear. And modes come from the glyph itself, so no skeleton or keypoint prior is needed.

A finite-element eigenproblem is assembled from the vector outline. Solving it yields the glyph's softest modes, both for the whole letter and for each of its parts, and these let it bend. The same problem has zero-energy solutions, which are rigid translations and rotations, and these are applied in closed form to each part, so parts can also move as blocks.

To animate, a frozen video diffusion model supervises only two things: the amplitude and the phase of each mode. Because the modes are smooth along the outline, and are driven by a few whole-cycle harmonics, the noisy gradients are confined to smooth motion that loops seamlessly. The only outside prior is a list naming each letter's moving parts, generated once for the whole alphabet by a language model.

Shape and motion are separated by construction. A single base outline is sculpted toward the concept, and the modal drive cannot alter that outline. A letter can therefore be animated without being reshaped at all, leaving motion by itself to express the idea.

Compared with Dynamic Typography and AniClipart, the method produces more articulated and smoother motion, at comparable or better alignment with the concept. It tears the glyph less than Dynamic Typography does. Human raters prefer it, including in a frozen-shape setting, where the outline is held fixed and motion alone has to carry the concept. Raters also preferred its results over Astra, listed as GPT-6.

The title's pun is also the method. A struck chord rings in the ways the instrument already allows, and these letters move in the ways their own outlines already allow. Nothing about the motion is borrowed from a skeleton or a keypoint template: the letter supplies it, a language model names the moving parts, and a video model only decides how loudly and when each mode plays.