Process

A transbliteration is not a translation and not an obliteration — it is both: a poem carried across four centuries by rewriting it entirely, keeping the argument and losing the skin. Each of Shakespeare’s sonnets is first put through machine translation into Spanish, where it comes back as near-nonsense. The nonsense is then read for sound rather than sense, and whatever it triggers is written down. What survives is not the sonnet’s language. It is what the sonnet was for.

The method turned out to describe how the pictures are made, too. A diffusion model starts from pure noise and, held to a text, makes it slightly less noisy, then does that again, and again, until an image is standing there. It is the same loop: a noisy text, a brain, and enough passes.

Every line of the finished poem becomes a prompt — material, period, place, light — and generates an image of a sculpture in a garden. Each set of images is grown from a single fixed seed, so a style can be judged by how far it bends to the words rather than by luck; it is also why one line often echoes the pose or the spacing of the line before it. The strongest images are lifted off the page into true 3D models.

The garden they stand in is a maquette. The sculptures are meant to be built, and the votes are the beginning of the argument about which ones.

Technical specifications

Medium

Poetry, AI-generated imagery, and 3D digital sculpture

Dimensions

Images 1344×768 px; sculptures as glTF/GLB 3D models; texts in print and web

Tools

Transbliteration

Original poetic rewriting of all 154 Shakespeare sonnets by the artist

FLUX

Line-by-line image generation from structured prompts (material, artists, setting, era), run in Google Colab

TRELLIS

Image-to-3D generation of sculpture models (GLB)

Latent interpolation

Slerp walks between line images, rendered as video

DynamicDiff

Custom Python/Manim tool animating each sonnet morphing into its transbliteration

Site

Hand-built static site (Python/Jinja2, three.js, model-viewer)