a model reads your words · lays out a scene · SVG + canvas draw it
Type a description. A model turns it into a scene of shapes — positions, sizes, colors — and both an SVG layer and a canvas layer draw the picture. Not diffusion (no photo of a dragon) — procedural: the words become primitives the model arranges into an image.
◆ SVG layer
◆ Canvas layer
type a scene and the model will draw it
known words (local): sun · moon · house · tree · cloud · star · hill · water · mountain · bird · flower
honest note: this is not diffusion text-to-image — a standalone artifact can't call an image model. This is text → structured scene → drawn picture: the model returns a list of primitives (circle, rect, triangle, star…) with positions and colors, and the artifact draws them. In the Claude.ai artifact runtime the live API composes scenes from free text; standalone, a local keyword library handles the known words. Same primitive schema, same renderer, both layers.
TEXT → A MODEL RETURNS A SCENE OF PRIMITIVES → SVG + CANVAS DRAW THE PICTURE
PROCEDURAL, NOT DIFFUSION · LIVE API IN THE RUNTIME · LOCAL KEYWORD LIBRARY STANDALONE
TEXT TO PICTURE · A PURPLE PAPER · SERIES E · JUNE 2026