Stable Diffusion maker launches StableCanvas, a vector-native model for scalable assets

AI · 5 min read

StableCanvas shifts from raster-first generation to a vector-native approach; it directly generates Bézier paths, groups and named layers, aiming to give designers output that's ready for production scaling and editing. The model supports style controls and can emit multiple optimization levels for print or screen.

Stability AI added exporter options to keep semantic layer names, preserve color tokens, and generate accompanying CSS variable snippets for web implementations. There's also a 'componentize' mode that packages repeated motifs into reusable symbols.

Open-source versions of the model and tooling were announced, accompanied by community-driven datasets for vector evaluation. While the output reduces manual tracing work, experienced vector artists still find edge cases—complex blends and certain type treatments—need manual polish.