Meta open-sources LlamaDraw: a sketch-to-vector model for UI components

AI · 4 min read

Meta open-sources LlamaDraw: a sketch-to-vector model for UI components

LlamaDraw is a 6B-parameter model released by Meta AI that translates raster sketches into structured UI components (buttons, inputs, nav bars) and exports editable vectors with CSS-like style descriptors. The model pairs a vision backbone with a small code generator that outputs component skeletons and interaction suggestions in a domain-specific language.

Meta emphasized permissive licensing and provided a training recipe plus a dataset of anonymized sketch-to-component pairs. LlamaDraw is optimized to run on consumer GPUs, enabling local inference for designers worried about IP leakage. Benchmarks indicate accurate component recognition on common mobile and web layouts, though complex custom widgets still require manual cleanup.

Figma and several independent plugin authors already announced support for LlamaDraw, offering preview tooling and one-click export to component libraries. Meta encouraged community contributions to improve niche widget recognition and extend behavior scripting capabilities.