Meta opens Llama 3o series with fine-grained layout control for designers

AI · 4 min read

Meta's Llama 3o family is designed to help designers generate UI mockups with exact layout control and composable component constraints. The models accept layout grids, token maps, and component manifests as conditioning inputs, allowing teams to request specific placements and behavior rather than freeform visuals.

A notable addition is the 'layout mask' primitive, which forces generated elements to conform to a supplied grid and preserves element semantics (buttons remain buttons, text blocks stay editable). Llama 3o also outputs metadata useful for downstream conversion to code, including named layers and suggested ARIA roles.

Meta is offering hosted inference and a lightweight on-prem runtime for enterprise customers. The company claims the Llama 3o models were trained with a heavy emphasis on reproducibility and rule-following to reduce unexpected variation in outputs.