Meta unveils LlamaX-3 with multimodal prompts tailored for product designers

AI · 6 min read

Meta's LlamaX-3 focuses on bridging early-stage ideation and high-fidelity mockups by accepting sketches, screenshots, and natural-language instructions in the same prompt. The model improves on previous Llama variants with a new attention scheme that preserves layout context across longer conversations, reducing the need to re-upload assets during a single design sprint.

The release includes a set of pre-trained adapters specifically for UI tasks: icon harmonization, typography pairing, and responsive-layout suggestion. Meta also provided a reference Figma plugin and a browser demo that lets designers convert hand-drawn flows into editable vector frames with suggested accessibility annotations.

Privacy controls were a headline: LlamaX-3 supports ephemeral session keys and an on-premises variant for enterprise design teams that need to avoid cloud inference. Meta said the model will be available under a commercial license with tiered pricing that favors small studios and research labs.