Meta extends Llama Visual for UI prototyping with live component outputs

AI · 5 min read

The extended Llama Visual includes prompts and adapters specifically built for interpreting design screens and producing framework-ready components. Designers can upload screenshots and get componentized results in React, SwiftUI, or Jetpack Compose, with prop suggestions and state outlines.

Meta highlighted internal experiments where the model cut implementation time for simple screens by nearly half. The release also includes a moderation layer and heuristics to avoid generating copyrighted content when designers provide third-party imagery.

Meta will make the model available to partners and has released an SDK for companies wanting to host fine-tuned versions on private infrastructure. The company notes that the model is optimized for collaboration with engineers and is not a substitute for production-quality engineering review.