Anthropic launches Claude Nova: a vision-and-code multimodal model tuned for designers

AI · 5 min read

Anthropic’s Claude Nova represents a step beyond simple image-captioning models by combining high-resolution image understanding with deterministic code output for design platforms. Nova is tuned on paired UI screenshots, design tokens, and front-end repositories so it can suggest pixel-aware edits and produce production-ready React or Flutter components.

The company emphasized safe UX workflows, adding layers of style consistency and licensing checks for third-party assets. Nova’s API supports prompts like "make this hero section mobile-friendly and output CSS variables," and outputs both an edited image preview and code scaffolding.

Design teams who tried the beta said Nova dramatically reduced handoff friction between visual designers and engineers, but some flagged concerns about hallucinated CSS class names and edge-case accessibility suggestions. Anthropic plans iterative model updates and a community feedback system to refine layout fidelity.