Multimodal models such as GPT-4V and Claude 3 can directly convert design mockup screenshots into runnable front-end code, with accuracy already reaching a level usable in production environments.