DYNAVis is a visualization editing tool that combines natural language interfaces (NLIs) with dynamically synthesized UI widgets. It allows users to describe edits in natural language, and then generates persistent widgets for further modifications. The tool reduces the gulf of execution and enhances interactivity by blending NLIs with GUI widgets. Users can either describe edits in natural language or directly request dynamic widgets. DYNAVis uses a large language model (LLM) to synthesize widgets based on user input, enabling users to explore and perform multiple edits efficiently. The tool supports both chart design edits and data-related edits, allowing authors and readers to refine visualizations and interactively explore data. A user study with 24 participants showed that users preferred DYNAVis over NLI-only interfaces due to ease of repeated edits and immediate visual feedback. The system design includes a dynamic widget synthesis engine that translates user inputs into HTML and JavaScript code for widgets. The tool is implemented as a web application using React and TypeScript, with a backend in Python. The system architecture includes a data summarizer, chart engine, and widget engine. The user study demonstrated that DYNAVis significantly reduced the effort required for visualization editing, with users finding it easier to perform repeated edits and benefiting from immediate visual feedback. The tool also allowed users to explore and coordinate multiple edits more effectively. The study results showed that participants using DYNAVis found the tasks easier and completed them faster compared to using an NLI-only interface. The findings highlight the effectiveness of combining NLIs with dynamic widgets in improving visualization editing efficiency and user experience.DYNAVis is a visualization editing tool that combines natural language interfaces (NLIs) with dynamically synthesized UI widgets. It allows users to describe edits in natural language, and then generates persistent widgets for further modifications. The tool reduces the gulf of execution and enhances interactivity by blending NLIs with GUI widgets. Users can either describe edits in natural language or directly request dynamic widgets. DYNAVis uses a large language model (LLM) to synthesize widgets based on user input, enabling users to explore and perform multiple edits efficiently. The tool supports both chart design edits and data-related edits, allowing authors and readers to refine visualizations and interactively explore data. A user study with 24 participants showed that users preferred DYNAVis over NLI-only interfaces due to ease of repeated edits and immediate visual feedback. The system design includes a dynamic widget synthesis engine that translates user inputs into HTML and JavaScript code for widgets. The tool is implemented as a web application using React and TypeScript, with a backend in Python. The system architecture includes a data summarizer, chart engine, and widget engine. The user study demonstrated that DYNAVis significantly reduced the effort required for visualization editing, with users finding it easier to perform repeated edits and benefiting from immediate visual feedback. The tool also allowed users to explore and coordinate multiple edits more effectively. The study results showed that participants using DYNAVis found the tasks easier and completed them faster compared to using an NLI-only interface. The findings highlight the effectiveness of combining NLIs with dynamic widgets in improving visualization editing efficiency and user experience.