Recent visual editing methods have demonstrated impressive performance in modifying the content of a given reference image based on natural language instructions. However, these methods simply follow the instructions as given, without engaging in critical thinking or deeper reasoning. In this paper, we introduce Chat2Edit, a novel text-based image editing system that addresses these limitations by employing a thinking-command framework, which involves three key steps: Prompt Creator, Command Executor, and Feedback Loop. This approach ensures that the edits are both accurate and contextually relevant. Additionally, the system provides live feedback to the model, allowing it to track and update action plans, handle exceptions, and suggest editing parameters. To evaluate the system, we propose a diverse dataset, Visual Edit Instructions (Vis-Edit Instruct), containing 90 scenarios designed to reflect real-world editing demands. Experiments demonstrate that Chat2Edit outperforms advanced editing systems such as ChatGPT and Gemini, achieving superior accuracy in understanding user intent and delivering reasonable results, while maintaining the model’s adaptability. Code available at: https://github.com/nghialt3670/chat2edit-python.git .