This paper introduces a novel framework that integrates Large Language Models (LLMs) into process modeling to enhance the accessibility and efficiency of Business Process Management (BPM). The framework leverages LLMs for the automated generation and iterative refinement of process models starting from textual descriptions. It incorporates innovative prompting strategies, a secure model generation protocol, and an error-handling mechanism. The framework also features an interactive feedback loop, allowing for refining the generated models based on user feedback. A concrete system is implemented that extends the framework, providing robust quality guarantees on the generated models and supporting exporting them in standard modeling notations such as BPMN and Petri nets. The framework addresses the control-flow perspective of process modeling, which forms the basis for data, resource, and operational perspectives. It uses the Partially Ordered Workflow Language (POWL) for intermediate process representation, ensuring soundness and enabling the generation of hierarchical models. The framework employs prompt engineering techniques, including role prompting, knowledge injection, few-shot learning, and negative prompting, to effectively utilize LLMs for process modeling. It also includes error-handling mechanisms to address both critical and adjustable errors, ensuring the reliability of the generated process models. The framework was evaluated with two state-of-the-art LLMs: GPT-4 and Gemini. GPT-4 demonstrated strong performance in generating process models, efficiently resolving errors, and integrating user feedback. In contrast, Gemini showed weaker performance, struggling with error resolution and feedback integration. The framework's models were found to be sound and executable, whereas the TA framework produced unsound models. The evaluation highlights the superiority of the proposed framework in generating sound and executable process models. The framework's use of POWL as an intermediate process representation ensures soundness and provides robust guarantees on the quality of the generated models. The framework's ability to streamline process modeling tasks underscores the transformative potential of generative AI in the BPM field. Future work includes expanding the framework to incorporate additional process perspectives, conducting extended evaluations, and exploring direct BPMN generation without intermediate representations.This paper introduces a novel framework that integrates Large Language Models (LLMs) into process modeling to enhance the accessibility and efficiency of Business Process Management (BPM). The framework leverages LLMs for the automated generation and iterative refinement of process models starting from textual descriptions. It incorporates innovative prompting strategies, a secure model generation protocol, and an error-handling mechanism. The framework also features an interactive feedback loop, allowing for refining the generated models based on user feedback. A concrete system is implemented that extends the framework, providing robust quality guarantees on the generated models and supporting exporting them in standard modeling notations such as BPMN and Petri nets. The framework addresses the control-flow perspective of process modeling, which forms the basis for data, resource, and operational perspectives. It uses the Partially Ordered Workflow Language (POWL) for intermediate process representation, ensuring soundness and enabling the generation of hierarchical models. The framework employs prompt engineering techniques, including role prompting, knowledge injection, few-shot learning, and negative prompting, to effectively utilize LLMs for process modeling. It also includes error-handling mechanisms to address both critical and adjustable errors, ensuring the reliability of the generated process models. The framework was evaluated with two state-of-the-art LLMs: GPT-4 and Gemini. GPT-4 demonstrated strong performance in generating process models, efficiently resolving errors, and integrating user feedback. In contrast, Gemini showed weaker performance, struggling with error resolution and feedback integration. The framework's models were found to be sound and executable, whereas the TA framework produced unsound models. The evaluation highlights the superiority of the proposed framework in generating sound and executable process models. The framework's use of POWL as an intermediate process representation ensures soundness and provides robust guarantees on the quality of the generated models. The framework's ability to streamline process modeling tasks underscores the transformative potential of generative AI in the BPM field. Future work includes expanding the framework to incorporate additional process perspectives, conducting extended evaluations, and exploring direct BPMN generation without intermediate representations.