Transformer-based models have become the dominant standard for natural language processing (NLP) tasks, many of which involve text generation. However, producing coherent and contextually relevant text is not solely determined by the model’s architecture or training quality; the generation parameters significantly influence the output as well. This work presents an in-depth exploration of the key parameters that govern text generation in transformer models. We provide an empirical analysis of the core parameters, examine various decoding strategies, and discuss methods for controlling the creativity and coherence of the generated text. Furthermore, this work presents insights into fine-tuning these parameters to optimize performance for specific applications. Our findings aim to guide practitioners in achieving higher quality and task-appropriate text generation using transformer models.