In recent years, the widespread use of large-scale language models, such as ChatGPT, has facilitated the generation of various documents. Moreover, numerous studies have been conducted on automatic dialogue generation using large-scale dialogue models, with accuracy improving daily. Most automatic dialogue generation targets chats, Q&A, manuals, and so on. However, automatically generating dialogues incorporating humor, such as those in Manzai scenarios, remains challenging. In this study, we explore the potential for generating dialogues that include humor by employing several existing large-scale language models. Specifically, we focus on Manzai, a form of Japanese comedic content, as a case study for humorous dialogue. For this purpose, we utilized models such as Llama2, Llama2-Chat, and ChatGPT to generate a Manzai scenario automatically. Additionally, we fine-tuned Llama2 and Llama2-Chat with various datasets to automatically generate humorous dialogues and compare the outcomes.