Large language models (LLMs) have taken the world by storm, showcasing rapid growth, scalability, and diverse applications across multiple areas. Given their multifaceted capabilities, we hypothesize that a system can be modeled leveraging the potential of LLMs like Mistral and other generative models such as stable diffusion to generate playable levels in a game engine. To test our hypothesis, we introduce a multi-modal architecture combining the Mistral 7b LLM with the stable diffusion image synthesis model through LangChain for fully automated level generation. In this paper, we present the project methodology, detailing the system architecture and its functioning. Furthermore, we discuss the evaluation metrics used to assess each essential system component. The system was able to generate fully playable level in the Godot game engine quite well. The quantitative evaluation conducted on our finetuned LLM, our model outperformed both GPT-3.5 and GPT-4 on BERTScore by 2.53% and 2.70%, respectively. Whereas, CLIPScores obtained for each of the generated assets suggested that there was coherency between the generated assets and the prompts used to generate them.