Animating Virtual Characters in Unity Using Generative AI: A Prompt-Based Approach

Creating realistic and expressive character animations is a major challenge in video game development and interactive applications. Traditional methods, such as keyframing and motion capture, demand considerable time and resources. Recently, AI-driven text-to-motion models, especially diffusion models, have emerged as a promising alternative, allowing for the automatic creation of animations from textual descriptions. This thesis offers a comparative analysis of different motion generation models, focusing primarily on diffusion-based techniques. The evaluation examines crucial factors like motion quality, realism, adherence to prompts, and usability in real-time applications. To connect AI-generated animations with game engines, a specialized tool was developed to enable the seamless integration of these models into Unity, providing a practical workflow for developers. In addition to technical evaluation, this work explores whether text-to-motion models can effectively express emotions through movement. By drawing on insights from body language research, a structured approach was created to enhance motion prompts for generating emotionally expressive animations. An experiment was conducted where participants identified emotions in AI-generated animations. The responses were analyzed to evaluate the strengths and weaknesses of these models in producing believable emotional expressions. Finally, this thesis points out significant limitations of diffusion-based motion generation, such as the quality and diversity of training datasets, which greatly affect the expressiveness and generalization of the generated animations. The study concludes with suggestions for future research directions and improvements to enhance the adaptability and quality of AI-generated animations for upcoming applications in gaming and interactive media.