GDC 2025 Session Viewer

This talk will introduce Deepforge, a deep neural network designed to transform armor pieces for various races in World of Warcraft (WoW). This technology drastically reduces the time required for artists to fit 3D armors from days and weeks to mere minutes. By automating the fitting process, our tool empowers artists to focus more on fine-tuning the details and enhancing the quality of their work. We will explore the model design, training, and deployment, showcasing its unique features and practical applications on mesh fitting for WoW. The primary goal for this project is to ensure high-quality mesh generation suitable for production use, from preprocessing to postprocessing, combined with a robust infrastructure for efficient and reliable model performance. Attendees will gain insights into how Deepforge enhances game asset creation, making it a valuable tool for developers.

Takeaway

Attendees will learn how we train a deep learning model to automate 3D armor fitting in WoW, reducing artist workload and enhancing asset quality. They'll understand the model's design, training, and deployment, and gain insights into preprocessing, postprocessing, and infrastructure for efficient and reliable performance.

Intended Audience

Professionals such as Data Scientists, Research Scientists, Machine Learning Engineers, Software Engineers, and Technical Artists. Ideal attendees should have a basic understanding of neural networks and machine learning, and an interest in automating 3D asset creation and enhancing game development processes.