GDC 2025 Session Viewer

How are machine learning models and gen AI being used to power up Call of Duty development tools? The vast environments built for Call of Duty demand innovative solutions for managing an ever growing and extensive asset library. Multiple studios within Activision must efficiently find and retrieve 3D models, photogrammetry assets, animations, materials and more. Since traditional search methods, such as lexical string-matching, fail to capture the context of a query, handle synonyms and scale to manage an extensive asset library, Activision studios developed an AI-based semantic (multimodal) search engine that allows users to find and retrieve assets using natural language and images like concept art or reference pictures. Also, learn how large language models are incorporated into the studios' workflows to perform Retrieval-Augmented Generation (RAG) to assist developers.

Takeaway

After this presentation attendees will have a better understanding of how LLM's can be used in various game development workflows and how multimodal neural networks can help artists and engineers work more efficiently. The goal is to inspire attendees to implement these techniques into their own pipelines and tools.

Intended Audience

There are no prerequisites for this talk other than having tried ChatGPT-like systems. The talk is intended for Artists (Animators, Environment Artists, Character Artists, etc.), Technical Artists, Technical Directors, Pipeline Engineers and Tools Engineers.