DPhil students contribute to breakthrough AI model for creative gameplay design

Three DPhil students from our department Gunshi Gupta, Shu Ishida and Tarun Gupta were a part of a collaborative research that presents WHAM (World and Human Action Models), a generative AI system capable of simulating dynamic, editable game scenarios based on human input.

The team developed a generative AI model called WHAM that can simulate gameplay by learning from human interactions in video games. Instead of just generating visuals or random sequences, WHAM can produce consistent, diverse, and editable scenarios, letting creatives explore ideas by tweaking game scenes and seeing how they evolve. The goal of this study was to better support the creative process in game development, particularly during early-stage ideation when designers are experimenting with new characters, levels, or mechanics.

Creative industries like gaming, film, and design increasingly rely on tools that help brainstorm and prototype ideas. By enabling machines to understand and extend human creativity, tools like WHAM can reduce the time and effort needed to bring imaginative ideas to life, whether it’s designing a new video game level or visualising an interactive story.

"It’s exciting to explore how AI might become a meaningful creative partner. With WHAM, we’re showing how technology can begin to support and amplify human imagination in designing interactive worlds." - Gunshi Gupta.

The most exciting breakthrough was seeing the model not only generate plausible futures in a game but also remember and adapt to creative changes made by a human, like adding a new character or object, and naturally incorporating that into ongoing gameplay. It was a small but powerful step toward true human–AI collaboration.

‘‘The power of WHAM lies in its ability to remember and adapt to a designer's creative input in real time, making it more than just a scenario generator - it's a collaborative partner. This could fundamentally change how creatives approach game development, particularly in the early stages, by significantly speeding up the process of prototyping and exploring innovative ideas.’’ - Tarun Gupta.

Creative iteration is often time-consuming and costly. Our proof-of-concept WHAM model shows that we can train AI to be more intuitive partners for humans during the early, messy phases of ideation. In the future, this kind of technology could democratise creativity by giving small teams or individual creators the same prototyping power as large studios.

Read the research Nature publication here: https://www.nature.com/articles/s41586-025-08600-3#author-information