Early Prototype Explores Interactive World Models for Fallout-Style RPG Gameplay

A developer demonstrated an early prototype combining the MiniMax H3 video generative model with a 12B Gemma Vision LLM to create interactive RPG gameplay loop. The system converts 2D isometric scenes inspired by Fallout into 3D volumetric environments that react dynamically to player actions. This project illustrates how video world models and vision-language models can be integrated to simulate game engines entirely through generative AI. It points toward a future where game worlds, mechanics, and visual responses are generated on the fly rather than traditionally programmed. The setup runs in 10-second frames split into two phases: 5 seconds of direct user interaction via WASD or prompts, followed by 5 seconds of LLM reactivity where the Vision LLM calculates RPG dice rolls, health changes, and enemy responses. The prototype currently renders low-resolution 352p, 3-step video clips designed for integration with the Vascura FRONT frontend.