AI makes Doom in its own game engine — Google's GameNGen project uses Stable Diffusion to simulate gameplay

Google Research has successfully developed GameNGen, an AI-driven game engine that generates playable, original Doom gameplay entirely through a neural network. By leveraging Stable Diffusion to process player inputs and previous frames, the system creates new visual content with remarkable fidelity and cohesion. This achievement marks a significant milestone in AI-generated media, demonstrating that complex, interactive environments can be simulated in real-time rather than pre-rendered or statically constructed. The study addresses critical limitations of existing generative models, such as frame flickering and visual degradation over time. To maintain consistency, the team employed a self-correcting mechanism using Gaussian noise to repair context frames, ensuring that the generated visuals remain stable during extended play sessions. This technical innovation allows the model to preserve logical continuity and visual quality, solving the "photocopy effect" that typically hinders long-form AI video generation. This development is highly relevant to the open data community as it showcases how diverse datasets can train AI to replicate complex systems without explicit programming. By training agents to play the game extensively, the researchers created a rich dataset that enables the model to learn physics, logic, and aesthetics autonomously. It highlights the potential for open datasets to fuel autonomous game development, potentially lowering barriers to entry for creating immersive, procedurally generated experiences in the future.

Source: tomshardware.com
Published on 2024-08-29

₹100 crore sought to reach a compromise on MUDA site issue, alleges KPCC spokesperson »

Open Data

Related news