Daily 10/28: Genie 4 adds sound to world simulation as NVIDIA and Epic reshape game AI infrastructure
Google DeepMind has released Genie 4 with real-time audio and editable scenes. NVIDIA is bringing neural rendering and AI characters to RTX Spark, while Epic Games is moving to remove restrictions on training AI with Unreal Engine and MetaHuman outputs.
- [Google DeepMind] Genie 4 Brings Real-Time Audio and Scene Editing to World Simulation ↗
Google DeepMind has released Genie 4 as a research preview. Users can type prompts during a simulation to change the scene, while the model generates real-time music, ambient sound, and interactive dialogue that match the world. Genie 4 also aims for more realistic people and physics while maintaining coherent scenes for up to two minutes. The release pushes interactive world simulation closer to game prototyping and virtual production workflows.
- [NVIDIA] RTX Spark Brings Neural Rendering and Local AI Characters to Gaming PCs ↗
NVIDIA has introduced RTX Spark, a compact PC and laptop platform with up to 128GB of unified memory. DLSS 5's 3D-guided neural rendering starts with NBA 2K27, while the October model update runs twice as fast as the September version and uses 20% less VRAM at 4K. Nemotron 3.5 Bolt E4B supports local AI character interactions, and RTX Kit brings eight neural rendering SDKs covering features such as skin and strand-based hair. Game developers can combine graphics generation and character interaction locally instead of sending every interaction to the cloud.
- [TechPowerUp] Epic Games Moves to Relax AI Training Restrictions on Unreal Engine and MetaHuman Outputs ↗
Epic Games CEO Tim Sweeney said the company is working to remove EULA restrictions that currently prevent AI training on Unreal Engine and MetaHuman outputs. Developers would still be able to allow or restrict training through the terms attached to each game, while creators hosting content on Fab can use the NoAI tag. The position has drawn criticism because Epic removed an AI-generated Fortnite clone while considering its own training use of engine outputs. Game engine terms are becoming a direct control point for deciding who can use generated scenes and characters as AI training data.
One line: World models and local graphics hardware are making game creation and interaction more immediate, while engine licenses are becoming the battleground over who may reuse those results for AI training.