Building a believable 3D game world takes a team. Terrain artists, lighting designers, audio engineers, layout specialists—each role contributes to the task and brings life into the plot in their own way. This is exactly whenDolby researchers set out to automate that process with AI.

GIVE (Generating Immersive Virtual Environments) is a multi-agent AI framework where specialized agents—each focused on visuals, terrain, lighting, spatial layout, or audio—collaborate inside a game engine to generate immersive 3D environments from a simple text prompt.

How Dolby Is Building 3D Game Worlds from Text

Type "a foggy coastal town at dawn," and GIVE gets to work. The tasks follows a sequence:

A manager agent interprets the intent

A planner breaks it into tasks.

Expert agents execute — adjusting textures, placing assets, tuning ambient sound.

The result is a scene built directly inside the engine, ready to explore.

What sets GIVE apart from traditional models is its self-reflection loop. After generating a scene, a vision-language agent reviews the output, flags inconsistencies, and feeds improvements back into the system—much like a creative team reviewing early concepts. This minimizes errors and multiple iterations. In real-time, experimental results show richer, more coherent environments compared to single-model approaches.

For Dolby, this research points toward a future where immersive world creation is faster, more accessible, and guided by AI that collaborates as naturally as the teams it's designed to support.

Meet the AI Studio: Dolby Is Building 3D Game Worlds from a Single Line of Text

{{ pageEducationalSlot.headingContent[index] }}