XR AI Spotlight From 3D GenAI to the End of Game Engines
Dec 3, 2025
Tejas Kulkarni, CEO of Common Sense Machines and a former MIT researcher, dives into the fascinating world of 3D generative AI. He discusses how CSM transforms images into game-engine ready 3D assets, explaining key metrics like geometry and topology. Tejas underscores the importance of modularity in production workflows and predicts a future where neural networks fuel game engines. He also explores the societal impacts of democratized creation, sharing advice for creatives on embracing AI's potential and understanding its cultural nuances.
AI Snips
Chapters
Transcript
Episode notes
Perception vs Topology In 3D Evaluation
- Human evaluation in 3D GenAI prioritizes appearance but misses topology and parts.
- Common Sense Machines focused on geometry, splat rendering, topology, and parts to improve perceived quality.
Parts Are Core To Game Engine Design
- Game engines impose representation constraints that make parts essential for animation and interaction.
- Tejas predicts three engine classes: current programmatic, splat-based, and future world-model-driven engines.
Offer Parts Or Kits For Production Use
- Use a parts-based or kit system to make AI assets production-ready for AAA pipelines.
- Offer both auto-assemble and kit-based outputs so artists can choose lower friction or precise control.