StoryOS
Deterministic AI video via 3D point-cloud scene proxies.
StoryOS is a demo project being developed for a media company to address inconsistency issues in AI video regeneration. The goal is a deterministic system where video generation produces consistent, repeatable results rather than random variations.
The problem
Current AI video generation tools suffer from inconsistency in video regeneration. When you regenerate a video or make small changes, the output often varies unpredictably: characters change appearance, scenes shift, and details morph. That makes continuity and control in production workflows impossible.
My role
I'm working on 3D rendering and point cloud technology to solve this. By creating 3D representations of characters and scenes using point clouds, we keep geometry and appearance consistent across regenerations, so when a video is regenerated or modified, the core 3D structure remains stable.
Technical approach
The solution builds 3D representations of characters and scenes using point clouds. Maintaining consistent 3D geometry across regenerations means characters and objects retain their appearance and structure.
My work focuses on the 3D rendering pipeline and point cloud generation, the foundation for preserving character identity and scene structure even when videos are regenerated or modified.