I’ve always had a question: If we use Gaussian splatting point clouds to replace light cards (since people often mention that Lumen has shadow artifacts and struggles with complex surfaces), wouldn’t this approach naturally avoid those issues? Moreover, could the anisotropic properties introduced by spherical harmonics (SH) inadvertently create better global illumination effects than traditional methods? I’m curious about how feasible this approach actually is in practice. What are the main technical and computational challenges we might encounter with this substitution?
1 Like
That’s a really thoughtful question! Using Gaussian splatting to replace light cards does seem like a promising way to mitigate some of Lumen’s known issues, especially with complex surfaces and shadow artifacts. The added benefit of SH potentially improving the anisotropic lighting response is also an interesting angle.
From what I’ve seen, though, one of the main challenges would be the computational overhead rendering dense splat point clouds in real-time, while integrating them seamlessly into Lumen’s existing GI pipeline, could be quite demanding. Also, ensuring temporal stability and avoiding noise or flickering might require additional filtering or reprojection steps. Would love to hear from anyone who’s experimented with this workflow in practice!