GridVids
Sign in with YouTube

Voxel GI with World Space Canvas

Jed037 May 20, 2026 15:11
Create an account to hide videos, mark them watched, or save them to Don't Miss.
Don't Miss
Create an account to hide videos, mark them watched, or save them to Don't Miss.
Mark watched
Create an account to hide videos, mark them watched, or save them to Don't Miss.
Hide
Open in popup Open on YouTube
Share video

Share video

Copy the GridVids link or iframe code.

Video link
Iframe code
Iframe content

It also now supports oriented bounding boxes. Internally, it now also uses analytical normals through the help of rounded box SDF function that returns the gradient alongside the SDF value (see https://iquilezles.org/articles/distgradfunctions3d/) World Space Text is also voxelized by constructing a bounding box for word batches (which are generated based on TextMeshPro info available). Custom logic is implemented to break batches if certain properties do not hold for contiguous slices of characters (for example, text colour). This can be extended to allow breaking off batches based on average character height to make voxels fit more tightly to the world space font. I've optimized the global illumination pass by using indirect dispatch to run the cone tracing on visible voxels only (since interior or fully empty voxels will not be sampled and be of any use). This is implemented using an AppendStructuredBuffer and a prior pass that just checks for surface visibility. This can be also extended to support camera culling, since not all voxels will be visible at all times. There is no temporal accumulation present, so spuriously regenerating / toggling surface voxels should not cause any artifacting or ghosting effects.
Science & Technology