I think its being caused by the Bayer pattern itself. As more dots are added the patterns aren't equal in brightness. Most visible here in a 2x2 Bayer pattern.

Still some brightness banding occurring that I'm not happy with though. Easier to see it when the image is blurred.

Fixed the SDF antialiasing a bit since some people were checking it out.

(Before/after)

Posted the code here:
bsky.app/profile/matt...

Didn't do much more but have cleaned up the code and released it on Github to demo the implementation.

github.com/mattdevv/SDF...

As in to reduce the banding you remap the brightness using a gradient, and the gradient is precomputed from the ratio of black vs white in each slice of the 3D texture? Sounds interesting but yeah harder with different shape dots.

More work to do, there is some precision issues in the texture coordinates, and need to smooth out the sdf at low angles.

Comparison of different size Bayer matrices: 2x2, 4x4, 8x8, 256x256

Thanks, mine is in Unity too. I would like to release it once it has more features. Looking to implement other sized Bayer matrices next week.

Yep for each pixel I find its fractal level to subdivide the UV the correct amount. Then the UV is shifted to the center of each 2x2 Bayer sample and scaled by the fraction into the fractal level. Finally I min() the 4 SDF samples together and apply a hard edge.

No textures are used in dithering (except for albedo on the cube). Dot shape can be changed to any SDF equation.
(Circular Dots, Star Dots, Stars SDF, Stars UV coords)

Amazing Video!

Tried my hand at implementing this just from the video but with SDFs instead of a 3D texture. Turned out pretty good but missing a lot of the quality touches from your final shader, however I can now switch the shape of the dots easily (See reply).