Popov72

That will potentially break code if you move the define... Currently, people can change `uvOffset` or `vSpecularUV` at the location of this define and it will be taken into account...

Yes, that's true the change can be made earlier... I think that's ok to move the define, you should just drop a line in the `Breaking changes` section about that...

Arf! @deltakosh is the boss so do as he suggested :)

Here's what I could come up with using the code [there](https://jcgt.org/published/0003/04/04/paper.pdf):  As you can see, there are a lot less artefacts (only at the base of the ball reflections)...

Another follow up: |Without the backface depth buffer|With the backface depth buffer| |---------|----------| ||| Look at the reflection of the balls on the floor and the reflection of the red...

> Just wonder how it behaves with a scene with walls or high mesh density.. ? It has all the problems of SSR, so depending on the scene it can...

Blurring added. No blur:  Some blur (`roughness=0.1`):  Lot of blur (`roughness=1`): 

The PG I'm using for my testings: https://playground.babylonjs.com/#PIZ1GK#639 Note that it works only with the geometry renderer at the time and not with the prepass renderer yet.

Yes, the prepass renderer configuration was missing. When fixed, the problem with the display was because the MRT is not created right from the start by the prepass renderer and...

Good catch, I have also fixed it in my sources and a couple of other ones related to linear/gamma space for the reflectivity. I still have some problems when comparing...