Thanks for sharing! However, I am running into an issue where the output does not fall back at all when disoccluded, tested on a new project.
Digging through the fragment shader, it seems that the depth stored in the alpha isn't being read properly, so disocclusion is never detected, because the shader thinks the depth is always the same. Turning the grid size up in the config gives very obvious incorrect projection.
I've only had issues with world scales that are too small or too large. If the effect is working but you're experiencing problems, try scaling the entire scene.
A large grid size will always increase projection errors, since the static cubemap position can end up differing too much from the camera capture position. Try to keep it around two or three character steps in size.
If you want to debug disocclusion, uncomment line 307 in cuby_marching.gdshader. This will highlight all disocclusion areas in magenta (the "Fallback to Cubemap" option must be enabled).
If you encounter other issues, ensure that you're using Forward+ and Godot 4.5.1 or 4.6.3.
If you still run into issues, feel free to reach out — I’ll be here to help until everything is working properly.
Downgrading from 4.6.2 that I was on to 4.5.1 seems to resolve the issue. I can see the magenta appear.
On 4.6.2 and 4.6.3, there's no magenta, and whenever the probe changes position, things don't seem to align properly and there's a strange smearing.
With the same settings in 4.5.1, it looks how I expect, similar to the video.
Good to know — I’ll investigate and reply if I find the reason for this issue. It’s probably related to the compositor.
The smearing is caused by the Temporal Accumulation on the cubemap compute shader. It only affects geometry in the mid to far field ('TA Start' and 'TA End' on the configuration).
Thanks for the feedback!
