Add Blend Distance property to ReflectionProbe

[lander-vr]

Supersedes #61416

This PR adds the "Blend Distance" property to reflection probes. This parameter is exposed to offer artists control over how quickly a reflection probe blends into the scene.
Some context on why this property is necessary: #61416 (comment)

This PR builds on @Calinou's initial work, however I've entirely reworked how the blend parameter is calculated and smoothed.
This is for two reasons:

• The blend is now calculated for each axis of the probe separately. This is necessary to keep an even blend-distance on each side of the probe bounds. In other words, this avoids the blending being warped based on the proportions of the AABB.
• The blend smoothing before had an issue where it would strangely reduce the strength inside the probe.

Blend Distance is a float representing the length of the blend in meters, with a default range of 0 to 8, or greater.

I've added an interior wireframe to display the extends of the blend-distance within the AABB of the main probe. This makes it clear at what point the blending stops, and where the probe emits "full strength". (For the sake of presentation, I commented out the lines between the corners and center_offset for this GIF, this would be the case if #99920 gets merged)

Wall test scene, replication of the example scenario I gave in the context comment:

Comparison of the same scene to 4.3:

4.3 - unadjusted probes	4.3 - adjusted probes	This PR

Unadjusted probes have the same bounds as in the PR.
Adjusted probes have enlarged bounds, which is why the wall in the adjusted probes get stronger reflections. However, since this needs to happen on both sides of the wall to accommodate for the view from that side, you get bleeding.
Notice how despite that adjustment and being well within the bounds of the probe, the wall is still significantly darker due to not receiving full reflection strength.

[jcostello]

Looks awesome. I notice that you can blend above the bounds of the probes. Is that correct? I tried and I got nice results but I wonder why above the bounds

[lander-vr]

Looks awesome. I notice that you can blend above the bounds of the probes. Is that correct? I tried and I got nice results but I wonder why above the bounds

I don't think limiting the property based on the size of the box really gives any UX improvements, if anything it could lead to unwanted changes of that value if the user accidentally makes the bounding box smaller than the max blending distance for those bounds. It would also feel needlessly restrictive, in my opinion. The actual blending is min(reflections.data[ref_index].blend_distance, box_extents[i]), so the blending itself is clamped based on the extents of the probe: even though you can drag the blend distance further, you shouldn't actually see a visual change if your probe is too small. You can however then increase the bounds of your probe and the internal bounding box will appear again keeping that blending distance. I think this feels a bit more flexible and nicer.

24_12_03_15_02__godot.windows.editor.x86_64_JzzmqvKW5A.mp4

[Calinou]

Tested locally, it works as expected in Forward+ and Mobile. This new blending makes a lot more physical sense overall.

Testing project: test_reflection_probes_mirage_palace.zip

master	This PR

Additionally, this allows for better fake area lights using bright custom ambient colors:

In Compatibility, I get a shader compilation error and the scene looks fully white. This doesn't occur in master.

ERROR: SceneShaderGLES3: Fragment shader compilation failed:
0(1085) : error C1503: undefined variable "reflections"
0(1085) : error C1503: undefined variable "ref_index"
0(1087) : error C1503: undefined variable "reflections"
0(1087) : error C1503: undefined variable "ref_index"

[jcostello]

Looks awesome. I notice that you can blend above the bounds of the probes. Is that correct? I tried and I got nice results but I wonder why above the bounds

I don't think limiting the property based on the size of the box really gives any UX improvements, if anything it could lead to unwanted changes of that value if the user accidentally makes the bounding box smaller than the max blending distance for those bounds. It would also feel needlessly restrictive, in my opinion. The actual blending is min(reflections.data[ref_index].blend_distance, box_extents[i]), so the blending itself is clamped based on the extents of the probe: even though you can drag the blend distance further, you shouldn't actually see a visual change if your probe is too small. You can however then increase the bounds of your probe and the internal bounding box will appear again keeping that blending distance. I think this feels a bit more flexible and nicer.

The only visual change I see using blender distance further is that you control how much of the sky blends. But maybe we can have a separate slider to control how much the sky blends when using inside=false

[lander-vr]

The only visual change I see using blender distance further is that you control how much of the sky blends. But maybe we can have a separate slider to control how much the sky blends when using inside=false

I'm not entirely sure I understand what you mean. Is this not the behavior you would expect?

Keep in mind that this PR is not really supposed to solve any of the current sky blending issues, this PR just deals with how things blend, not which things. To fix sky blending we need a priority system and sky visibility masks for reflection probes.

[lander-vr]

In Compatibility, I get a shader compilation error and the scene looks fully white. This doesn't occur in master.

Should be fixed now! :)

[jcostello]

I'm not entirely sure I understand what you mean. Is this not the behavior you would expect? !

Maybe its the expected behavior but it feel strange to go beyond the limits of the probes. Like in my mind I dont know what it means to blend further that. Like I said, the only thing I notice was the sky.

To me, if the probe is only 1x1x1m, the maximum blend distance should be .5m. But I like better the unity approach that is with a float value from 0 to 1. Meaning 0 not blending at all and 1 the maximum distance from the center of the probe. Does it make sense in this case?

[lander-vr]

Maybe its the expected behavior but it feel strange to go beyond the limits of the probes.

I'll add a bit more context in the tooltip.

Like I said, the only thing I notice was the sky.

This is strange, I cannot reproduce this. Once the blend distance reaches it's internal limit, there's no visual change for me when going far beyond that value. Looking at it pragmatically though, I don't think it matters too much. In normal use cases the blend range will only be a small part of the box.

To me, if the probe is only 1x1x1m, the maximum blend distance should be .5m. But I like better the unity approach that is with a float value from 0 to 1. Meaning 0 not blending at all and 1 the maximum distance from the center of the probe. Does it make sense in this case?

I think this is mostly a case of getting used to this behavior if you're unfamiliar with it. In the end they're not vastly different.
A % value of the bounds is more abstract in my opinion (For example % of which side?), and causes varying blend distances with different sizes probes.
To me a distance makes more sense, it's more predictable. A meter is a meter, regardless the size of the probe. The probe can have dimensions of 500m x 10m x 10m, but when the blend distance is set to 1 meter, it will still be 1 meter of equal blend on all sides.

[Calinou]

It works great in Compatibility now 🙂

[lander-vr]

Rebased to resolve conflict. :)

[Repiteo]

Thanks!