Localized-IBL implementation

[ZacD]

What do they mean by that?

This: https://docs.unrealengine.com/en-us/...etricLightmaps

But it only fixes the issue where you can bake lighting.

[scha]

This: https://docs.unrealengine.com/en-us/...etricLightmaps

But it only fixes the issue where you can bake lighting.

Sorry, maybe I don't understand something, but if you can bake lighting you don't need the IBL, no? I don't understand what they mean by "static baked lighting AND dynamic sky/sun".

[ZacD]

I think you could technically bake lighting in interior areas and still use a dynamic sun, I haven't tested blending the two. Like imagine you had a game like PUBG with lots of houses but a large open map. Obviously you couldn't bake the lighting on the terrain, but you could bake lighting within the buildings.

[scha]

I think you could technically bake lighting in interior areas and still use a dynamic sun, I haven't tested blending the two. Like imagine you had a game like PUBG with lots of houses but a large open map. Obviously you couldn't bake the lighting on the terrain, but you could bake lighting within the buildings.

I see. I wonder if there is some documentation on that, because i don't really understand how to use baked light and the movable light together. I don't really understand if movable skylight will use lightmaps correctly. Also, how it works with world composition (for large world). We tried something like this: https://www.youtube.com/watch?v=oz9oRyuJTfs , but it uses 2 skylights in the same level, probably will break something.

[Chosker]

interesting.

so it seems you can use a Stationary SkyLight and a Movable DirectionalLight together. After baking the lighting you'll get the skylight baked into lightmaps + volumetric lightmaps, but you'll be free to change the SkyLight color+intensity (since it's stationary), as well as the DirectionalLight entirely (since it's movable)



here's a quick experiment with a cave entrance.

the Skylight is fully visible outside with a natural transition towards the inside.

With The Skylight being Stationary I could change the color+intensity to simulate different times of day. And since the DirectionalLIght is fully dynamic it could be a rotating sun light.



this setup seems useful for mostly static scenes with changing lighting conditions, which is really the same setup that the Localized IBLs work best at

[jonimake]

interesting.

so it seems you can use a Stationary SkyLight and a Movable DirectionalLight together. After baking the lighting you'll get the skylight baked into lightmaps + volumetric lightmaps, but you'll be free to change the SkyLight color+intensity (since it's stationary), as well as the DirectionalLight entirely (since it's movable)



here's a quick experiment with a cave entrance.

the Skylight is fully visible outside with a natural transition towards the inside.

With The Skylight being Stationary I could change the color+intensity to simulate different times of day. And since the DirectionalLIght is fully dynamic it could be a rotating sun light.



this setup seems useful for mostly static scenes with changing lighting conditions, which is really the same setup that the Localized IBLs work best at

That does't change the reflection environment though, does it? Meaning that even though the diffuse lighting might be correct the specular reflections would be wrong.

[Chosker]

That does't change the reflection environment though, does it? Meaning that even though the diffuse lighting might be correct the specular reflections would be wrong.

that is correct. but the same applies to Localized IBL with a rotating DirectionalLight - the rotating sun won't modify the IBL
(unless you try the trick of capturing several probes per time-of-day with different lighting and blending them in/out as the time cycle goes, but you'll quickly reach the 341 probe limit)

[eobet]

alright the pull request was finally rejected: https://github.com/EpicGames/UnrealEngine/pull/4856

TL;DR:[/LIST]
I mean what

Very disappointing. So both Unity and Unreal have very poor real-time gi, and in both cases, the community is trying to make up for it while the official channels seem to be just relying on Nvidia to do the work for them?

I don't get this on Epic's part, because they want to be AAA and many AAA titles have real-time GI today.

[Chosker]

Very disappointing. So both Unity and Unreal have very poor real-time gi, and in both cases, the community is trying to make up for it while the official channels seem to be just relying on Nvidia to do the work for them?

I don't get this on Epic's part, because they want to be AAA and many AAA titles have real-time GI today.

just to be double clear: my Localized IBL system does not produce real-time GI in any way. it relies on pre-baked probes that cannot be recomputed at runtime.

[scha]

I don't get this on Epic's part, because they want to be AAA and many AAA titles have real-time GI today.

Because everybody will have real time raytracing on gpu soon, including real time GI? (<- btw, that's my noob guess)

... but you'll quickly reach the 341 probe limit

Could you put those captures in dynamically loaded sublevels, like lighting scenarios?

[Sn_a_ke]

Because everybody will have real time raytracing on gpu soon, including real time GI?

ray tracing is almost useless for global illumination ...
there is not enough power even for top-end video cards, while medium-sized player configurations will not be enough.
you can not even think about it in the coming years (5-10).
I think for ray tracing there will be more intellectual applications than letting beams in all directions and counting every rebound for light...

[Chosker]

Could you put those captures in dynamically loaded sublevels, like lighting scenarios?

I've never tried this but I doubt probes have any loading/unloading capabilities

ray tracing is almost useless for global illumination ...

it's actually happening already. just look up UE4 RTGI and you'll see what's already possible.
but I agree from "what's just now possible on high end hardware" until "technology is mature and the majority of users have adopted it" we are still quite some years away

[scha]

I've never tried this but I doubt probes have any loading/unloading capabilities

I tried now, it seems to work. Reflection captures data is stored in the same "BuilrData" uasset file as the lightmaps, and the light scenarios basically are mede to load a "builtdata" to a level. So, I created 2 light scenarios for my Japanese rooms and just put a capture in each. One capture has an offset and the brightness different from the other. And it looks like on screenshots (1 is without light scenarios, 2 and 3 is with different scenarios enabled.) It's very easy to setup, you can try that yourself.

[Nick!]

alright the pull request was finally rejected: https://github.com/EpicGames/UnrealEngine/pull/4856[/LIST]
I mean what

Well that’s frigging aggravating. Volumetric lightmaps are not a solution as long as they don’t work with level streaming. I can’t believe the person who rejected that is so unfamiliar with the engine they don’t get that.

Right now now there’s a huge gulf in UE4s lighting when it comes to interiors that aren’t fully static or fully dynamic (which look poor), and they keep refusing to implement anything to get a decent middle ground like other engines have been using since last generation like local, controllable IBLs. You either go all static with dense lightmaps where nothing can move (unless it looks like a Scooby Doo prop that sticks out where the lighting doesn’t match well because level streaming forces you to use the old UE3 style dynamic ambient lighting tech) or your project has to run on space hardware from the future for fully raytraced lighting that can’t run on modern consoles or the bulk of Steam users’ mid range PCs.

This solution or if they bothered to support volumetric lightmaps to implement level streaming would have filled in that middle ground between fully dynamic and fully static to be in line with other engines, but I guess they stubbornly don’t want anyone targeting that spec without investing in rewriting parts of the engine in a new branch.

God this is stupid and frustrating.

[scrcr0]

I guess on the upside, even if they aren't going to implement it as an official engine feature you can always use Chosker's branch instead.