While I have no idea how that would be implemented, it does make me think back to Distance Field Indirect Shadows and the like. That was a per-mesh distance field, that could use interpolated GI from baked probes to cast accurate indirect shadows. I wonder if one could have a per-object surface cache parameterization as well, such that you could have a totally baked scene with only a handful of dynamic objects, and only need to pay the surface cache cost for a few of them.
All that said, Epic is clearly moving in the broad direction of entirely dynamic lighting, but the performance just isn’t there yet, at least not on consoles.
It would be nice if we could “pre-calculate” the probes and stuff, unless we specifically put a “box” there that says: calculate this all the time please.
that said, I am not that sure which part of lumen needs what amount of compute.
But… then again, with Graphics Hardware advancing fast (aside from midrange -.-), this would be obsolete quite quickly.
While I am largely with you on the obsolescence, given how messy this console generation transition was, there are no guarantees that we’ll get a serious next-gen console upgrade anytime soon (barring some unexpected technical breakthrough). A solution that can ship on current consoles will matter a lot to scale, even as anyone with a PC gets access to real-time path-tracing and many other things consoles couldn’t possibly support.
unless we specifically put a “box” there that says: calculate this all the time please.
I had that same idea but I came across one scenario where is might not be workable due to innate architecture of Lumen: Moving lights would invalidate the pre-calculated.
Funny enough jblackwell mentioned interpolated GI which could solve the pre-calculated profiles(time of day) of the world.
MGSV(2014) used ambient lights/probes(skylight blockers) for every hour of the day and every weather change. The ambient lights used 1/4th resolution cubemaps(not sure what was base resolution) of their static surrounding and then interpolated the those ambient lights as the time of day and weather changed.
I would say this worked visually and performance wise extremely well EXCEPT for a one kind of scenario. When any character would go beneath a strong light in a dark area, the model would not receive any lighting except from the strong light which resulted in the objects being almost pitch black in most areas which looked terrible.
With a “GI calculate box”. You could solve that by attaching that box to any dynamic or stationery/destructible such as the character to absorb proper bounce light.
This is what I mean by lack a per scene optimization tools if anyone was wondering about that when I first said it(I think Miguel1900 or Arkiras). More tools and ideas to save computing on areas where we don’t need computing.
Personally, I think we’re not getting these tools because UE targets FN. FN is a uncommon and worst case scenario because you can’t bake anything(because everything is destructible) meanwhile most open worlds(and successful ones) mostly consist of static objects like the environment.
But, hey I’ll throw out some ideas here.
Hi. I have noticed that High Quality Translucency Reflections is not working in Unreal 5.3.1. Can’t get any improvement on reflection quality of translucent meterials by checking “High Quality Translucency Reflections”. It works in 5.2.
i gotta add… some options required for this type of sceneario are not properly applying in the postprocessing editor nor when switching into game mode. they work however when you enable them via commandline on begin play or map load. that’s for proper ingame control to only use them where you need them.
it’s not brilliant but decent. translucent has issues with depthbuffering. it’s rendered in a different pass, afaik. masked does not have that problem. it’s just random solid pixels. hence why i tried it. ez
brilliant would be if it’d dither in the reflection too. it’s only seethru in the first path, not the bounces. but… that’s a lot of noise and random raytraces thru polygons. not very efficient, i think. in a gaming scenerio you don’t need this kind of precision, anyway. imo
Lumen already supports dithered translucency in way with distance field reflections, I wonder if there’d be a way to hook the proper components up. Dithered translucency always seemed like a very smart solution to transparency (NVIDIA had a groundbreaking paper on stochastic transparency), and I’m sad it never really took off by and large.
A good sample of gamers hate dithered masked objects.
Unreal is famous for it. Hence why so many UE games force TAA.
Which again, people hate forced TAA and hate games that depend on them.
I know I’m being redundant but know one else will stand up for players against TAA on this thread.
@jblackwell
Unreal is years behind translucency. Ever heard of deferred texturing?
It’s what allowed the PS4 and PS5 to display the massive forest(translucent grass etc) in Horizon forbidden west at crispy native resolution.
brilliant would be if it’d dither in the reflection too
Lumen reflections use temporal jittering which means the jitter in the reflections should reconstruct the dithered object in the reflection no?
Dithered translucency always seemed like a very smart
This might be true if temporally dithered effects didn’t require the entire image to be smeared in motion/basic gameplay because temporal filter(DLAA, TAA, etc) dependendency.
A smarter way to use dithering effects would to have a material that leaves 2 frame trail across the g-buffer. Basically do what TAA does but on a per object instance.
All this talk about reflecting translucent materials made me think to test how SLW is working these days. I figured it being technically opaque may make it a good substitute for glass in some cases.
Long story short screen tracing is having issues with SLW.
Single layer water does not receive perspective correct screen traced reflections, presumably due to having its own separate depth pass. Instead, it will be projected onto the depth of whatever opaque surface exists instead.
i gotta be blunt and say, idgaf what gamers think. they get fed a new meta whenever youtubers or tech companies throw up a new buzzword. i think taa still looks good. the stability on dithered objects is not always there, but if it’s dialed in well there are not alot of artefacts. this dithered glass hack for sure is on the edge of what looks good. taa handles it fairly well tho. if it’s needed, it’s there for testing, how good you can make it look.
That is strange. It makes sense as you lay it out, but I was having SSR issues with SLW a while ago, and I just couldn’t figure out the cause, good work.
Something I’m trying to figure out with regards to water reflections: in testing 5.3, I discovered that liquid niagara fluids are now actually present in the lumen scene, with the water surface represented opaquely by what I believe is the albedo of the water. Strangely enough however, the water rendering system still doesn’t appear in the lumen scene.
This is very confusing imo, because the fact that epic got a volumetric fluid simulation to show up in their lighting cache, but not a tesselated sheet or heightfield, is interesting from a development perspective. Although I have no doubt you could art the problem away, it’s strange to have screen traces cast such strong GI from water without any fallback.
