That does seem to normally solve a lot of problems. Lumen scalability is still definitely WIP, and cinematic tends to yield the expected quality, with occasionally interesting results.
On the bright side, lumen just got an update for improved object lighting in distance fields, especially for lower scalabilities. I’ve yet to test it, but they said it vastly improves lower-tier reflections, and I’d imagine that will make scalability a little easier.
I’m sorry if this was answered before but can we expect better reflections in 5.1? At the moment Lumen is not usable for reflective environments at all and you cant combine it with raytraced reflections.
With software lumen the reflections are completely dependent on the quality of the mesh distance fields, check the lumen reflection visualization views, and make sure your meshes have enough distance field resolution, and that they accurately represent the scene in the debug views.
This is hardware lumen using hit-lighting. Reflection quality 4.
The problem is that there is no reflection depth. It just reflects diffuse and tries to fill in with screentraces so everything behind the camera is gone. At the moment we are trying to render a ship that has reflective paint and the water reflects as a dark diffuse surface unless you are viewing it from an angle where screentraces can take over.
(just imagine the silver ship from star wars episode 1 flying over a water surface)
Lumen does only support a single bounce of offscreen reflection, yes. On-screen is covered by screentraces, up to a number of bounces I don’t recall.
I asked Wright when we could expect this feature to be instituted, and he basically stated not for a while. In most cases, a single bounce is all you need, and the cases that need more generally involve either RT with baked lighting or path-tracing.
By the way, SingleLayerWater and translucency isn’t supported by lumen at all in its’ current state. That could be part of the problem as well.
Thats bad news I dont necessarily need lumen to support this but I’d be nice to get any solution to this problem. Thinking about it there should be a lot of cases where reflective surfaces show up in a reflection. Just get close to a window in a city and the skyscraper behind you wont show any reflections. Btw I’m using an opaque water material to work around the single layer water issue and it works surprisingly well. I think for now I’ll try to solve this with some scene captures and fake reflections.
If you need a good implementation out of lumen reflections, I heavily reccomend @BananableOffense 's work. Their reflection cheats appear performant and relatively simple to implement all things considered. They even figured out a strategy to fake the appearance of caustics, truly clever stuff.
Lumen GI is pretty great in my experience, but Lumen reflections present a really interesting series of design decisions when it comes to reflections and it still requires some designing around. Unlike with cubemaps (and perhaps SSR) lumen reflection quality really can’t do rough reflections well, which is bad because that’s the exact sort of thing you need to hide the discontinuities between the primary view and the lumen scene.
I’m assuming this won’t come in 5.1, but 5.2 should include some upgrades to lumen reflections, quality at least and preferably perf. And imho, support for multi-bounce reflections, irregardless of the cost. It’s cheaper than path-tracing and faster than baking lightmaps with RT shadows.
100K+ dynamicly deforming Nanite instances running at a playable* framerate. No LODs, no HLOD, no native billboarding. I was told that Nanite with WPO comes with its’ own limitations that aren’t quite apparent to me yet, but still, cause for excitement.Obviously, this isn’t a real game senario test, and the opaque surfaces do make occlusion culling a lot more efficient, but I tested masked geo and it seems to work well. That being said, there were some blocky artifacts that I believe were the product of some optimizations (having Strata on doesn’t help), but overall pretty good.
Basically, you’re not getting ‘unlimited geometric detail at a flat cost’, but you are getting a lot more bang for your buck, with a lessened need to optimize.