Are you saying we could afford more light bounces?
Perhaps some tweakable parameters to choose the distance of voxelization for example, so anyone can adjust it depending on their needs and computer.
Multiple bounces are a different story, cause even if I improve a lot the tracing code, time increases exponentially with the number of bounces. I’ll implement a different approach to that, tracing at Voxel level and using the prev frame Voxel grid, so that will give a rough approximation to the other bounces. Coupled with the accurate first bounce should look good enough and have a small performance hit
That’s already I place actually, you need to define the bounds for the Voxel grid and the Voxel size. Also you can make the bounds follow the camera. Another thing to keep in mind is that I’m using the lowest lod to voxelize, but most of epic’s showcase scenes, or even the starter content have no lod in place.
Downloading that, first thing tomorrow ! 
Just pushed it to git. For some reason 4.9 created a reflection capture, so when I started the demo it looked strange. In any case, AHR works correctly with precomputed lighting, so it’s just deleting the probe if you don’t like it.
Sorry for the stupid question but how can I download ?
I have search a little in but I cannot find it 
Cheers!
Here : https://.com//UnrealEngine
Make sure you are logged in on and your account is linked to your Unreal Engine account.
Thanks Comrade! Downloading 
Hey . Downloaded it clean. Gave it a spin. Deleted probes like you suggested.
Nothing seems unordinary right now. Works well, though I couldn’t see a performance increase.
Also reflections are reflections still broken ?
Did you ran with the -DX12 switch?
Also, yes, they still are. Working on that right now.
Hmm no sorry. Can you tell me how to do it ? I’ll try ASAP.
You run it from visual studio or from the exe?
From visual studio you have to go the the UE4 project, right click -> properties and under debugging add -DX12 to the Command Arguments box. Then open with ctrl + f5 as usual.
If you run it from the .exe, think that creating a shortcut with the -DX12 option added should work, but hadn’t try it.
Hey guys, fixed reflections. Was a problem with the initial displacement math, and ue was getting confused and overlapping texture registers.
All working now.
Hey , something I’ve seen done else where in another GI solution is “infinite” bounces. I have no idea how any of your coding is done but I’ll post what they said in case you can figure something out from it: " is done with a clever feedback loop. With each frame, indirect lighting gains an additional bounce. Enlighten actually uses a similar approach for rendering infinite bounces (which took me months of using it to actually notice). The visual drawbacks of doing are nearly unnoticeable."
If you’re interested, the full post/thread is here.
Yep, that’s what I’m gonna do for the voxel grid. The thing is that I’m not currently doing any pre-process on the voxel grid, so I need to add a new stage to trace at voxel level. Not a big problem though, and could be split over a few frames so the performance hit is minimal.
In any case, yeah, feeding the last frame grid to the current one is kinda the standard way to do multiple bounces on voxel based stuff, I mean, I have seen it a bunch of times before.
Thanks for the tip anyway, and that screens look really good.
Considering there’s a limited trace distance I’m picturing some very odd energy variance, where corners get tons of bounces and are fantastically overly bright while isolated areas are far too dark. So far as I know unlimited bounce hack, while it sounds good on paper, isn’t actually used as such for the most part. But maybe what’s in my head is just wrong.
Either way, the main problem I can see is the temporal trailing. If the idea is that it’s dynamic and moveable you’ll get really odd looking temporal instability, where light kind of follows really slowly to “catch up” to the scene updates. It could be a neat specific art effect, kind of reminds me of the Discworld novels and how light is supposed to be “slow” there. But it will probably be unwanted. And looking at the linked thread, yeah the artifact I mentioned of odd energy variance is very much present there. The super bright areas near the light and dead darkness where it’s too far away, doesn’t look as bad as I’d thought though.
Hey , just tested AHR with a test level from my game and looks to be working quite well! will grab the reflections fix and try out DX12 before posting some screens
The thing is that you normally give a smaller weight, like 0.5 to the prev one, so you get something more similar to a moving average. Also remember that light looses energy on each bounce, so even if you have a lot of bounces contribution is quite small after 3 or 4. You do have a point with the limited range thing, but don’t think it’ll be much noticeable. Guess I need to try and see.
Tbh, temporal trailing shouldn’t be that bad, did a few tests a while back and actually looked better, as it smoothed Voxel changes.
EDIT: also is what NVIDIA is using to bring multiple bounces in vxgi if I remember correctly
Well, trying it is the only way to really see, hope it works out! Like I said, looking at the later pictures in that linked thread, the energy difference is there, but it’s not really that bad. Sure lights have really bright spots around them and distant corners now get overly dark in comparison, but it’s not the end of the world.
You would think latex would work like how the ti-84 text used to look before mathprint.