Baking results are wrong when using a stationary skylight. It looks like there’s a bug with the sky bent normal baking algorithm.Anyone know the reason?
Thank you for working on this.
Also would firefly removing simple filter not help with the remaining fireflies after denoiser? I noticed those white fireflies are the only thing that requires several more GI samples. Also auto disable realtime when baking would be nice.
Also is there a reason why gpu can’t do precomputed AO?
I’m experiencing the same with any light actually type, any stationary lights are super off and produce no dynamic shadows on anything…
Hey, new to this plugin. I am compiling the engine from source and I was wondering if this disrupts things there with the repo? I do have to share the engine with other team members to be able to work on the project.
I also noticed you are modding the source which is conflicting with modifications on our fork.
Also, would it be possible to have this in a standalone plugin someday?
would it be possible to have this in a standalone plugin someday
Not possible. More precisely, that’s not the path I want to take as I will start to do more engine modifications (especially the runtime part) later. I have the source of this modified GPULM on github (see the release list post). You need to perform a proper merge with your engine fork.
Is there one for 5.7, noticed your repo but it seems to only exist for 5.6 and 5.8.
Also, a few more doubts -
- Merging on source means I do not need to download things from the binary release?
- Can I use this with Forward Rendering?
Thanks for the prompt response.
For 5.7, anything before commit ‘Merge tag ‘5.8.0-release’ into gpulm-no-luoshuang’ is what included in the 5.7 release.
- The only thing you need is OpenImageDenoise*.dll in Engine\Source\ThirdParty\Intel\OIDN\bin\ and Engine\Binaries\Win64\ (for GPU denoising)
- The baked lightmaps and volumetric lightmaps should work, though I’ve not tested if the baker will run. You may need to manually switch back and forth to see
edit: I’ve tested that GPULM should work when Forward Shading is enabled.
So I don’t need to switch between forward and deferred to bake? It’s been years since I saw this.
Thanks for the clarification!
感谢！高尚的共产主义战士
Release 5.8.0 2026.07.06
A newer version is available below
Back to release list
- Add support for spline meshes
4 Likes
Hello, can you tell me if it’s possible to build the project using this version of GPU Lightmass? Currently I am on 5.7.4 and it gives me unknown cook failure
Yes. I tested on several of my test projects and all of them work.
Release 5.8.0 2026.07.08
A newer version is available below
Back to release list
- Add support for baked Ambient Occlusion
- Add support for Material AO Mask
7 Likes
Is AO not available in older versions? I am using version 5.6.1.
Some issue I think, unless it is limitation. See the round artifact on ground on the left, its both shadow and light from the right spotlight that has attenuation ended early. If I would extend it it would land exactly where the curved shadow is on the ground. This comes from the fact that if not for attenuation, the right one would overlap the left one.
Also would it be possible to have Volumetric Lightmap Spherical Harmonic Smoothing and Diffuse Boost?
GPU lightmass:
CPU Lightmass:
How do the lights behave in the viewport path tracer?
Sent you message with this project file. I cannot recreate the scenario in other project. No idea what is going on.
Release 5.8.0 2026.07.13
A newer version is available below
Back to release list
- An issue with stationary local light shadows has been fixed
- Add support for global diffuse boost and per material diffuse boost
7 Likes