Graphics [RELEASED] Lumina GI -Real time Voxel based global illumination for URP-HDRP, no SDFs baking

Without the new optimization (Tested in 5 years old Dell XPS Laptop - 1050GTX GPU)


Adding the new dither optimization


With the new optimization (Tested in 5 years old Dell XPS Laptop - 1050GTX GPU)


Lumina GI URP v1.1
- Added new optimization controls for the skipping of polygons when very close by in a uniform way. Enabled by set one to the DitherControl.x and regulated by the divide factors y-z-w (cut back on trace directions, trace steps and reflection steps). With a divider by 2 in all optimizers get a direct 10fps gain (40fps versus 30fps) with minimal visual impact.

This optimization is now ready to use and will soon be available in v1.1 of the system.

 

Comparisons with the new optimization activated and without (higher fps images have optimization activated with a factor of 2). Tests run in a 5 years old Dell XPS Laptop (1050GTX)

Lumina GI URP v1.1 is now released at the Unity Asset Store !

- Added new optimization controls for the skipping of polygons when very close by in a uniform way. Enabled by set one to the DitherControl.x and regulated by the divide factors y-z-w (cut back on trace directions, trace steps and reflection steps). With a divider by 2 in all optimizers get a direct 10fps gain (40fps versus 30fps) with minimal visual impact.

Lumina GI URP is a fully real time, voxel based, true global illumination system (not screen space), that will keep the GI even if the light is out of the camera view, thus creating a fully realistic and stable GI representation.

The addition of the new optimization in the HDRP Beta version is the current work in progress, will post results from HDRP soon as well.

ARTnGame - Instagram - Facebook - Twitter - Discord - Youtube - Reddit - Discount Offers

 

UPDATE:

The addition of the new optimization in the HDRP version is finalized and in the testing phase, will post results from the new HDRP performance soon.

ARTnGame - Instagram - Facebook - Twitter - Discord - Youtube - Reddit - Discount Offers

 

Lumina GI HDRP work in progress.

Added the new URP optimization method in the HDRP version as well, for a controllable performance gain on three factors (tracing steps - density and reflection steps).

The upper image is without the optimization and lower with the optimization activated with a factor of 2 for a gain of around 10fps in a 1050GTX GPU.

ARTnGame - Instagram - Facebook - Twitter - Discord - Youtube - Reddit - Discount Offers

 

New demo with LUMINA


Without LUMINA


With LUMINA

Lumina GI URP work in progress on new demo scene.

ARTnGame - Instagram - Facebook - Twitter - Discord - Youtube - Reddit - Discount Offers

 

UPDATE of work in progress on HDRP version.

Lumina GI HDRP 2ond Beta will be available for download soon, the new optimization mode has been tested extensively and adapted to multiple of the demo scenes.

The new optimization method adds a controllable performance gain on three factors (tracing steps - density and reflection steps), like in the URP version.

ARTnGame - Instagram - Facebook - Twitter - Discord - Youtube - Reddit - Discount Offers

 

Lumina GI URP work in progress on new demo scene, testing mesh based lighting. The meshes can be in the GI solution casting light even when the emitting mesh is invisible to the main camera, if required.

The lights opposite of the curtains also create a proper translucency effect, due to the voxel based real GI solution.

ARTnGame - Instagram - Facebook - Twitter - Discord - Youtube - Reddit - Discount Offers

 

Lumina GI - Showcase of difference with screen space solution like in Unity HDRP method.

In voxel solution the GI is always accurate and can have light bounce even in corridors behind the areas the main light hits, also the color is spread based on the actual color surface placement and not in screen space where color is in areas that make no sense.

The lights opposite of curtains also create a proper translucency effect, due to the voxel based real GI solution.

So the voxel solution is the only way to get proper realistic lighting in a scene.

The HDRP version of LUMINA is also finalized and in last stages of testing, with a release coming soon. Lumina GI HDRP Beta is also now available for download for all users of the URP system.

ARTnGame - Instagram - Facebook - Twitter - Discord - Youtube - Reddit - Discount Offers

 

Lumina GI - New demo showcasing more options on the tweaking of the system.

ARTnGame - Instagram - Facebook - Twitter - Discord - Youtube - Reddit - Discount Offers

 

Hi nasos,

great work. i love that you took SEGI and continue with it, very good!

1. somebody asked if the "wet look of SEGI" is also here. Most screenshots you posted do INDEED look wet. but isnt that due to metallicness and smoothness being both setup for materials as very high? could you do a screenshot with materials with metallicness 0 smootnhess 0, then mettalicness 0.6 smoothness 0 and metallicness 0.6 smoothness 0.6 and finally mettallicness 1.0 smoothness 1.0

2. Somebody asked: "Is there anything that must be done at shader level? like use a special material or something, or is this compatible with everything?" i think you missed the question or perhaps thoughts it is a question about "emissiveness". i want to ask this same question but for the receiving end. especially transparent materials. i have seen only opaque materials being rendered with Lumina in all your screenshots. SEGI had of course big issues with transparency (out of the box).

3. and another question about instancing: how does the voxelization work? a secondary camera? what about foliage instanced on terrains, what about generally instanced meshes or drawh mesh instanced calls... are they not or automatically or manually included in Lumina results?

4. could you post a screenshot where you take your 256³ voxel grid and scale it on a bigger landscape, just to give an impression how the low resolution looks then upfront?

thanks for your good work, i will keep this thread subscribed for sure!

 

Hi, thanks for the interest in the system and detailed questions, really helps with focusing on next steps of testing

I will check all in both hdrp and urp versions and get back on each point asap.

 

q1-2.HDRP - various surfaces and
transparencies




q1-2 URP
various surfaces and
transparencies


q1-2 HDPR two different floor surfaces on same lighting - LUMINA settings



q3. URP - Instanced Indirect grass





q4.URP voxel grid scaled


q4.HDRP voxel grid scaled

Hi,

Here is some tests i made for the above, with images per question number and pipeline for reference

The wet look is mostly due to surface properties, the system is full automatic, so not need to change any materials, also seems to work ok with transparent in those first tests, same for instanced Indirect grass from my GIBLION Anime scene maker asset, i tested this only in URP for now.

 

Thank you very much for the effort you spend to make the screens, very appreciated! you could ahve them annotated, because some look like you disabled GI to show the difference, but then have strange artifacts or contents that seem like they are GI just with other settings ... can be confusing

Let me review them for you. Sorry if i sound harsh somewhere, i dont mean harm, just trying to objectively observe them.

FYI i am also working with SEGI and have some experience with it. i am using built-in standard rendering thought (and want to stay using that for my project).

1. transparencies images: in HDRP i see no effect from the transparencies. which means they look correct to me, but the GI is not applied? and what are those red effects and the strangely, mirrored reflections. some kind of reflection or refraction shader? or is it the SEGI reflection tracing that does these strange effects? as this usually works well, i would guess very strange material settings, can you show them? the second and third confuse me, as the third looks without GI but the transparent sphere top center looks very strange.

1.b in URP it looks good to have the emissive transparent spheres orange, red ,cyan in background show up in the voxels. but shouldnt the red sphere get affected by the orange one? so again i feel like GI is not applied (cast onto) the transparent objects. (while we can clearly see that the transparent objects to affect the voxel GI properly). and as usually the voxel grid resolution makes it difficult for them to look good on ground, here more smoothing would help. a typical SEGI "problem" that should be solveable with more processing power. did you add anything yet beyond the gaussian mip blur?

1.c for the folliage in foreground on URP, which looks 1bit opaque (so should/could be defferred?) i am not sure if i can see a faint green cast in the shadow. perhaps? the foliage bottom right in HDRP (looks strangely blurred texture?) seems to correctly cast green, maybe even a bit to much on the curtain?


2. the two different floor surfaces look strange. obviously the wet one has GI and SEGI typical path traced reflections on ground which look "wet", but the top picture is unclar to me. the wall in back shows reflectionms (SEGI?) but near red and green and the sphere we see NO GI at all? or maybe the blue cast is SEGI "fakeGI" from the sky? its hard to tell, because the light angle is different and one loses the comparability with the shot bellow. i was hoping to find a difference in settings on right hand side inspector, but it is not SEGI / LUMIA.


3. URP instanced direct grass: here the orange flowers look good, and the plants and grass seem to cast green on the ground. but an on/off comparison would have helped, as we dont know how greenish the ground texture is.

3.b i dont see the orange glowing ball effect neither the deffered opaque broad leaved plant, nor the grass, nor any of the white scanned objects. perhaps the voxel resolution and exactly these camera angles did not help here?


4. voxel grid scaled: these are the most important screenshots, i think, for everyone interested to do open-world like big games these days. we can see veyr well that it looks good with 300/250 aswell as 100/50 aswell as 84/82 aswell as 284/282 for voxels and shadow space respectively. nice!

4b. on HDRP the orange flowers seem to be forward rendering / real transparency and look wrong. this is usually the big pitfall of SEGI, that the lighting of transparent surfaces (PFX/VFX/Foliage) makes them look bad, and one must switch to opaque (1bit) deffered for shaders as often as possible. but one ones to do that these days anyway, to get uniform lighting and much faster unified rendering. AFAIU most modern foliage packages offer their plant shaders (with wind, seasons, etc etc) as deffered.

any idea what to do with real forward transparent particles, though? i mean if we use VFX graph to create tens of thousands of particles, its possibly easy to even make them opaque and "blend" by size only. but if someone wants to (re)use (existing) shuriken cpu particles, then the few hundres to thousands only make good effects when properly alpha blended. what is your strategy to make these work with SEGI/LUMINA? i mean if your customers buy it, then complain "it doesnt work for this out of the box", what tips/tricks/tutorial do you give them to de-escalate their uninformed anger?

4c. what is strange to me are the last two HDRP images, where the 84/82 voxel size makes good foreground AND greencast from plants on left background stones. the one with higher voxel res (i mean bigger size, less voxels per m²) interestingly has NO green cast on left stone anymore, but the foreground still looks nicely green cast (to me).

4d. i think the most noteable is that for the first URP screenshot with 300/250 one could say that the low voxel density makes it look very wrong, the sofa in the middle getting a lot of green. but honestly it looks great, it feels properly to have all that green.


if you havent stopped readding by now, here are a few new questions i saw:

5. obviously what remains is that down-tuning of over-saturation SEGI has, as right now it always look "in your face" intense, which is good for demoing and screenshots, but looks strange in production of realistically styled games. your opinion on this?

6. do you have any guess why SEGI was not picked up so much in the past 5 years? is it just the performance? did you solve that SEGI uses a lot of cpu (main thread) time per frame? did you solve that SEGI uses more cpu time when the pixels are scaled up (e.g. 3840x2190 versus 1920x1080)?

6b. do you want to elaborate what exactly your optimization with x/y/z/w stepping is (which you have shown in previous screenshots in this thread)? i think your performance boost (on an older notebook) from 30 to 40fps is tremendously good and important. you should advertise this a lot and IMHO inform users what (more or less) exactly you have done to boost performance in this way.

again, thank you for your time, great that you are making SEGI commercial and bringing it to the URP and HDRP people

 

Hi, thanks for the detailed feedback and suggestions, will check on all and get back asap as well.

Note that about transparency i though the questions is if works at all, not about receive or cast GI, this is much more tricky as transparency has issue with image effects in general and is one of the biggest problems in graphics in general, so i will do more tests and study on this, but i would consider transparencies not to cast or receive GI as default and move from there.

I gather that transparent objects could still cast GI from the overall object or have some fallback trick for this, like have a duplicate mesh using cutout shader in GI layer and true transparent in camera layer, this way might contribute correctly to the GI for example (though not sure about cutout as well). About transparent receive GI, perhaps i could apply the effect in a render order after transparent, i will check on this in URP first where is easy to define the order.

 

thanks for your time, looking forward to another reply from you.

let me comment on what you wrote so far:
transparencies NOT producing GI is not a problem, as we can always make invisible emitter e.g. my red explosion shuriken cpu pfx can simply have an expanding sphere with red emission, invisible to normal camera, visible to voxelization of GI. this side is easy to workaround!

but transparencies not receiving GI (due to deffered) is a terrible problem, because you will see the lighting BEHIND the transparent object, as the post process pass is done after transparent shaders are written. so for this a clear and practical solution needs to be taken. one idea i need to try out is to simply render all alpha transparencies (e.g. pfx) to its own render target and add it at latest post process (but then its hard to mix additive and pure transparency, and one needs to make sure they are allready premultiplied to allow for simpler adding in one pass).

i will be happy to hear and see if you try this out, too. i am currently sitting down every night tuning the voxel counts and other performance aspects to reduce the cpu impact as much as possible, gpu impact is easily adjustable by having less objects (only specific layers) and specific GI casting object versions (LOD meshes, etc) ... but cpu impact from SEGI on 4k is still annoyingly super high.

also one other question:

7. do you know if ALL modern drivers can allow more than 256x256x256 volume texture? on my nvidia it is easily possible to go higher? and second question: do ALL modern drivers allow non power-of-2 sizes, again on my nvidia it works well with e.g. 900mb of video ram useage (instead of the "default" 450mb on 256x256x256 "high" setting of SEG). did you experiement with this?

/EDIT: because i reread what you wrote: (though not sure about cutout as well) i can guarantee you that with SEGI cutout (standard shader atleast) works well when voxelizing. for sure the mipmapping of that texture could cause strange colors or to much darkening (e.g. small green flower in texture with big black background, even if the alpha cutout is set correct, as mipmap it will look mostly dark black and the green gets lost). i guess that we can transfer the SEGi cutout situation to LUMINA as well, altough i have little idea if it could break in URP or HDRP ... i fear you have to reevaluate that. wishing you sucess with this

 

I will check this also, if can put the pass before transparent for example

I tried 512x3, have it also as option, named Insane , it run on my 1050gtx but was very slow, so would not recommend it

 

Lumina GI HDRP video tutorial WIP.

The system will be released soon and will be upgradable from Sky Master ULTIMATE and Lumina GI URP with a very big discount.

ARTnGame - Instagram - Facebook - Twitter - Discord - Youtube - Reddit - Discount Offers

 

Lumina GI HDRP and URP new demos WIP. The 1st image is the new pixel art mode in URP version and the rest from the latest HDRP demo.

Lumina GI HDRP and URP are upgradable from Sky Master ULTIMATE with a big discount.

ARTnGame - Instagram - Facebook - Twitter - Discord - Youtube - Reddit - Discount Offers

 

Will it work on distant objects, for instance mountains etc? Like in the image?

 

Hi, the system voxelizes the space around the player at a user chosen level, the larger the distance the larger the voxels so less detail goes to near GI, the far distance beyond the voxels receives only a general ambient

I post sample videos of outdoors for reference. The voxel solution also follows the player if needed to cover infinite maps.

 

Thanks a lot for the anwser and congrats for this excellent asset.

 

Thanks

I have now added multiple more optimizations in HDRP and will come in the next update.

 

Hi nasos,

sorry that your thread in reddit/r/Unity3d went a bit .. critical. i think it is good if you take that criticism to your heart, and try your best.

FYI what i am currently doing is still SEGI on built in pipeline:

i have optimized SEGI now so far, that (at lowest acceptable quality setting) i am having 1-2ms cost of CPU and half to 2ms cost on GPU (2560x1440px on few years old i7 + GTX 1070 desktop). i am drawing skinned mesh renderers, physics driven objects, a terrain and a few hundred dynamic objects in an open world test scene.

i am very happy with the results so far, even though i use voxelization at 1m³. there is still room to improve as i have not yet aded further optimizations by user CK nor user ninlilizi.

also i have particles rendered into the GI volumes, and properly added as last draw without Gi "strangeness". but they still do not receive GI (which would require a special shader and some tricks to not do cone tracing for multiple overdraws).

i still did not solve transparent reflections, these as well as GI shaded particles would IMO best be solved by drawing them first into an additional render target, and when this is finished drawing a final SEGI pass over that target before combining it onto the previous rendered geometry (including the first SEGI pass there).

still looking forward to get my questions answered ;-)

 

Hi, thanks for the update , sure plan to answer all, still doing tests and solving some critical issues in HDRP version, so hopefully can do this weekend

I just made it work in DX12 too, since some users in HDRP version got a crash with the swapchain issue, which does not happen in DX12, so hopefully this will solve that problem with latest Unity and DX12 version use for the HDRP.

 

I'm confused, if I want the HDRP version, which products do I have to purchase exactly?

Also does this work without a directional light? If I wanted to use this indoors with point, spot and area lights only, is that supported?

 

Hi, the Lumina HDRP is for hdrp pipeline, the LUMINA is the original asset for the URP.

The system supports only directional atm and meshes with emission as light sources, so can definitly be used indoors, just not directly with spot and point lights.

You can attach a sphere to point light for example with emit material and exclude it ftom camera layers, but add it to gi layers, to emulate gi from the point light for example.

Or directly use emitting meshes.

 

Vulkan - Ubuntu Linux - NVIDIA RTX 2060 GPU






DX12 - Ubuntu Linux - NVIDIA RTX 1050 GPU

Lumina GI HDRP - Update on supported platforms

Due to users having issues with DX11 in some latest GPUs (which issues do not happen in 1050GTX GPU that LUMINA HDRP was developed on) i have upped the minimum spec for the system to DX12, where the Unity bug is fixed and the crash does not happen and will reevaluate the minimum spec as I investigate the swapchain crash issue further.

I can also now guarantee Lumina GI HDRP runs in Linux with 2060 RTX in Vulkan, I have not tested it or confirmed the same with Vulkan on Windows yet though, will update on this asap.

ARTnGame - Instagram - Facebook - Twitter - Discord - Youtube - Reddit - Discount Offers

 

Ah okay, so what you are saying is, I only need the HDRP version and not the URP one as a base for it?

 

Yes correct, the original version is only URP and the latest is made for the HDRP specifically.

 

I got this error in Unity 2021.3.8f1:
ArgumentException: Input Axis Controller Right Stick X is not setup
What is the correct settings for it?
Never mind - I have changed the input system to the old one.
The default for HDRP is "Both"

 

Hi, i think is due to the sample controller, this is only for moving in the scenes so can use other in its place if need to have only a single input enabled (though i suggest enable both to be globally compatible)

 

The renders are really very nice, but I have RTX 2080 and suffer from the Unity/Windows crashes very often.
I have applied registry patch (TdrLevel), but it only helps with temporary possibility to run scenes,
and then the crashes still continues.
Any tips on changing any parameters to default Lumina prefab to gain some better stability.

 

Hi,

I have adapted the LUMINA HDRP system in the latest version to use DX12 where the crash does not happen, also i did not see a crash with RTX 2060 in Intel NUC with Linux and Vulkan, so hopefully both DX12 and Vulkan can be used to work around the Unity bug with DX11.

The issue does not happen in the 1050GTX i work on, so i assume is some bug introduced in the latest NVIDIA hardware for some reason, maybe was built with DX12 in mind and failed to adapt to DX11 properly is my guess.

Let me know if either option worked in your case as well. I have not tested the Vulkan option in Windows yet though.

 

Thanks, will try your suggestions.

 

Just want to add, that I have tested on the other computer with the different Nvidia graphics card (A1000)
and the swapchain Unity bug persists - editor crash, just after trying to run the scene.
How to disable DX11 and use only DX12 instead?
Will try the URP version to see, if it is stable on my system.

 

Hi,

It is automatic, just need to go to the Player settings, choose not to have automatic graphics API (disable checkbox) for Windows, add DX12 with the plus button and drag it on top of the APIs list. You can erase the DX11 one also after that.

The URP so far has not shown any issue on any harwdare on windows, probably because is so much more lightweight than HDRP in its base. Only on HDRP I had reports of the issue and only for DX11 in some Nvidia cards (seems to work on both DX12 and Vulkan so far, only DX11 is bugged)

 

Thanks a lot - switching to DX12 makes a huge difference.
Now I can enjoy the advantage of GI in HDRP and explore the possibilities offered by LUMINA.
Thanks again.

 

Also make sure to download the latest version from store, as this has the dx12 enabled system.

 

Great , let me know for anything else.

 

Just one more question.
If the scene is too bright after adding LUMINA HDRP prefab to it, how to make it darker?
Right now I am decreasing the Weight in Volume LUMINA in the prefab added to the scene.
Am I doing in the right direction?

 

There is two modes that can make a difference in the overall brightness, when the Alpha in sky color is zero or above zero. When above zero creates an ambient sky light, so should also regulate everything lower or higher to adapt to it.

So first choose the alpha above zero to include the sky (for outside world for example), then regulate the sliders above and below and then can regulate the main GI intensity, near GI intensity and secondary bounce intensity and the occlusions to lower the overall light.

Note also there is a contrast value that should be set to one or near one and is for micro adjustment of overall contrast.

Is best to remove reflections when regulate the first GI, so can get a good look before reflections and then add them if needed,

 

Thanks a lot for the explanation.
All the best for you.

 

Lumina GI URP ($9 upgrade from Sky Master ULTIMATE currently) new demos WIP.

ARTnGame - Instagram - Facebook - Twitter - Discord - Youtube - Reddit - Discount Offers

 

This engine needs a lot of improvements. I tested here on my Ampere 3090 but the effect is far away from exaptation: GI is a mess (problem with refreshing GI, missing GI when should not, reflections have misleading options or work completely bad or I get bad effect. I understand this is a beta phase so I will wait for the next update

 

LUMINA HDRP





LUMINA URP




Lumina GI URP and HDRP, new demos with screen space reflections (HDRP native and URP with new system coming in next LUMINA URP version)

Hi, indeed the system is currently under lot of refinements to perfection, note that reflections are by default blurry ones and should combine with screen space, i work on this as well for both urp and hdrp versions, images of the wip above.

Which pipeline do you use ?

The update rate is controllable in time and space, plus on demand if need to optimize performance in specific ways. Also GI follows the transform declared in the script, so can be set to follow the camera to see GI everywhere or be set on demand as needed to cover specific areas. The distance limit can also be set in the LUMINA scripts on camera.

 

nasos_333 I use HDRP+DXR but I would like to see and test also Voxel system working in extreme environments as I do by DXR.
I will test your plugin after the next updates (need to see in my way, clean override GI to see how it works or what is the GI quality or limits)
Need to test like this:


Because I have Ampere 3090 I can really push it...but for that I need working proper plugin

 

Hi, sure, will be working on refinements and some tests with similar environment as shown and post asap here and to store.

Note though that a voxel solution is not meant to replace a raytracing approach that would be 100% accurate and pixel perfect, is more of an emulation of such with some closeness.

 

Of course that is understood, but maybe with good GPU hardware and extreme settings (more voxel density, optimizations, low amount of recurse, etc) can be really interesting alternative for DXR....especially when we talk about reflection, which are nightmare in DXR

 

Sure, can try this and see what can be done. What is the issues btw with the raytraced reflections ?

 

....the computation power
Simple, amount of data to calculate Global Illumination in real time (DXR as Interpolation method + denoiser) is so big that is no left free power to calculate reflections (only raytracing without GI) even on Ampere 3090.
Here: on my channel, you can see pure DXR with Raytraced reflection without GI (because there is not yet a solution for that in Unity and generally everywhere)
...yes that is amazing stuff when we talk about this quality and speed (around 20.000+ times faster than traditional CPU offline system) but still, reflection is only Ray Traced without Global Illumination.