4.19 Physical Lights

[MysteryBlues]

Hi all,

I am following this thread since it started and also was playing around with the system. Eventually, I started to put together all the information in a comprehensive talk which I held at the Unreal Meetup Munich last week.

As I learned a lot by you guys (many thanks to all of you) I wanted to share my slides with you which should be a good reference for looking up lux values etc. You can find them here: Physically Based Lighting in Unreal Engine 4

For me, it worked best to have a mixture of the above mentioned workflows meaning using the Sunny16 chart for initial camera, sky and sun light setup (using the 35-125k lx range), then measuring the received light by using a white fully rough card or sphere and comparing it with the tables until I reach a correct looking light condition which matches with the values. Especially the breakdowns of sun and sky illuminance for some conditions in the lux table help a lot! Finally, I convert the camera settings to EV values for setting up auto exposure and applying some offset to the min ev value.

Probably worth mentioning that I was solely using fully dynamic lights for now.

Thanks for that! Very comprehensive quick guide.

[2car]

Bugs I've encountered while using the physical units, most are a result of the share amount of lux used and the needed support of the old unitless system:
...
Although shelved, LPV is blown as well due to required exposure and/or lux levels.
...

It seems to me LPV's break as soon as you start increasing directional light brightness ever so slightly.
At 128K lux you more or less only get primary colors. :/

[neighborlee]

A visual example of using a photographed HDR with photometric numbers in Unreal


Pixel Inspector luminance readings on white sphere:

• sun light only: 4500 (cd/m²) = 14,137 lux
• sky light only: 1770 (cd/m²) = 5,560 lux
• combined: 6150 (cd/m²) = 19,320 lux

These match (close enough to) the illuminace readings I took on location

• 18,900 lux in sunlight, facing sun
• 5,000 lux in shade, facing same direction
• sun+sky - sky = 13,900 sun intensity

My actual Sun Intensity is 18,000 lux because its color is very saturated, something else to take in to account.
Skylight intensity is at 1.0 "cd/m2", although I still hope they remove the misleading unit label.
Pixel inspector show the deepest blue sky at about 720.0 luminance (cd/m²)
The skydome material emissive had a multiplier of 23,170 or 14.5 EV.
The HDR middle exposure (mentioned earlier) was at EV₁₀₀ 13.32, so I guess you can't just take the EV number and plug it in to your skydome emissive intensity, but it's close-ish. Again the difference is probably due to color saturation. It's always good to have illuminace measurements to be sure, but with most HDRs you'll never know the camera settings or illuminace at time of shooting.

The results came out okay. I can use manual exposure or EV100 numbers and it looks pretty close to what I'd expect. Something else to be aware of is I'm not using pre-exposure, so numbers around 65,000+ will start clipping, (at least on my hardware they do).

Btw, 'what kind' of a camera is required to do that/are there TUTS for this procedure ? I suspect you're losing quite a few with this- artists won't have much of a clue( most) .
Bringing this into a POV for my own current 2x2 5101x5101 world comp level...only a few stair brushes basically , few trees-deffault sandrock landscape texture- one directional light while enough so far, there is one part of terrain that isn't working as intended, and given 'lighting isn't artist friendly atm, how do I go about getting THIS picture to look like this editor image ?

Thx

[neighborlee]

It seems to me LPV's break as soon as you start increasing directional light brightness ever so slightly.
At 128K lux you more or less only get primary colors. :/

Looks like thread is VERY dead, wonder why , so did you ever get your lighting and if so what was workiing for you ?

thx

[2car]

Looks like thread is VERY dead, wonder why , so did you ever get your lighting and if so what was workiing for you ?

thx

No I have not gotten it to work.
At the moment I think we will have to choose to go with either physically based lighting or LPVs. As the two cannot be mixed.
I wish someone would implement DDGI (aka nVidia RTXGI) or something similar to what Unity is currently doing. LPVs just dont seem to be very future proof.


https://developer.nvidia.com/rtxgi-technology

[neighborlee]

I've tried LPV not seeming to do much https://docs.unrealengine.com/en-US/...mes/index.html , but I'll have another look. I have a semi large 2x2 world comp world,,and lighting there is almost impossible atm, VERY inconsistent and I still can''t get all areas right as some look decent and others completely washed out with light, fake looking- and very dim shadows to boot. Too complex atm, too many settings imho anyway,,,hard to know even how they all interact!

Physically based lighting HM but not great , and available as 4.22 but very basic stuff from what I'm seeing.

RTX would be nice, I assume you mean raytraced,and that's fine, but I won't be able to use it atm as yes I'm very near to being able to afford a gtx 1060 6gb which is now (with new drivers) a MINIMUM to do all of this, but even then you don't get everything !!

So no, that's going to leave out a huge swath of artists and others, but then of course those are the breaks with lighting .
Cheapest I kind find and atm anyway outside my ability to pay for https://www.newegg.com/msi-geforce-g...-231-_-Product ,and THAT is refurbished o_0 . Yes what I'm doing is on the budget of someone who is disabled , but of course Seismic <grin> . My dev budget is first likely going to be for an AM4 system and max cores so lighting won't be so grunt like.

This, https://www.newegg.com/asus-geforce-...-257-_-Product , seems to be a minimum for non refurbished , cheapest I can find but eventually here doable. LOL, the last card I paid almost that much was an awesome Gtx 8800 , & my current GTX 950 2gb cost me like $119.00, easy.

[rosegoldslugs]

LPVs don't scale well with high brightness values it seems. I managed to get them visible at least, but even in lower arbitrary values if the LPVs are increased separately they still cap out at white. Need to dig some more.

RTXGI would be great, but I'd take any improvement over Volumetric Lightmaps to be honest. Like decoupling them from requiring Lightmaps/Static Lighting permutations, more control with unique volumes(with/without runtime raycasting like in DDGI), etc.

Also neighborlee, if you plan on doing anything with RTX you'll need something much higher than a 1060. Yes it's now DXR-enabled, but you can't really do anything useful- it's more for testing.

[2car]

LPVs don't scale well with high brightness values it seems. I managed to get them visible at least, but even in lower arbitrary values if the LPVs are increased separately they still cap out at white. Need to dig some more.

RTXGI would be great, but I'd take any improvement over Volumetric Lightmaps to be honest. Like decoupling them from requiring Lightmaps/Static Lighting permutations, more control with unique volumes(with/without runtime raycasting like in DDGI), etc.

Also neighborlee, if you plan on doing anything with RTX you'll need something much higher than a 1060. Yes it's now DXR-enabled, but you can't really do anything useful- it's more for testing.

As far as I understood from the "Ray traced irradiance fields" (that is the basis of RTXGI) presented at this years GDC, RTX was not a requirement. It would just be able to cast more rays within the allotted ms budget.

[rosegoldslugs]

As far as I understood from the "Ray traced irradiance fields" (that is the basis of RTXGI) presented at this years GDC, RTX was not a requirement. It would just be able to cast more rays within the allotted ms budget.

I believe you're referring to 2 separate GI methods from NVidia powered by RT cores. It's confusing because NVidia calls everything RTX, when that's the hardware/brand name, and they purposefully change names to resemble RTX(i.e. see the Square Enix path traced demo being referred to ray tracing with RTX... two separate things and fully path traced images are way more impressive, but there's no "PTX" to market right now) Btw did you get to work on that?

RTXGI is the standard method of sending rays to the scene with x samples per pixel
DDGI is another method that uses raytracing on top of previous methods of Spherical Harmonics/probes to improve visibility/light leaking, dynamic updates, etc

I actually double checked on this and NVidia's main page changes Morgan's writeup from DDGI to RTXGI:

NVidia: https://devblogs.nvidia.com/rtx-glob...nation-part-i/
Morgan McGuire's site: https://morgan3d.github.io/articles/...dgi/index.html

[neighborlee]

As far as I understood from the "Ray traced irradiance fields" (that is the basis of RTXGI) presented at this years GDC, RTX was not a requirement. It would just be able to cast more rays within the allotted ms budget.< yes.

[2car]

I believe you're referring to 2 separate GI methods from NVidia powered by RT cores. It's confusing because NVidia calls everything RTX, when that's the hardware/brand name, and they purposefully change names to resemble RTX(i.e. see the Square Enix path traced demo being referred to ray tracing with RTX... two separate things and fully path traced images are way more impressive, but there's no "PTX" to market right now) Btw did you get to work on that?

RTXGI is the standard method of sending rays to the scene with x samples per pixel
DDGI is another method that uses raytracing on top of previous methods of Spherical Harmonics/probes to improve visibility/light leaking, dynamic updates, etc

I actually double checked on this and NVidia's main page changes Morgan's writeup from DDGI to RTXGI:

NVidia: https://devblogs.nvidia.com/rtx-glob...nation-part-i/
Morgan McGuire's site: https://morgan3d.github.io/articles/...dgi/index.html

Yeah, the RTXGI and DDGI I am referring to is the same thing. They just changed the naming to get "RTX" in there, but I think it just makes it more confusing.

from your first link:
"We first presented RTX GI as Dynamic Diffuse Global Illumination (DDGI) at GTC’19 and GDC’19 in the NVIDIA sponsored sessions."

No, I did not work on the path traced demo. I was on Kingdom Hearts III and FF7R.

[klusky27]

Hello everyone!

I hope this thread is not dead
Sorry if you already answered my questions but I really tried to read all the responses!

So now, are some of you actually working with physical lighting values? Do you still have issues with the exposure and with the material reflections?

I would like to use this physical 'method' in the project I am working on. I am working on huge environments (big caves with high ceilings) and I tend to find reaalllly difficult to have indirect lighting from my lights on the players. Right now I am using a 3 lux intensity for my directional light and 1 for the skylight (which captures a skybox whose texture brightness is set to 1).
Even though I can achieve good lighting for my environment (also by adding some fill lights here and there), the volumetric lightmaps are suuuuper dark, so the players are also really dark. They don't fit with the environment at all, and I don't want to add more light since I want to keep a mysterious and dark mood.

My first thoughts were that I use really low values for my sky and sun and all the other lights, so they don't create so much indirect lighting.
I tried using physical lighting values on a test scene: I increased the sun to 100 000 lux, increase my skybox texture to 5000 cd/m2, and changed the exposure mode to 'manual' with the proper settings in the camera so that it is not super bright. I baked the light, and looked at my volumetric lightmaps, and theeen it made a lot more sense. They were not black anymore and fitted well in the environment. With the same test scene, I did the lighting using the other method (sun = 3 and sky = 1) and achieved approximatively the same look, but after baking, the volumetric lightmaps were dark.

So I guess using physical lighting values may help! But I don't wanna f*ck the whole project and the materials and stuff if there are still some issues regarding this way of doing.
Would you advise me to go with that? If no, how do you do to make your players brighter when you don't wanna make the environment brighter? Maybe I do something wrong in my lighting ...

Thank you a lot for your response! I am struggling soo muuuuch with that

[rosegoldslugs]

Physical values shouldn't really change much, you can visually get the same result without physical values - they just make it more consistent/easier. When you set up your arbitrary values, how close did they match your physical setup? Did the sky change other than the intensity? And, how was your exposure? A lot of people skip the exposure step, which has a huge impact on all lighting.

Regardless of how you do the lighting, there isn't a simple way in stock Unreal to boost characters only other than material tricks or character-only lights with lighting channels.

[klusky27]

Thanks a lot for your response!

Here two screenshots of the test I did on a scene I took from the market place.



On the left, it is from the non-physical lighting values, and on the right, it is the physical lighting values.

'non-physical' method settings:
Dir intensity : 4 lux
sky light intensity : 1
HDR texture (captured by the sky light) : brightness of 1

I set a fixed exposure that looks good to me for the interior of the train.
Exposure : Set to Auto Exposure Basic
Min brightness : 0.2
Max brightness : 0.2


'physical' method settings:
Dir intensity : 128 000 lux
sky light intensity : 1
HDR texture (captured by the sky light) : texture multiplied by 25000 (to have a luminance around 5000 cd/m2 in the clouds). But other than the brightness, the texture is the same.

Exposure : Set to manual
Camera settings :
Shutter speed : 100
ISO : 100
Aperture : 6.8


Everything else is the same (lightmass settings etc..).

The exterior lighting looks the same also on the two methods and both contributions of the sky and the sun look the same. Other than the emissive that don't work with the physical values (I guess I have to increase a lot the emissive's values), I see some other differences. In general, physical values give more accuracy and realism I think...

But I agree that the overall brightness of the character doesn't change drastically... So going with the physical values in my project won't really solve my problem of having dark characters... I will need to cheat on that.

[Kel Solaar]

I'd also like to get some more info about this exposure formula (Exposure = 1 / (1.2 * 2^EV100)).
Where exactly does this come from? in particle the 1.2 value?
I assume by exposure here you mean the average scene luminance in cd/m^2 that the game will adjust to?

For future reference, we spent a few hours digging and found out where the 1.2 value is coming from: It is actually normative as per ISO12232:2006, i.e. an ISO Standard derived value, referenced in Wikipedia, in Moving Frostbite to Physically Based Rendering 3.0 (and confirmed by that bit of maths in Unity code):

// Compute the maximum luminance possible with H_sbs sensitivity
// maxLum = 78 / ( S * q ) * N^2 / t
// = 78 / ( S * q ) * 2^ EV_100
// = 78 / (100 * 0.65) * 2^ EV_100
// = 1.2 * 2^ EV
// Reference: http://en.wikipedia.org/wiki/Film_speed

Cheers,

Thomas