Realistic light values and camera settings for day/night shots

[garipodelu]

Hello guys!
Our company workflow has slowly evolved the last few years. Being reluctant to change it took some time to get people used to the new way vray works.
Our library has a ton of custom models and shaders that we use in every project but now we are facing a new issue.
In our effort to make the whole process more plug and play we struggle finding the best settings for the lights, cameras and light materials so that you have to tweak a minimum of settings and get consistent results in the same scene for day and night shots.
To be more specific, after we render a day shot and everything looks ok, when we render a night version the lights are too bright or too dim and you have to spend time and try to adjust that.

I know that it's hard to achieve the perfect realistic result, especially if you are not using ies files, but I was wondering if you guys can share some tips from your workflow.

[joconnell]

So you've got three things that add up to a final exposure as you said, lights, materials and camera settings. If you look at a photo you can't be certain if something was very dark, the camera was underexposed or the light was dim. Likewise you could have a photo of a very bright room or someone may have just overexposed. To get accuracy you have to remove some variables from the equation and I reckon the cheapest was is to get a grey card from a photo store.

A "middle" grey in photo terms is 18% grey-ish and corresponds to a vray color texture with 0.18 as it's value. If you render this in our standard linear it'll have a mid grey diffuse. If you take an slr camera and your grey card out with you, you can put it into various night or day lighting scenarios and take a reference shot. If you meter off the grey card in your camera so it'll expose to make that middle grey, you'll have a good reference point. This is why you always see grey balls on vfx shoots, the silver one is so we get a mirror reference of which angles lights were coming from. Next when you're making a scene, you can have a look through your reference library and see if there's something similar to what you're making. If you look at the exif data of the image you've taken (lightroom, adobe bridge, xnview, irfanview and so on will show you this) it'll tell you what the camera settings were for that image. You can make a vray camera with the same settings, make a plane object with our 0.18 vray colour map as a diffuse and line it up roughly as it was in your reference picture. since you know the camera settings from your image, you know your grey card is 18%, the last variable to tweak is your lighting intensity and colour. Of course there's a bit more complication to this in that you should make your light sources at the same distances from the grey card in your 3dsmax scene as it was in real life, this also won't take into account gi so you'll only be in the ballpark, it'll still be pretty good though.

Another option is a lux meter which you place under a light (ideally only one light on so you're not taking readings of multiple lights in one go) and then hit a button to record the lux or brightness value of that light at that distance from it. Vray has a luxmeter helper so you can recreate this situation - make a light, make a vray helper and make sure it's the same distance between the two in 3dsmax as it was between your real life bulb and lux meter. The vray helper has a calculate button which will give you lux values and then all you have to do is keep tweaking the vray light's multiplier until you get the same lux value as you had in real life. A caveat is that most lux meters have a margin of error - it might be 10% margin on a cheap one (about 15 dollars on amazon) and 5% on something that's a few - again any of these is way better than just guessing though.

What big vfx places normally do is have a few very carefully calibrated look development scenes. They'd normally have a room in house where there's no artificial light getting in and a few lights that they leave in the same places and turn on / off manually. They recreate this in 3d or shoot a hdr of the room with the lights on and then measure and calibrate the hdr very carefully until a known reference object (normally something simple like a grey ball) can be rendered in 3d from that hdr and get a very close match to a reference photo of a real world grey ball. once this is verified, all materials are made and checked in something like this test lighting. Since it's a controlled room, as you get objects or materials you want to recreate you can simply bring them into that lookdev room, shoot some reference pictures with the same lights turned on and then start matching.

Lastly you can try to verify that the brightness of your materials is accurate. Something like a nix or cube sensor is a small device that you can hold against any surface, press a button and it'll give you the rgb value of the surface. It's not smart enough to separate what part is specular and what part is the diffuse colour but again it'll give you a good measure of how bright the material should be overall. What screws up renders mainly is people using diffuse values that are too bright or dark, then they have to use bad lighting values to compensate.

Food for thought and also a horrible rabbit hole to disappear down

[garipodelu]

Hello joconnell and thank you for your detailed response.
I was more interested in the workflow process of setting up the scenes and what is the most efficient way of doing things so that you minimize the time spent on tweaks.
I believe most people are eyeballing stuff and was curious if anyone found a better way of doing things.

[^Lele^]

Hello joconnell and thank you for your detailed response.
I was more interested in the workflow process of setting up the scenes and what is the most efficient way of doing things so that you minimize the time spent on tweaks.
I believe most people are eyeballing stuff and was curious if anyone found a better way of doing things.

In general if you have stuff misbehaving under different lighting conditions it's got to be an issue with the shaders, provided the lights and camera settings are in the right ballpark.
VRScans would take shaders out of the equation, if you could use them.

Other than that, as John said, you will need to get into the calibration realm, reproduce a few staple materials under controlled conditions (f.e. two or three plastics), and derive the new ones off those, without altering albedo (in other words: value) too much.
You'd then use a furnace/ganzfeld test to verify you are indeed energy preserving and around about the values you need for the new materials.

There is -to my knowledge- no way to reproduce just any material through principling alone, as most real-life counterparts exhibit very convoluted curves, with -often- very specific responses way off what one would expect.
Which is also one of the main issues of bad shaders: with production pressure mounting, and a fixed or quasi-fixed camera, or lighting scenario, one can quite quickly get to a "one-shot pony" shader, liable to break the heck down as soon as anything at all moves off the original balance.
It looked right, alas it was never.

Further to this, physical camera response in low light has got to be severely non-linear, and heavily camera/film dependent, so i am not sure you'd be able to find an exact match from within the render engine alone (ie. you may need post and the loathed tweaking anyways.).

It's a great rabbit hole to get lost into, i don't know why John's so negative about it: it's like an old pub we keep meeting each other at, decade after decade...
I may feel ever so slightly offended he doesn't like the place as much anymore. XD

[garipodelu]

Thanks Lele
Can you please elaborate on this statement "physical camera response in low light has got to be severely non-linear, and heavily camera/film dependent"? I'm not sure I fully understand what are you referring to.

In my office people are trying to make "presets" so that everything is more efficient but I'm not sure this is ok or that it can be achieved so I tried to do some tests.
By "presets" I'm referring to one camera for day shots, one for night shots, and a set of domes with HDRs that you just import into the scene and render.

During my experiments I came across this situation:
one image has the hdr with the multiplier set to one while the other image has the hdr's intensity set to 16 (this is how it was set up by the guy in charge with setting up the library)
As you can see the first image looks better than the second (at least for me).
So after doing this I realized that all the images made with this dome were incorrect because the hdr didn't have the right exposure to start with.

When using vray sun+sky it's easy to balance everything out but with hdrs I still find it hard to settle on a good multiplier value (and there are few of them with good quality).
I understand that if I'm making my own hdrs there are ways to measure the exposure and adjust things more accurate inside max, but when dealing with third party assets I still can't find a logical way of doing things.

[joconnell]

Ooh actually that's a useful thing - Lele if I'm not wrong, the vray scans materials will render after your license runs out, just with a watermark though? As in could we use them as a starting point of roughly what a wood / metal / plastic should look like in our lighting setup?

I'm still a huge fan of calibration, just finding the time to run through it has been tough. I've all the gear I need for it here though, just need to make the effort!

[^Lele^]

I was referring to DSLR which one could use for night shots as reference.
@joconnel , yep, that's the case!

[joconnell]

Should we do something like settle on a paint can manufacturer who ships world wide and does a neutral grey paint, spray a few paint chips and then have that scanned as a target? Of course paint manufacturers have slight differences in batches too but it might be a good base for a universal approach?

[^Lele^]

Ivaylo Katev is being summoned to answer: he'll know better what the picture looks like

[Ivaylo Katev]

Should we do something like settle on a paint can manufacturer who ships world wide and does a neutral grey paint, spray a few paint chips and then have that scanned as a target? Of course paint manufacturers have slight differences in batches too but it might be a good base for a universal approach?

what is the goal, to have a car paint base that can be colored?

[DanSHP]

Thanks Lele
Can you please elaborate on this statement "physical camera response in low light has got to be severely non-linear, and heavily camera/film dependent"? I'm not sure I fully understand what are you referring to.

In my office people are trying to make "presets" so that everything is more efficient but I'm not sure this is ok or that it can be achieved so I tried to do some tests.
By "presets" I'm referring to one camera for day shots, one for night shots, and a set of domes with HDRs that you just import into the scene and render.

During my experiments I came across this situation:
one image has the hdr with the multiplier set to one while the other image has the hdr's intensity set to 16 (this is how it was set up by the guy in charge with setting up the library)
As you can see the first image looks better than the second (at least for me).
So after doing this I realized that all the images made with this dome were incorrect because the hdr didn't have the right exposure to start with.

When using vray sun+sky it's easy to balance everything out but with hdrs I still find it hard to settle on a good multiplier value (and there are few of them with good quality).
I understand that if I'm making my own hdrs there are ways to measure the exposure and adjust things more accurate inside max, but when dealing with third party assets I still can't find a logical way of doing things.

We have had issues judging this too. Take a look at the HDRI pack below.

https://3dcollective.es/en/producto/...2pack-pro-eng/

Each HDRI gives you a specific multiplier value (8,16,32 etc). Those that do not have a multiplier value are ok set to 1. I'd recommend reading how and why these values are used for each HDRI, it's all there in the link provided.

So with the multiplier out of the way, it's just the camera exposure to adjust, and to ensure all materials aren't too light or too dark.

[joconnell]

what is the goal, to have a car paint base that can be colored?

My sneaky plan is so that anyone doing vfx can buy a can of the same spray paint, anywhere in the world, and create a grey ball that they can photograph as reference and then use the vrscan shader as a pretty close reference to calibrate their hdri's with. I know there's small variations between batches of a paint colour so this still some inaccuracy but it'd be better than nothing!

[jasonhuang1115]

My sneaky plan is so that anyone doing vfx can buy a can of the same spray paint, anywhere in the world, and create a grey ball that they can photograph as reference and then use the vrscan shader as a pretty close reference to calibrate their hdri's with. I know there's small variations between batches of a paint colour so this still some inaccuracy but it'd be better than nothing!

Chaos certified neutral gray vrscan...
I would totally buy the spray paint!

[^Lele^]

Sh*t, there go the royalties! ^^

[Ihno]

My sneaky plan is so that anyone doing vfx can buy a can of the same spray paint, anywhere in the world, and create a grey ball that they can photograph as reference and then use the vrscan shader as a pretty close reference to calibrate their hdri's with. I know there's small variations between batches of a paint colour so this still some inaccuracy but it'd be better than nothing!

Yes that's an awesome idea. +1