What is the Native White Balance in BM cameras?

Is there any tech info what is the real "native" WB setting for different BM cameras? It is known that most sensors are adjusted for Daylight but is there some specific Temp/Tilt settings that used as true original White Balance starting point in BM cameras RAW controls? Is it 5400, 5600, 6500, or maybe some specific Temp/Tilt setting?

I just attempt to extend my workflow and in addition to Exposure (Gain in Linear gamma) i try to figure how to adjust WB in ProRes exact same as WB in RAW. From my quick tests and other people suggestions it seems perfect starting point require:
- Use native scensor WB setting.
- Adjust RGB Gain (or Temp/Tint tool in Primaries) in Linear gamma and in native camera color space (or in ACES color space or in XYZ color model).

As I understand it's best "balanced" at at around 5000K. Again speaking as not an engineer, just a lay person, but this the white point where the RED, GREEN and BLUE gains are all at the same levels.

Also, as we discussed in the other thread, colour checkers are important references if you're doing profiles, but they also only represent a small handful of tones.

I've found after many years of doing deep looks at sensor and colour reproduction, the light you use can GREATLY affect the result of any colour critical assessment. I also try to have skin tone as well.

I avoid LED lights all together because they're often missing some tones, or deficient. If the light isn't being emitted by the light source, how can that colour then be reflected by the object ? (the Ryobi thread).

It's EASY to test this as well.

I now use tungsten lighting GELLED to be the WB light I want to test. So If I want 5600K I use tungsten lighting and correct if with CTB to get to 5600K.

That way I'm getting one of the most flat and KNOWN light sources at the white point that I want.

Then I know the colours being reflected back are what the sensor is seeing, not what the light source is emphasising.

It's worth it to invest in a good colour temp meter, one that's suited to LED and discontinuous light sources.

I like the Spectra C800. It has TCLI and SSI, which are both much more accurate indicators than CRI (Which it also has)

https://c800.sekonic.com/

JB

With some WB RAW vs WB Temp/Tint tool in Primaries in linear gamma compare tests i figure that for BMMCC WB starting point is 5500K or 5600K

So if we keep ProRes source at 5500K or 5600K and at native ISO800, we can adjust WB and Exposure in Resolve exact same as if it was RAW.

It works!

i remember this from Captain Hook about cst plugin
viewtopic.php?f=2&t=110298#p609487
may be related on bmd sensors native wb

As an Arri engineer told me, it's a property of silicon sensors.
They all have their native optimum around 5,000 K. So, if you use daylight around 5,500 to 5,500 K, it's all good. Tungsten without gels will starve the blue channel (and to a lesser degree green) of light and you'll get much more noise by trying to compensate for that.

LED lighting is very difficult since even the best ones out there don't have a true continuous spectrum.

Yes, yes. Tungsten light always better to use blue gels or with 80A 80B lens filter. My example with Tungsten is for test only. Color reproduction is far from perfect without blue filters. It just illustrates match between WB RAW vs WB Temp/Tint tool in Primaries in linear gamma when source Temperature was set to 5500K.

Last image just illustrates how ColorChecker may help to fix colors in complicated situations.

Some additional tests:




Sounds insane, but WB Temp/Tint tool in Primaries operated in ACES or XYZ color space and Linear gamma produce even better result than original WB RAW controls. Temp/Tint tool in Primaries is rather limited, so larger color spaces require 2-3 node copies of same tool to adjust WB far enough.

And some color corrected DNG vs ProRes422 tests. Perfect WB match.

I avoid LED lights all together because they're often missing some tones, or deficient. If the light isn't being emitted by the light source, how can that colour then be reflected by the object ? (the Ryobi thread).

It's EASY to test this as well.

I now use tungsten lighting GELLED to be the WB light I want to test. So If I want 5600K I use tungsten lighting and correct if with CTB to get to 5600K.

That way I'm getting one of the most flat and KNOWN light sources at the white point that I want.

Do you also avoid HMI lights? If so, and you need high output, what do you use?

Do you also avoid HMI lights? If so, and you need high output, what do you use?

HMI's are what you have to use sometimes for output. Nothing beats them for their efficiency.

HMI's are great for a night exterior. When you need to paint a forest or a landscape.

On stage however, I tend to stick with tungsten. On my last show I uses dozens of Arri T12's (a 10K tungsten fres) and for more output I like the 24K Dino / Maxis (24 x 1 K lamps).

Catherine's bedroom in The Great is lit by three 24K Dinos for level. I then use 10K Molebeams for hot sunlight accents. Mole beams are incredible for "projecting" a beam of light across a set.

So here's my lighting list from early on for the show.(it got changed a bit) I had to light across four stages so each stage has it's own full time package.

https://www.dropbox.com/sh/kmave8q60wqb ... fQlNa?dl=0

Then there's a travelling stand by lighting package I'd take to each stage. There's also in a separate truck a full time location package for when we go out to a location, that's the equivalent to the stand by package on stage. It's presumed that IN ADDITION to this package on location there would be a pre-rigged package. In episode 5 and 6 I had a lot of night exteriors with carriage scenes traveling for real and that took a dozen Arri M90's and half a dozen cherry pickers.

So HMI ? Good for Night exteriors. Sometimes good for a Day interior on location when you're mixing with daylight. But my location daylight package is pretty small, because like a lot of Australian DP's, we tend to use a lot more bounced and overhead's rather than lighting with HMI's on day exteriors.

JB

If you look at the spectral distribution, nothing beats Tungsten for consistency!
Even natural daylight has some spikes and lows.

I am not sure what you mean by daylight spikes and lows since the sun covers a full spectrum but of course time of day and weather conditions make daylight a very inconsistent light.

So I’m sure almost no multi-kazillion feature would rely on daylight without ‘lighting’ the scene for predictability. Exception being those who closely watch for the colour of the natural light they want to capture for their aesthetic preferences.

Our sun is a star like any other, and you can even analyze its spectrum for physical processes and the elements represented in it, as astronomy does. In open space it would even have discrete bands in the spectrum.
It's our atmosphere which helps to smooth it to some degree, but it's by no means continuous.

Have a look here: https://www.esa.int/ESA_Multimedia/Imag ... r_spectrum

So any further info from BM tech support about native sensor color temperature? Is it 5500K? Is it the same between cameras based on Fairchild Imaging and Sony sensors?
Another question - is it any real life cons about transforming to XYZ color space (color model) and back with Resolve CST node? Any warnings about possible image degradation? Technically XYZ is not a Color Space but a Color Model different from RGB. It can formally include infinity color gamut, but usable only for some special tasks.
Temp/Tint in XYZ/Linear gamma works really nice, but RGB gain in XYZ/Linear gamma produce slightly different, more "overloaded" look. Not sure why.

Here are some workflow examples to understand the evolution of this concept:

1. BMDFilm source -> WB, Exposure(ISO) in RAW -> Transform to REDWideGamutRGB/REDLog3G10 -> Noise Reduction, Gain, ColorChecker correction, Contrast -> Transform(Compress) to Rec709 -> Film emulation LUT, Sharpen
Cons:
- WB and Exposure (same as ISO) adjusted in RAW (no keyframes support, no different WB/Exposure settings for different parts of same clip)
- WB/Exposure adjusted inside native "limited" camera color space.
- It is impossible to apply Noise Reduction before WB/Exposure. Noise Reduction applied after huge WB/Exposure adjustments to already amplified noise will always produce lower quality result.
- Workflow is not directly usable for ProRes sources. ProRes Log sources require baked in-camera adjusted WB and ISO.

2. BMDFilm source -> WB in RAW -> Transform to REDWideGamutRGB/REDLog3G10 -> Noise Reduction -> Transform to REDWideGamutRGB/Linear -> Exposure (Gain in linear gamma) -> Transform to REDWideGamutRGB/REDLog3G10 -> Gain, ColorChecker correction, Contrast -> Transform(Compress) to Rec709 -> Film emulation LUT, Sharpen
Cons:
- WB adjusted in RAW (no keyframes support, no different WB settings for different parts of same clip)
- WB adjusted inside native "limited" camera color space.
- It is impossible to apply Noise Reduction before WB. During large WB adjustments color channels will produce different amount of noise.
Pros:
- Exposure adjusted outside RAW controls (full keyframes support, different Exposure settings for different parts of same clip are possible).
- Noise Reduction applied BEFORE huge Exposure adjustments to original non amplified noise will always produce higher quality result and cleaner image.
- Workflow is partially usable for ProRes sources. ProRes Log sources require baked in-camera adjusted WB and untouched native ISO.

3. BMDFilm source -> Noise Reduction -> Transform to XYZ/Linear Gamma -> WB (RGB Gain in Linear gamma), Exposure (Gain in Linear gamma) -> Transform to REDWideGamutRGB/EDLog3G10 -> Gain, ColorChecker correction, Contrast -> Transform(Compress) to Rec709 -> Film emulation LUT, Sharpen
Pros:
- WB/Exposure adjusted outside RAW (full keyframes support, different WB/Exposure settings for different parts of same clip are possible).
- WB adjusted inside unlimited XYZ color space (color model).
- Noise Reduction applied BEFORE huge WB and Exposure adjustments to original non amplified noise will always produce higher quality result and cleaner image.
- Workflow is 100% usable for ProRes sources. ProRes Log sources require baked in-camera untouched native WB and untouched native ISO.

[CaptainHook]

As an answer to the original question of this thread, here's what I wrote to a similar question on BMCuser back in 2016 :

The cooler your light source, the more red (vs blue) gain you need. The warmer - the more blue (vs red) gain you need. So in terms of red+blue gain they end up adding up to the same as you change colour temp,

Image sensors in general are more sensitive to green light, so when a scene has equal amounts of RGB light - you actually need red and blue gain to rebalance it. Going for a light source that's as magenta as possible will mean less Red and Blue gain is necessary. In that situation you would set a tint value (speaking in FW4.0) to -50.

In terms of where red/blue would be basically equal in gain you're looking at :


URSA Mini 4.6K - around 4000K (-50 tint)
URSA Mini 4K - around 3800K (-50 tint)


But at the same time remember that the blue channel tends to be noisier so you may prefer a higher temperature where you gain blue less and the amount of red+blue gain would still be similar since you just lower one while increasing the other.

-----

What's your goal, to reduce noise?

Go for a cooler light source (like HMI's or even Kino's which are often closer to 6500K) and then add minus green - so the light source is blue and magenta-ish which means you don't have to gain up the red/blue channels too much. The other thing is i've noticed is in general CFA's on sensors tend to have "better separation" with daylight light sources versus tungsten, so you'll likely get more "cleanly separated colour". On the hand, many think human skin responds arguably nicer under tungsten light sources so it's totally up to what you're trying to achieve.

If your definition of "native white balance" is the point where the RGB gains are equal then yes it varies per sensor. I could look at calculating the numbers for the P4K/P6K if people really want to know but I don't think it matters much personally (I think they will be closer to 3200K compared to the 4.6K being around 4000K as mentioned in my quote above). Ideally you want to white balance in "sensor/camera space" (adjust channel gains), and then you would use an interpolated matrix (white balance specific) to transform from sensor/camera space to a more common space like XYZ/Rec.709/BMD Wide Gamut/etc.

Doing white balance in XYZ space is not as ideal but it's perfectly fine as a substitute or second choice if you don't have the information or ability to get into sensor space. Assuming all of this is also in linear.

If you are using Gen 4 colour science than using the CST node to go into XYZ is generally fine too, but Gen 3 or lower and the Resolve CST plugin assumes your source white balance setting is 6000K (a somewhat arbitrary value chosen by the Resolve team). The CST plugin also doesn't do any chromatic adaptation for your white point if you are to change into a space with a different one, but they recently added a seperate plugin specifically to handle white point adaptation.

Blackmagic RAW simplifies all this as in the SDK we do everything in the background including chromatically adapting the white point for your target gamut/space etc if needed.

[CaptainHook]

I am not sure what you mean by daylight spikes and lows since the sun covers a full spectrum but of course time of day and weather conditions make daylight a very inconsistent light.

Compare the "smoothness" of Tungsten here compared to daylight (taken from https://indiecinemaacademy.com/understa ... rocedures/)

(Click the image to expand)

STANDARDS-tungsten-daylight.jpg (150.1 KiB) Viewed 8011 times

Thanks CaptainHook. That info helps a lot.

If your definition of "native white balance" is the point where the RGB gains are equal then yes it varies per sensor.

Under "native white balance" i imagine some specific color temperature setting that used as starting point in RAW WB processing. Not sure if it works exact like this.

As you may see from earlier posts, based on some tests i came to color temperature 5500K for BMMCC:

Test 1:
I used different DNG sources shoot under different light conditions and attempt to adjust them to neutral grey patch.
I set RAW WB to 5500 -> Transform to BMDfilm color/Linear gamma -> Adjusted RGB gain.
This gives me 99.9% exact same result as WB adjusted in RAW.

Test 2:
I set RAW WB to 5500 and in BMDfilm color/Linear gamma i try to shift temperature to very cool and very warm side. Colors always look similar to extreme cool/warm WB adjustments in RAW.

P.S. I guess some very tiny variations because some tiny mistakes in manual RGB adjustments and because some chromatic adaptation shifts in CST node, which as you explain expects 6000K for processing for BMDFilm.

Test 3:
When i set RAW WB to value different from 5500, and do the same tests - WB adjusted in BMDfilm color/Linear gamma never match to WB adjusted in RAW. Color patches looked very different.

From this point i decide that RAW processing use some "native white balance" starting point described in metadata (similar to native ISO). And to adjust ProRes WB in exact same way as RAW WB, ProRes source file also should be recorded in "native white balance" in-camera.
Maybe it is not a sensor, but some "native white balance" point of Color Space used by each camera.

I don’t know if it’s a Resolve bug or maybe I’m misunderstanding the math. If you put a CST in a linear node and start messing with the gain the image quickly falls apart. If you separate the functions into discrete nodes it behaves normally. It’s like there’s something within the process that grounds the floating math and bakes the gamma/gamut into the image.

Maybe try and separate the nodes and see how that affects your results.

Good Luck

Did you change node gamma with right click? Stay away from that quick option because it is not always work as expected. It is not compatible with tools like ColorChecker.
I recommend use only normal CST dual node chain:
[log to linear]->[Gain adjustment]->[linear to log]

What was the result when recorded at something other than 5500k?

Good Luck

Loving this discussion. So far, I didnt realize the importance of WB for the sensor and gain of each color channel.
But always preffered to shoot daylight or tungsten because you can really see they show its true colors. Pun intended.

[CaptainHook]

Maybe it is not a sensor, but some "native white balance" point of Color Space used by each camera.

I haven't read all your steps fully, but maybe what you're seeing is from most common gamuts/spaces having a D65 white point (since most consumer displays are also D65) including BMD Wide Gamut/Rec.709/etc. Have you tried with 6500K? By the way, at higher colour temperatures you can make much larger changes in kelvin without too much of a visual difference. But closer to tungsten smaller kelvin changes are much more noticeable.

I will post some further tests.
Here are cool and warm original DNG files if someone want to play and compare. I shoot those tests long time ago on BMMCC without OLPF filter, so there is some moire and IR pollution.

I will post some further tests.
Here are cool and warm original DNG files if someone want to play and compare. I shoot those tests long time ago on BMMCC without OLPF filter, so there is some moire and IR pollution.

Thanks Dmitry, color charts are my Candy Crush. Speaking of white balance, has Resolve’s temp tool always tracked wb properly? I feel like back in 14 it was straight red/blue bias and now it curves properly. Maybe I’m confusing Resolve with a camera I was testing.

Good Luck

Here is another test example with my universal daylight test shoot. This time to adjust WB i use RGB gain in linear gamma with Lum Mix set to 0.0.
Original DNG in attachment.
Please note, this and all other tests done in simple YRGB project.

Reference uncorrected look: WB in RAW set to 5500K -> Transform to REDWideGamutRGB/REDLog3G10 -> Contrast 1.2 -> Transform(Compress) to Rec709


WB in RAW set to 2500K -> Transform to REDWideGamutRGB/REDLog3G10 -> Contrast 1.2 -> Transform(Compress) to Rec709


WB in RAW set to 5500K -> Transformed to BMDFilm/Linear Gamma -> WB (RGB Gain R0.65 G1.03 B1.66) -> Transform to REDWideGamutRGB/REDLog3G10 -> Contrast 1.2 -> Transform(Compress) to Rec709


WB in RAW set to 8500K -> Transform to REDWideGamutRGB/REDLog3G10 -> Contrast 1.2 -> Transform(Compress) to Rec709


WB in RAW set to 5500K -> Transformed to BMDFilm/Linear Gamma -> WB (RGB Gain R1.15 G1.00 B0.84) -> Transform to REDWideGamutRGB/REDLog3G10 -> Contrast 1.2 -> Transform(Compress) to Rec709

Some further updates:

1. To match Temp/Tint in linear gamma to RGB gain in Linear gamma,
RGB gain in Linear gamma should be adjusted with Lum Mix set to 0.0

2. If look very close on pixel level there is still some tiny difference between RAW and non RAW WB adjustment quality. RAW WB adjustment always looks cleaner in some color patches, produce less aliasing-like rainbow artifacts and same time better preserve fine texture. It probably means that WB in RAW adjusted somehow on different level before debayering to RGB.


3. If use RGB gain in Linear gamma with Lum Mix set to 0.0 there is no any difference in colors between 2000K, 5500K or 10000 WB RAW starting point adjusted back to normal with RGB gain in Linear. (i apply Contrast 1.2 to these last examples)

Loving this discussion. So far, I didnt realize the importance of WB for the sensor and gain of each color channel.
But always preffered to shoot daylight or tungsten because you can really see they show its true colors. Pun intended.

+1

Great information here, thanks for taking the time to share it.

Now let's switch DNG source and CST nodes to ColorScience Gen4.
As you may know in ColorScience Gen4 Native sensor color space always transformed under the hood to smaller unified BMD WIdeGamut Gen4 color space. This additional pre-processing provides better balanced colors for start, but as side effect in ColorScience Gen4 you have no access to native sensor color space.


It appears that with ColorScience Gen4 It is impossible to match WB adjusted in RAW to WB adjusted in Linear RGB Gain
WB adjusted with RGB Linear Gain in ColorScience Gen4 produce huge shifts in colors. (these examples processed with slightly increased Exposure and Contrast 1.1)