Personal View site logo
Make sure to join PV Telegram channel! Perfect to keep up with community on your smartphone.
Daylight vs. Tungsten, and the winner is...
  • Out of curiosity I decided to compare Daylight to Tungsten to see if one was superior to the other, fully expecting Sunlight to win out. I instead discovered an odd luma/chroma shift between 5500K and 3200K even when white balanced in cam.

    And I was curious to see if Daylight resulted in less noise in the shadows... Well, it resulted in the opposite.

    To Sum up: After extensive testing - The Chroma and Luma shift demonstrated in the video occurs regardless of hacked or unhacked firmware, regardless of patch, and regardless of WB method whether it be standard presets, manual sampling WB, or dialing in Kelvin number. And it occurs regardless of whether it is actual sunlight or CTB. Unless someone can offer evidence to refute this, I think this is now a known issue.

    In conclusion: Daylight does not render accurate colors on the GH2, whereas Tungsten does.

    I've also edited this post to reflect the conclusion I came to after further testing, and also altered the title of this thread to help newcomers to the thread.

  • 43 Replies sorted by
  • Audio fixed

  • @shian I know you mentioned noise issues when setting manual white balance on the GH2 but have you ever used a WhiBal card or similar calibrated white balance reference card for manually setting the WB on the GH2? It appears many conventional gray cards are not totally color neutral.

    I'd be interested to know if something like a WhiBal card for manual WB would still give the discrepancies you are seeing between the GH2's daylight and tungsten settings.

  • if you check the description on vimeo you see that I WBed by dialing in 5500 and 3200 K and not by sampling gray card.

    I balanced the exposure by making sure it was at 50% in FCP. That's what I meant by balancing for 18% gray.

    The white balance was not touched other than dialing it in.

  • @shian If the GH2's Daylight and Tungsten preset settings are giving you such different results then I'd try something like a Whibal card and perform a custom WB for each light source instead of dialing the WB. If the color discrepancies still exist then...???

    I know you are reluctant about using custom WB with the GH2 because of noise but it may be worth doing a test in this case in order to see what's going on with the color shift.

    FYI I use a Sekonic L-358 in incident mode and an 18% gray card for exposure. I'm well aware of the 12% versus 18% exposure difference. I ceased using gray cards for WB some years ago as I found different gray cards gave me different WB readings and that's why I bought a WhiBal card.

    I'm also no expert in the field of video color correction. My background is photography (I'm new to this video caper!)

  • that's my next test...

  • I did not watch the video - but are you using Any Canis Majoris settings? These are known to use different formulas in the matrix than "normal" to achieve different results in different lighting.

  • nope - Quantum 9B

  • @shian Couple of things to think about.

    Do you have a colour temperature meter ? Because simply putting CTB on a tungsten light often won't get you 5600K. You have to meter it. It can also CHANGE during the day as heavy industry turn on and off the load on the electrical grid around you. You can get +/-30v on 240V near where I live. That can change the CT during the day as you're testing, even if you're metering it.

    And for that matter, most of the time, uncorrected tungsten lights aren't ever 3200K in my experience. Variance in the age of globe, dust on the reflector and the operating voltage being not what it should be will all affect the CT of the light.

    I just did some similar tests with the BMCC and the Tungsten lights I used were sitting at about 3000K. I ended up using 1/8 CTB on one lamp and 1/4ctb on the other to get it to 3200K at the target.

    You should also try to diffuse the light. Although it might look like the light has an even field in terms of brightness, it rarely is across the field of the light it's projecting. Bouncing it and using multiple lights bounced then going through diffusion should even that out so that you know the exposure is even across the target.

    So I use two lights and then bounce them into a 12x12 ultra bounce and then put that through a 12x12 1/4 grid. Also known as a WEDGE or BOOKLIGHT. Very flattering too for people FYI.

    Bouncing light like this also warms them up, so again, this is where you need a CT meter to confirm you're getting what you think you're getting.

    You also mentioned you used some "scrims" to "knock it down". So did you use scrims in one setup but not the other ? This will also affect the CT of the light shot to shot if you did this.

    You should try this with multiple cameras to also eliminate variances in the camera setups themselves. Ideally you'd shoot two at the same time so you can then also match their operating temperatures (having turned them on at the same time)

    You might want to consider looking at what OPTICAL CT correction filters do to the image instead of dialling in the electronic CT correction. Going the other way, many find they can get better results, especially with regard to noise by using an 80C or similar filter on a camera in daylight CT using 3200K lighting. Don't forget to allow for the exposure correction.

    Many sensor CT native temps are 5000K, not 5500 or 5600k. Maybe look at shooting at that electronic setting as well. It might be the 5500 and 5600 settings are fudges.

    It's interesting, although I wouldn't expect ANY camera to be perfectly matched doing this kind of exercise.

    There are so many variables. I also find that the charts themselves are very hard to get consistent results from. They can be affected SO MUCH by the precise angle you hold them at relative to the camera and the light source. Try it with a grey card while you're metering or shooting and watch the values jump hugely.

    Now I think the macbeth is an OK reference, but I've always found a grey card is more meaningful, because it has both grey and colour.

    Also, take a look at what is probably the worlds worst website selling what I think are the world's best test charts. This guy makes charts mainly for machine vision cameras but also some photo applications. He has a nifty grey card that has colour opposite spots that when summed together make GREY, but they can more easily reveal a magenta - green shift. So many don't realise you can still hit a WB of 5600K but be massively GREEN or massively MAGENTA whilst still being spot on 5600K. WB only shows the white point, not the evenness of the spectrum. These cards reveal if there's a colour bias in your lighting. In other words, they more accurately reflect the evenness of the spectrum your test lighting is emitting because if a colour dot disappears or is more visible, then you know you have a spectral bias in the light you're using to illuminate the target.

    If you're in the US and don't want to try and order with the mainly german website then try the DOP shop.


    (and why is it that you yanks love 5500K and everyone else is 5600k ?)

  • @johnbrawley Thanks good stuff to think about and try.

    I did use diffusion on my first test, but had to move the light position to match the EV readings. I wanted to eliminate that factor. (I did still observe the same luma and chroma shift)

    I also did one shot at 5600 and another at 5700 to see what difference it made -- none, but since I did not run through the exposures with them, I did not include them in this demo. ( They also had the same shifting)

    I used actual sunlight both direct and indirect vs. tungsten and got the same thing. (It probably made no difference, but I obviously had to stop down considerably for the direct sun)

    The only thing I haven't tried is manual white balance on a white card as well as a gray card, I will try that this week some time. I have avoided it, because of a noise issue that I first noticed back in January, which might just be my particular camera. I'm hoping to test it on other GH2s this week.

    I'd be interested to see this test on another camera. I'm shooting with 2 of them on Friday (One running Q9b and the other Unhacked), so I'll try to see if we have time to check them as well.

    Definitely considering the fotowand card.

  • So you don't have a colour temperature meter to check against ? I think you should try to get your hands on one...


  • A serious one doesn't come cheap – the problem with modern light sources is, that you need more of a spectrometer to find out what CRI your light has…

  • @nomad Which is why tungsten is a great light source for these kinds of tests. It's very stable, known and has just about the flattest spectrum emitted of any light.


  • Hmm.. there is definitely something strange in color reproduction of GH2... anyone any idea how to match GH2's colors to Canon? So far I have no 100% success... I can get close, but still not good enough for me :(

  • @shian What does the MBR Color Corrector do with this footage?

  • @sam_stickland it's windows only, so, no joy.

  • OK, so just tested it using, same setup only with diffusion this time, manual WB by sampling, and it's still doing the exact same thing.

    And tested it on unhacked GH2 as well, same parameters, same results. SO it's not a hacked matrix issue, it appears to be a GH2 issue.

    I'm looking at the manual WB noise issue now.

  • @shian If you can send me a link to the original footage I'll give it a try. I been wanting to try out this software, but I haven't wanted to splash out on a Macbeth chart just to test it!

  • @sam_stickland try this -

    But it'll be better once I've got some footage in a real environment with the chart in the shot for a few seconds.

  • I have a Macbeth chart, but my suspicion is that WB creates some problems. I suppose I should test the custom WB against the built in ones and color correct using Macbeth in Speedgrade which will automatically balance the colors. @Shian you could download the trial of Speedgrade if you don't have CS6 already to use the Macbeth chart analyzer and get a very precise readout of the color shifts.

  • @DrDave EDIT - looked at it in SG, and its the same thing. Used the color match function, and the shifts are still there.

    To Sum up: The Chroma and Luma shift demonstrated in the video both occur regardless of hacked or unhacked, regardless of patch, and regardless of WB method whether it be standard presets, manual sampling WB, or dialing in Kelvin number. And it happens regardless of whether it is actual sunlight or CTB. Unless someone can offer evidence to refute this, I think this is now a known issue.

    In conclusion: Daylight does not render accurate colors on the GH2, whereas Tungsten does.

    Also of note: The manual WB noise issue I observed tends to be confined to Quantum v5 and v2, It is not there unhacked, and is not there in Quantum v9B.

  • @ Shain "In conclusion: Daylight does not render accurate colors on the GH2, whereas Tungsten does."

    This is something drewnet (aka: @onionbrain here?) mentioned in one of his hacked GH2 videos on youtube, that the sensor output looked best (cleanest) if left in tungsten mode, even while shooting in daylight, then CC'ed in post. Personally I'd try an 85 filter first, but haven't tested it.

    His video:

  • @shian - Have you tried shooting with a daylight source but putting a 85 filter on the lens to correct towards tungsten and then shooting with the camera WB to tungsten? It would be interesting to see where that landed on your scopes. (after correcting for the light loss from the filter.)

  • Gh2--don't try this outside :) Damn. That goes along way to explaining why I could not match the colors precisely, in addition to my rotten grading skills. I'll use the tungsten preset and shoot a Macbeth chart in the footy. @Shian, do you have a recommendation for Fluorescent?

  • @DrDave I don't yet. But, as you can guess, it'll get the same treatment as the other two now that I know there might be an issue.

    [I've edited the Header and initial post to cover what we've discovered to this point.]