I have noticed that the blue channel clips easily on both the q3 and the a7r6, even when the histogram looks fine during live view. It is recoverable most of the time, but it is annoying to adjust it in post. If the sky is in the frame, I have to "underexpose" by a stop or more relative to the histogram. I mention the q3 because it uses a Sony sensor, and that is where I first saw this behavior.
With the a7r6, I went outside, aimed the camera at the sky, and adjusted the zebra settings so that the zebras disappear right at a correct exposure. I landed on a setting of 85. Zebras are new to me; I had not used them before. YouTube suggested something around 107 to 109 (on the older a7R models).
I arrived at roughly -0.5 stops of exposure compensation, which keeps the blue channel easily recoverable, or prevents it from clipping altogether. This also depends to some degree on color temperature.
Is there a different way to handle this? Am I missing something?
snapsy wrote:
By what measure or observation are you determining that the blue channel is clipped?
Right, I do all processing in CaptureOne. It is "clipped" if blue channel value is 255 in the sky. Most of the time it is recoverable by adjusting color temperature, exposure, or highlights, but not always.
In this example, a good exposure is 1/2000 when blue does get clipped from the start, but it will look like it is "underexposed" on camera's histogram with a lot of space between highlights and the right edge of the graph.
In the picture above and zebra at 85 most of sky gets marked as overexposed. At 1/1600 most of the sky will be ok. At 1/2000 everything will look fine. If I increase zebra setting to 100 or more, the correct exposure gets shorter, and too dark for the rest of the frame.
olegkin wrote:
Right, I do all processing in CaptureOne. It is "clipped" if blue channel value is 255 in the sky. Most of the time it is recoverable by adjusting color temperature, exposure, or highlights, but not always.
I do not believe a sensor has anything to do with it, it's the camera meter's job to maximize sensor capabilities for any given scene. And indeed, my Nikon cameras tend to expose at -0.5EV relative to my Sony cameras when I shoot them together on the same photoshoot. Not always, but quite frequently.
That said, this could very well be the C1 engine issue, because mapping internal 16-bit luminance values to 8-bit values you're sampling, is the RAW converter's job. I do not have an a7r6, but I noticed that C1 uses a steeper contrast curve with a7r5 files relative to Sony's own or Adobe's profile.
I would play with RawDigger to investigate. I haven't used it in a long while, but IIRC it has a similar sampler tool which allows you to examine RAW data directly without the camera profile influence.
Perhaps someone can clarify this: When we view a live in-camera histogram on a typical Sony camera, aren't we actually looking at a JPG which would be embedded in the raw file ? Therefore, when the camera presents us with zebras, are we not also looking, not at the raw data itself, but something which has already had lossy adjustments applied ?
darbo wrote:
Maybe try shooting with a warmer white balance? Also in camera the AWB can be tweaked to set a hue shift away from blue.
I tried, but for picture above it needs to go into 7000k, too warm for the rest of the frame. I will try to shift blue and see what happens. Basically, I want to find a way to shoot with histogram or zebra guidance but with a minimum post processing to get to the reasonably good image.
Kenneth Lee wrote:
s someone can clarify this: When we view a live in-camera histogram on a typical Sony camera, aren't we actually looking at a JPG which would be embedded in the raw file ? Therefore, when the camera presents us with zebras, are we not also looking, not at the raw data itself, but something which has already had lossy adjustments applied ?
In-camera histogram visualizes the output of the camera's JPEG engine. I verified that with the A7r4 and A7r5 when I was setting those cameras as film scanners, where I needed precise control over the exposure in each of the 3 channels.
old-gregg wrote:
I do not believe a sensor has anything to do with it, it's the camera meter's job to maximize sensor capabilities for any given scene. And indeed, my Nikon cameras tend to expose at -0.5EV relative to my Sony cameras when I shoot them together on the same photoshoot. Not always, but quite frequently.
That said, this could very well be the C1 engine issue, because mapping internal 16-bit luminance values to 8-bit values you're sampling, is the RAW converter's job. I do not have an a7r6, but I noticed that C1 uses a steeper contrast curve with a7r5 files relative to Sony's own or Adobe's profile.
I would play with RawDigger to investigate. I haven't used it in a long while, but IIRC it has a similar sampler tool which allows you to examine RAW data directly without the camera profile influence....Show more →
You are onto something. I did not play with curve profiles for Sony cameras yet. I switched to 'linear response' curve and blue channel moved significantly. Looks like I will have to experiment more.
The only way to tell if any of the channels are clipped is to examine a RAW histogram. No camera made today presents a RAW histogram. If there is one I would like to hear about it. Zebras are a bit better but the last time I checked Sony camera only use the green channel for zebras.
In order to tell if any of the color channels are clipped in your RAW files you need to use something like Fast Raw Viewer or Raw Digger to examine the RAW histogram. That is also how you can determine how to set zebras to get the best possible indication in-camera, but Zebras are still not completely reliable in preventing clipped highlights.
olegkin wrote:
Right, I do all processing in CaptureOne. It is "clipped" if blue channel value is 255 in the sky. Most of the time it is recoverable by adjusting color temperature, exposure, or highlights, but not always.
As you may have intuited from just the responses so far, you're facing a potentially deep rabbit hole related to how clipping is affected and represented by the various players, including the camera and processing tools.
If you want to take the blue pill, you can limit the rabbit hole to empirically determining where the zebras place the blue channel for your typical shooting scene, using the process you've already started of experimentation and processing in your tool of choice to associate that to where clipping typically occurs.
If you want to take the red pill, you can read about the differences between raw and jpg/processed color channel clipping, how exposure aids like the histogram, zebras, etc.. are derived from camera settings, WB, and the embedded jpg in each raw file, how each color channel on a camera has different levels of gain/sensitivity and saturation points needed by cameras to achieve the range of white balances they support, how the depiction of clipping in post-processing tools are affected by the processing settings, etc...
RoamingScott wrote:
I'll ask the most important question. How was your sample photo irrevocably ruined by this supposed blue clipping?
It was not ruined, but it required an extra adjustment that affected the whole frame. In a case worse than this prepared example, the sky would go cyan, and even once recovered, it would look off. Also important to me: this is a case where the histogram is consistently misleading. But maybe if I started with a linear response curve, it would actually be on point. I will experiment more.
The in camera histogram is based on the JPEG the camera produces. The histogram in your photo editor is basically based on the JPEG image your photo editor is going to produce based on the RAW file data conversion + adjustments.
The data can be clipped in either JPEG, but not in the actual RAW data.
The blue channel in a JPEG is a 8 bit vlaue, so it goes from 0-255. The blue channel in an a 12 bit RAW file goes from 0-4,095. Since RAW files are usually linear, the brightest stop of light will use the range 2,047-4,095. Since JPEG files use a gamma correction to increase the effective dynamic range, the brightest stop of light will be approximately 191-255. So a value of 255 in a JPEG may be a RAW value of 4,015. So technically not clipped. This is not even taking into account RAW conversion adjustments like white balance, which may push up the blue channel in a JPEG even more than it is in a RAW file.
So what you are seeing in your final photo may be more of a preference or an editing artifact than the camera technically clipping data. Hope this helps you figure out how to get the images you prefer at the end of the day
olegkin wrote:
I have noticed that the blue channel clips easily on both the q3 and the a7r6, even when the histogram looks fine during live view. It is recoverable most of the time, but it is annoying to adjust it in post. If the sky is in the frame, I have to "underexpose" by a stop or more relative to the histogram. I mention the q3 because it uses a Sony sensor, and that is where I first saw this behavior.
With the a7r6, I went outside, aimed the camera at the sky, and adjusted the zebra settings so that the zebras disappear right at a correct exposure. I landed on a setting of 85. Zebras are new to me; I had not used them before. YouTube suggested something around 107 to 109 (on the older a7R models).
I arrived at roughly -0.5 stops of exposure compensation, which keeps the blue channel easily recoverable, or prevents it from clipping altogether. This also depends to some degree on color temperature.
Is there a different way to handle this? Am I missing something?...Show more →
Using the A7r-V to shoot lossless compressed RAW files only, with the zebra threshold set to 107+, I very infrequently noticed a couple of the brightest areas in waterfall photos had an odd cyan color cast. These are always long exposures (in the 0.5 second range) captured on cloudy days. The areas are small, but larger than tiny specular highlights, which really don't exist in these conditions. The very light cyan patch in the middle of a neutral highlight is very noticeable, and I found it to be unrecoverable except by using a brush in Photoshop to desaturate the spot, making it the equivalent shade of grey. If this was not a true overexposure, it would be recoverable in the RAW file, and it definitely is not recoverable.
I am certain there were no zebras on any of the offending highlights. I also know that setting the exposure 1/3 stop higher did create zebras on the same highlights.
I haven't changed any of the JPEG settings and all are still at their (auto) default. I have noticed that the AWB (based on the "as shot" white point in ACR) is always on the cool (blue) side of the ACR default for the conditions, and the ACR "auto" setting for the image, and what ACR measures based on me selecting a bright neutral area in the photo. Perhaps adjusting the JPEG white balance for the prevailing conditions in each photo would fix this, but fussing with that is a hassle considering that I don't even have the camera record a JPEG file, except the one it embeds in the RAW file.
Because this has happened only two times in 1.5 years and around more than 50 different waterfalls, I'm not very excited about it. In fact I'd forgotten about the issue until your post reminded me. My "plan" for now is to set the zebra threshold set to 106+, be more aware of the issue, and see what happens.
Thanks reminding me about this issue. Since our shooting situations are very different my "solution" probably won't help you, but I do think adjusting the WB setting may be on the right track, and a more proper way to go about solving the issue.
DWOfPaul wrote:
The in camera histogram is based on the JPEG the camera produces. The histogram in your photo editor is basically based on the JPEG image your photo editor is going to produce based on the RAW file data conversion + adjustments.
The data can be clipped in either JPEG, but not in the actual RAW data.
The blue channel in a JPEG is a 8 bit vlaue, so it goes from 0-255. The blue channel in an a 12 bit RAW file goes from 0-4,095. Since RAW files are usually linear, the brightest stop of light will use the range 2,047-4,095. Since JPEG files use a gamma correction to increase the effective dynamic range, the brightest stop of light will be approximately 191-255. So a value of 255 in a JPEG may be a RAW value of 4,015. So technically not clipped. This is not even taking into account RAW conversion adjustments like white balance, which may push up the blue channel in a JPEG even more than it is in a RAW file.
So what you are seeing in your final photo may be more of a preference or an editing artifact than the camera technically clipping data. Hope this helps you figure out how to get the images you prefer at the end of the day ...Show more →
Yes, I understand this. The solution for now is to underexpose 0.5-1 stop from histogram suggestion on purpose when there are a lot of blue in the frame. I am experimenting with editing starting with "linear" vs "auto" curves. Linear actually works very well on Sony. But I am not there yet. I tried linear curves before on Fujifilm and Nikon and did not find it rewarding except for some rare images.
olegkin wrote:
Right, I do all processing in CaptureOne. It is "clipped" if blue channel value is 255 in the sky. Most of the time it is recoverable by adjusting color temperature, exposure, or highlights, but not always.
Which film curve are you applying in C1? One of the Sony curves, or one of C1’s like “film standard” — I would apply a few different ones to your image above and see if any help the issue. I personally find C1’s film standard to be really well balanced, better than most manufacturer curves.
